In short: Owned by the whole committee (15:41). Link "just recently bought NVIDIA when it was really almost at the lows" — 17× against a 35× long-term average (20:44). Talkington holds but trades it: range-bound 210–230 until it clears its ~232–236 ceiling, "until it can break out, I would just sell calls" (16:53). Jensen Huang's "0% chance" of AI doom calms the "peak doomerism."
Everyone on the committee owns NVIDIA. Link recently bought more near the lows, when it traded at about 17 times earnings against a long-run average of 35. Talkington holds it but earns extra income by selling call options, which pay her now in exchange for giving up gains above a set price. She does this while the stock stays stuck between about $210 and $230, and expects a big move higher once it breaks above ~$232.
In short: Passing mention — in her anecdote, the growth stock a would-be dividend investor keeps running back to out of fear of missing out. No view on the company.
6:19I give up on it and I go back to buying Nvidia." And he was basically trying to have me convince him that he should own Kimberly Clark. And I said, "Buddy, you've answered your own question. You want to be a dividend investor, but you are not a dividend investor. You can't handle that fear of, you can't handle the patience, you can't handle the fear of missing out on the growth." So, it was interesting.
In short: Referenced only — Jensen Huang's line that "nobody uses AI better than Meta," which Muse is "proving again today."
20:27Not only beat them to it, but executed in excellent style. I am blown away by how Meta did this. It is incredible that they flanked both Gemini. They flanked ChateBT and Anthropic at reaching the consumer first. They really caught up. It's incredible to see. It reminds me of what Jensen Wong said when he said that nobody uses AI better than Meta.
In short: Skeptical framing from both. Eisman: revenue "up over 100%," but in the 10-Q "70% of accounts receivable is from five accounts… the largest company on planet Earth basically 70% of its revenue comes from five customers is very frightening" — and if lenders thought these were good credits "they'd make loans, but they're not doing it." Noble: "the fact that Nvidia has to announce… $250 billion package for Open AI to fund what they're doing… tells me all I need to know" — "funny money." No position stated here (Eisman was still "less long," Aug 28).
NVIDIA makes the chips that power AI, and its sales doubled. Both men see warning signs in how that is happening. Eisman notes that five customers owe NVIDIA 70% of its unpaid bills, so it depends on a handful of buyers. Noble notes that NVIDIA announced a $250 billion financing package for OpenAI, one of those buyers: in effect NVIDIA helps pay for its own chips.
When a supplier has to fund its customers, and banks aren't rushing to lend to those customers, it suggests the demand isn't paying for itself yet. Neither says he is betting against the stock here. It is a warning about the quality of the growth.
23:32this day can't understand why no one has comment on this other than a few substackers which is Nvidia's revenue in the quarter that was just recently reported was up over 100%. But if you go in the queue, it says that 70% of accounts receivable is from five accounts. [sighs] I mean, think about it this way that the largest company on planet Earth basically 70% of its revenue comes from five customers is very frightening.
In short: Passing mention — the week's recap items, not a stance. Sold with the semis on Monday's "long software, short semis" rotation after the pacing debate; "Truist initiated Nvidia at a buy with a 355 target… it's seeing higher pricing"; and "Nvidia announced another 20% price increase on… the H100, H200, B200, and B300… effective on October 1st. That actually led the stock to go a little bit higher." Context he adds: global semiconductor revenue a record $425bn in Q2 (+31%) — "despite all this discussion about a slowdown, semiconductor earnings have been the highest ever recorded."
Full passage: premium transcript (PDF).
In short: Named only as an example of the outsized winners the market is hunting ("the next Nvidia or the next Google or the next Micron"); he expects not to own the next ones and doesn't try to.
1:20:13And in theory that should be making it more difficult to do good fundamental research, because these tools are available to everyone. But in practice I almost get the sense that because it's become a commodity, people are just ignoring it, because they feel, well, what's the point if everyone else has the same access, and the focus instead seems to be on trying to find that small number of companies which are going to be the next Nvidia or the next Google or the next Micron, to take a more current
In short: Bill Baruch is buying more with his Apple trim proceeds (25:43). "We want to own the names more heavily that are going to drive the breakout in that Mag 7 ETF. We're buying more NVIDIA" — he expects the largest names to lead one of the year's two "risk-on thrusts."
Baruch runs a concentrated 10-stock portfolio. He sold part of his Apple stake because it looked expensive and put the money into NVIDIA and Amazon, which he expects to lead the next leg up in the biggest tech stocks.
In short: Used as the worked example rather than a pick — how to hedge big tech gains without paying tax: "Let's say I own a 100 shares of Nvidia and I've owned it for years… Instead, I could short 50 or 100 shares of Nvidia. That's called shorting against the box… I can wait until I think things will get better and then I can undo the short." A hypothetical, not a disclosed trade.
21:20How can you hedge that risk? Now, one way to do it is to short all or part of your positions against the box. What does that mean? Let's say I own a 100 shares of Nvidia and I've owned it for years. So, I have massive embedded gains. Let's also say I'm nervous. I want to reduce my risk. My problem is that if I sell my Nvidia position, I will pay 20% in federal long-term capital gains taxes plus the long-term capital gains tax rate of my state.
In short: Leading the AI-hardware bounce (0:54). Ethridge: the chip trade (NVIDIA, AMD, Broadcom, then memory/optical) "got away from us"; the committee is "offsides" being so heavy in chips and expects rotation to the Mag 7 spenders. Flatt: Nvidia is leading the push to make chips a financeable asset class.
In short: Top of "the entire AI chain, from Nvidia to the hyperscalers to Anthropic and OpenAI," which "all depends on the future health of Anthropic and OpenAI" — "if something were to happen to one of those two companies, then the chain would really fall apart." Still not adding to AI: "I've taken some… off the table." On the host's theory that Nvidia bought Hugging Face as an insurance policy for OpenAI: "there were other bidders… one of the reasons Nvidia might have been willing to pay more is to protect Hugging Face… you're protecting OpenAI… the entire ecosystem."
NVIDIA makes the chips that AI runs on. Eisman's point is that its customers, the big cloud companies, are in turn largely selling to just two AI labs, Anthropic and OpenAI. If either of those labs got into serious trouble, the whole chain of spending could unravel, NVIDIA included.
He isn't betting against NVIDIA, but he has sold some of his AI stocks and isn't buying more. He also entertained a host's theory that NVIDIA's purchase of Hugging Face was partly about shielding OpenAI from a potential lawsuit.
3:06What I would say is that. This is how I sort of think about the chain. — Sounded like you just described, that we're at some type of tipping point for AI if they're desperate to create a narrative because things aren't going. — Well, I think we're potentially closer. But I think the issue is that at the end of the day, the entire AI chain, from Nvidia to the hyperscalers to Anthropic and OpenAI all depends on the future health of Anthropic and OpenAI, because there's such a huge percentage of the entire chain. So if something were to
In short: The host asked why the stock is flat (~$200–210) despite ~150% compounded earnings upgrades. Deluard's answer: circular financing, the same customers, and a base effect. Nvidia's doubling comes from the hyperscalers doubling capex (>$1T across five companies), "you can't just keep doubling that… you run out of economy to eat." The AI narrative "is starting to weaken."
Nvidia's profits have exploded, yet the stock is roughly where it was last autumn. His explanation: Nvidia's growth depends on a handful of tech giants doubling their spending on data centers, already more than a trillion dollars. Some of that money circles back between the same companies ("circular financing"). Spending can't double forever before it runs out of economy.
He sees the market quietly doubting that those earnings will last, which fits his view that the AI story, now about four years old, is fading.
32:50market just going, we don't believe these earnings are going to come to fruition, and the market is just smart enough to know that? How do you read that? — Yeah. There's an awful lot of circular financing, same customers, the view that just the base effect, right — this doubling of the revenues or the earnings of Nvidia comes from the doubling of the capex of the hyperscalers, right? So we're already talking
In short: Context only: Jensen Huang keynoted Equinix's investor day — Equinix "has this relationship in place with Nvidia."
17:43The keynote speaker was Jensen Huang from Nvidia. So they have this relationship in place with Nvidia missionritical global internet connection platform very powerful network revenues were up earnings were up it's supported by cloud network AI demand obviously things to be concerned about power that's the one thing we concern for every data center reach where's the power coming from premium valuation again because of that sector being so hot the intense capital nature of data centers can they execute on development, concentration
In short: "If you look at a company like Nvidia, 27 times earnings, and they stated expectations of increasing earnings by 70%, that is a relatively low valuation" — high historical multiples are cheap against AI-driven earnings growth.
Nvidia makes the chips that train and run AI. Peterffy's point is simple math: a price of 27 times earnings sounds expensive, but if earnings grow 70% next year the stock gets cheap very fast. He thinks the broad market's high valuations are justified the same way, as AI spreads into every company's productivity.
12:01So if you look at a company like Nvidia, 27 times earnings, and they stated expectations of increasing earnings by 70%, that is a relatively low valuation, and a similar situation to a lesser extent is what we can expect everywhere else as AI permeates the economy. So it's not necessarily a focus on AI only.
In short: Investor, supplier and backstop at once — the circular-financing case in one deal. Buying $1.5B of non-voting Class N shares in a private placement at the IPO price, plus a separate $1.5B prepaid forward with parent Energy Global (paid in August, shares at 90% of the IPO price) — ~$3B of IPO-linked exposure, half of it fresh capital. Provides the compute at PORTS-Pike and "substantial credit support for portions of OpenAI's leases," so "its credit support helps SB Energy finance a campus that will exclusively host NVIDIA infrastructure… That tightly aligns NVIDIA's chip sales with SB Energy's ability to build the real estate." A disclosed author holding.
NVIDIA plays three roles in this one company: it sells the chips that will fill SB Energy's biggest campus, it is investing about $3 billion in SB Energy's shares around the IPO, and it guarantees part of the rent OpenAI owes. That last piece ("credit support") makes it easier for SB Energy to borrow money to build a campus that will run only NVIDIA hardware.
The point for NVIDIA investors: its chip sales are increasingly tied to financing it helps arrange for its own customers — critics call this circular. It supports demand today but means NVIDIA carries more risk if AI companies struggle to pay for what they're building. The author owns NVIDIA shares. Analysis, not a recommendation.
In short: "In a totally different bucket" from the neoclouds: "they forecasted 70% revenue growth for calendar 27 and you're paying right now a 14 times PE" against the S&P at ~18x and the public clouds in the low 20s — "not paying an outlandish valuation for the guy with the best chips on the planet." Also "pushing very hard in the open-source, open-weight model category." The tension is customer concentration as hyperscalers build their own ASICs, but "Nvidia is one that I feel like deserves a place in people's portfolio."
Nvidia makes the chips that power AI. Niles' case is price versus growth: the company guides to about 70% revenue growth next year, yet the stock trades at roughly 14–15 times earnings — cheaper than the S&P 500 (about 18x) and the big cloud companies (low 20s). You are "not paying an outlandish valuation for the guy with the best chips on the planet."
He also likes that Nvidia is backing open-weight (freely available) AI models, which he expects to handle most everyday AI work. The risk he names is that its biggest customers — Google, Amazon, Microsoft — are designing their own custom chips, so Nvidia depends heavily on a few buyers who are also becoming competitors. That's why he treats it as one position in a portfolio, not an all-in bet.
47:11Nvidia, you're paying, if you believe their forecasts, they forecasted 70% revenue growth for calendar 27 and you're paying right now a 14 times PE. The S&P is trading at I think it's 18 times right now. And the other, internet guys, or not internet guys, but public cloud guys like Microsoft, Google, Amazon, they're all trading in the low 20s.
In short: An unprompted aside while scanning the chat (possibly reading a member comment aloud): "Nvidia is a strong buy here probably. I should not have sat on the sidelines with Nvidia, but it's just incredible what that company continues to do."
In short: He couldn't model "what an NVIDIA is going to guarantee and what they're not going to guarantee and what their cost of capital really is," so he went top-down. Chip prices up "6x for NVIDIA" and memory ~7x have pushed the cost of building compute to "astronomical, ludicrous" levels against a revenue pool that can't carry it. The supplier to a build-out he calls a bubble.
NVIDIA sells the chips behind the AI build-out. Taylor didn't call for shorting it outright, but his numbers point against it: about $3 trillion of AI data centers will exist by end-2027, and wearing out that equipment alone costs around $400 billion a year. Revenue from the main AI customers leaves only about $40 billion to pay for it. Chip prices up roughly six-fold and memory seven-fold have made that gap worse. He also says it's unclear how much of its customers' financing NVIDIA is effectively guaranteeing. If the spending has to slow, the company selling the shovels feels it.
3:28Because I went through this and I was like, how can I figure out what an NVIDIA is going to guarantee and what they're not going to guarantee and what their cost of capital really is? And then CapEx has distorted everyone's numbers. So I just said, okay, erase all that. Erase. At the end of 27.
In short: Financing the AI party: $125B into OpenAI's $500B data-center project and reportedly a $10B lead anchor for the Anthropic IPO — "they have to keep this party going." Jensen Huang may shrug off the China question; Wall Street roadshows don't.
Nvidia makes the chips that power AI. Woo isn't judging the stock. He points to Nvidia as the company paying to keep the AI boom going: $125 billion toward OpenAI's giant data-center plan, and reportedly $10 billion to anchor Anthropic's stock-market listing. His point is that the AI industry is now funding its own customers to keep the party going, a sign of how much depends on the boom not stopping.
28:14If it goes poorly, the whole AI trade could basically collapse and then it's the end. But the point here is that they have to keep this party going. This is why Nvidia is now said to be interested in putting $10 billion into basically, and is a lead investor to anchor it and so on so forth.
In short: Analogy, not a pick: "look at Nvidia stock" — when value creation is recognized "the stock price can go nuts"; that has happened in semis, AI and gold but "hasn't happened yet in energy." Chip-price inflation also used to explain the crack-spread surge.
40:39And if there's graphite, graphene, lithium in Canada, those, hydrogen, all of those are going to be important. And you need to be looking at, a lot of them are smaller and speculative. But look what AI did from two years ago to today. Look at Nvidia stock. Look at the private market value of Anthropic.
In short: Down on the day with ASML as investors priced "a serious slowdown in the frontier of AI." In the motive analysis Nvidia is the cost line the labs want to shrink: "They have to keep paying Nvidia over and over again at a time point where they already have market share dominance." No stance on the shares.
27:23They have to keep paying Nvidia over and over again at a time point where they already have market share dominance. They already have all the customers using them. All the Fortune 500 companies, everybody's plugged into Claude Code and Anthropics products left, right, and center. What they're focused on now is monetization.
In short: Eisman, via Ed Zitron and his own read of the 10-Q: revenue "up over 100%," but "note 7 says that the top five direct customers of Nvidia accounted for 70% of accounts receivable" — "totally concentrated in a handful of names and a couple of those handful of names are not looking so good. Open AI certainly." A concentration flag, not a call (he remains "less long," Aug 28).
NVIDIA makes the chips that power AI, and its sales more than doubled. Eisman looked in the quarterly report for who owes NVIDIA money, and five customers account for 70% of it. Sales that big from so few buyers are risky, especially when one of those buyers, OpenAI, is losing a lot of money. This is a warning sign, not a bet against the stock.
13:27And so I said to him, Nvidia's revenue numbers were up over 100%. and he said, "Yeah, but and then he pointed out and I confirmed this later in the day that in the 10 Q there's in note 7 I actually looked this up — stuff that we used to do to financial stocks. So note 7 says that the top five direct customers of Nvidia accounted for 70% of accounts receivable in the quarter. Okay.
In short: Fundamentals untouched, sentiment exposed — and the name at the end of the funding chain he is worried about. The flywheel: "if they aren't supporting their cloud backlogs, how are they going to increase capex? … If they don't increase capex, who's going to buy Nvidia GPUs so that it grows earnings 30% every year and who's going to be buying all this memory? Now, it's not going to matter in the next quarter or two." The balance: "for the actual AI trade, Nvidia, memory, power, optical… nothing here changes the near-term capex number. No hyperscaler guide has moved… for now, this is sentiment risk, not fundamental risk." The near-term tape worry: "semiconductors could sell off and energy might be the only sector that rallies tomorrow because of this AI slowdown." From Goldman's TMT conference, Nvidia's own message: "the bottleneck has basically moved from chips to power. Demand is broadening rather than narrowing."
Nvidia makes the chips almost every AI system runs on. Nothing about its business changed this week — no customer cut spending, no data centre was delayed. What changed is the mood, after the heads of Anthropic and OpenAI both talked about slowing down the development of the most powerful models.
Singh separates the two. Fundamentally, he says, "this is sentiment risk, not fundamental risk," and the rule is to wait for real evidence: a delayed model, cancelled chip orders, or a company lowering its forecast. But he describes the chain that worries him. The AI labs depend on raising huge sums from investors. If Anthropic's stock-market listing slips and OpenAI cannot list at all, the labs may spend less on cloud computing from Microsoft, Amazon and Google. Those companies would then slow their own spending — and they are the ones buying Nvidia's chips and all the memory that goes with them. None of that shows up for a quarter or two, but a mere slowdown in growth can be enough to push the shares down, and he warned semiconductors could sell off the next morning.
Full passage: premium transcript (PDF).
In short: Dale, not Gromen — named only as the future tax target. With 85–90% of federal receipts coming from workers and AI displacing them, "they're going to start taxing where the money is… there's going to be a Wile E. Coyote moment where we realize, how does Uncle Sam pay for all this? And they're gonna go Nvidia." No company call from either guest.
1:30:14But ultimately, that's going to cause an unwind of the market because there's going to be a wow coyote moment where we realize, how does Uncle Sam pay for all this? And they're gonna go Nvidia. — That's who — Jensen Huang. So you're gonna get a big a big check in a few years, my friend. So enjoy the good times while they last, — right? Or going the right route of California where All right.
In short: Top of the revenue-growth sort: "Nvidia: +69.1%." The only name in the top nine of all three sorts — 5-yr diluted EPS CAGR 95.1%, 3-yr ROIC 128.4%, ROCE 104.6%, FCF margin 45.4%, on a $5.52trn market cap. Listed only; no commentary.
In short: Cautious (caveat: "not a semi analyst"): capitalizing semis margins into perpetuity ignores Chancellor's capital cycle — "as great as their products are," the "Jensen tax" incentivizes Google/Amazon to build their own chips, and margins mean revert. Could run longer than people expect.
This is a caution, not a short call — he stresses he isn't a chip analyst. Chip companies' profit margins are at record highs, and investors are pricing them as if that lasts forever. History (Ed Chancellor's "capital cycle") says fat margins attract new supply until they fall back.
The specific risk: Nvidia's biggest customers, Google and Amazon, are designing their own chips to escape paying Nvidia's premium (the "Jensen tax") and its supply bottlenecks — and trillion-dollar companies are exactly who you don't want as competitors. It could run longer than people expect, but the incentive is working against those margins.
25:45And with that caveat, I will say if I were a semi investor capitalizing these margins and growth into perpetuity, I'd be very concerned about the fact that Google and Amazon are now developing their own chips. Those are typically people you do not want to be competing with. And people refer to Nvidia, as great as their products are, as the Jensen tax.
In short: Held (top of her Sep 4 holdings list, per the host: "you've got really NVIDIA, which makes sense"). Her explicit case is about hyperscaler psychology, not the chip cycle: "I think the leadership at every hyperscaler thinks about this exactly the same way that there's no cap to the demand here… And as long as literally every hyperscaler C-Suite is thinking about this in those terms, I think you can comfortably buy NVIDIA. We'll sleep well at night."
NVIDIA makes the chips almost every AI data centre runs on. Wilde owns it, and her reason is simple: the people who buy those chips — the bosses of the giant cloud companies — all believe there is no ceiling on demand for AI computing, so they will keep spending.
Her argument is about buyer behaviour rather than chip specs: as long as every one of those executives thinks this way, the orders keep coming and "you can comfortably buy NVIDIA." The risk is the flip side — if that shared belief cracks, the spending can slow all at once.
In short: His inelasticity example: for the most inelastic mega-caps the flow multiplier is "approaching a hundred… $1 into an Nvidia is raising Nvidia's market cap by $100" — the mechanism concentrating the market. A market-structure illustration, not a call on the business.
Nvidia is his example of how "inelastic" the market has become. In textbook theory, a dollar of new buying barely moves a stock because some seller always steps in at a fair price. Academic work found that in reality each dollar of flow into the market added about $5 of market value on average from 1992 to 2019.
Green thinks that multiplier has grown as passive funds took over — to roughly $22 on average, and close to $100 for the largest, most index-owned names: "$1 into an Nvidia is raising Nvidia's market cap by $100." That is why the biggest stocks keep getting bigger. It's a statement about market plumbing, not about Nvidia's chips.
10:38Those multipliers are now approaching a hundred. And so $1 into an Nvidia is raising Nvidia's market cap by $100. That's causing the market to become more concentrated. It's causing all sorts of perverse effects. When you think about the market in that framework and you think about something like SpaceX, what they're really trying to do is make investors aware that bid will be there to provide liquidity for them to exit if they participate early.
In short: Subscriber view (rebutting the Zitron "circular financing" framing): semis were ~45% gross-margin businesses, NVIDIA runs near 80% (Eisman: "I think they were 74%"), so it is "extracting rent from hyperscalers and then using it to finance their emerging competitors" — labs, then neoclouds and open-weight models — a "modern-day Robin Hood" that may "accelerate the commoditization" of its own rents. Eisman's caveat: "this could be correct," but the current ecosystem depends on Anthropic and OpenAI, and one could fail before the challengers can replace their commitments.
This view is mostly a subscriber's, not Eisman's. NVIDIA keeps about three-quarters of every sales dollar as gross profit, far above what chipmakers historically earned. The reader argues NVIDIA is using that surplus to fund the companies that compete with its biggest customers — AI labs, smaller "neocloud" rental firms, and free open-source models (hence buying Hugging Face, the hub where those models are shared). Like Robin Hood, it takes from the rich cloud giants and arms their challengers, which should make AI cheaper for everyone over time.
Eisman takes the argument seriously and allows it could be right in the long run. His objection is about timing: today's AI spending leans on two money-losing labs, OpenAI and Anthropic. If one fails before the new challengers are big enough to replace its orders, the whole chain could suffer first — just as the first wave of internet companies collapsed before the second wave, like Google, succeeded.
12:10One viewer sent in a long email criticizing Ed's negative thesis on AI, and I quote it below. Steve, I'm surprised you characterize Nvidia's strategy as circular financing. Historically, semis were roughly 45% gross margin businesses. Nvidia is now operating at close to 80%. Economically, Nvidia is extracting rent from hyperscalers and then using it to finance their emerging competitors.
In short: A named single-name short expressed as options. "So one market has a big rally up we increase our put exposure, mostly S&P 500 but sometimes others like right now we have puts also in Nvidia." The company's own guidance is what he uses against it: "Nvidia said 70% growth next year. Okay. Well, that will make about close to $1.7 trillion of spending overall if the whole sector grows another 70%. So I think we're very much coming to the top, the peak of spending… I really don't believe that 1.7 trillion spending is sustainable. So when we go down, what happens to all these stocks? I can imagine that those stocks would go down a lot."
12:19It just became too hard. — Well, it is hard. It is hard. But since 2021 I added a feature to the fund, the way I manage. I reduce the number of shorts I have. So we're still short but then I have puts instead. So when a market has a big rally up we increase our put exposure, mostly S&P 500 but sometimes others like right now we have puts also in Nvidia. And when the market goes down then we sell some of our puts.
In short: The terminal recipient of the loop: the four hyperscalers "have spent over a trillion dollars in capital expenditure buying GPUs from NVIDIA," and the guest's premise is that this spending is being justified by revenue the buyers themselves fund. Zitron is on to argue "why he believes the AI trade is overheated." The post's dedicated section — "Looking Closer at NVIDIA's Numbers" — is behind the paywall and is not captured, so no specific claim about NVIDIA's own accounts is recorded here.
NVIDIA sells the chips. The four biggest cloud companies — Google, Microsoft, Amazon and Meta — have spent more than a trillion dollars buying them. That spending is the reason NVIDIA's numbers look the way they do.
The guest's case is that those buyers are spending in order to capture revenue from two loss-making companies they themselves are funding — so the demand behind NVIDIA's sales rests on a financing arrangement rather than on a broad base of profitable customers. If the funding slows, the chip orders slow with it.
An honest caveat on this page: the post's dedicated section on NVIDIA's own figures sits behind the paywall and was not captured, so the detailed argument about NVIDIA specifically is not recorded here — only the demand-side framing above.
In short: Lebenthal's single Mag-7 leadership name, and the one number Belski uses against the bubble case. Lebenthal: "whether it's Alphabet, whether it's NVIDIA in particular, which out of the seven is the one that I look to most for leadership… it's within a hair's breadth of an all time high. I think it's very likely that it takes that out… it's not going to look back and go down from an all time high… on around roughly 19 times forward earnings. We know what they said two weeks ago about revenue growth rate in the coming year, 70%, which by the way is supply constrained — so it could be higher if things on the supply chain become a little bit unstuck." Belski uses it as the anti-1999 exhibit: "remember how many times on this set we talked about 30 to 35 times multiple in the S&P 500 is where everything peaked in '99-2000. You have NVIDIA at 17½ times, the largest stock in the market. That tells a lot." Wapner's pushback is that the S&P's own multiple is higher than that.
Nvidia makes the chips that AI systems run on. Lebenthal's point is not that the story is new but that the price is close to breaking out: the stock is "within a hair's breadth" of its record, and in his experience a stock that clears a record high does not usually turn around and fall back through it.
The numbers he leans on are unusual for a company this size — about 19 times next year's earnings, with the company itself guiding to 70% revenue growth, and that growth figure is limited by how many chips it can physically make rather than by how many customers want them. If the supply bottleneck eases, the number could be higher, not lower.
Belski uses Nvidia for a different argument. Bubbles usually show up as expensive multiples on the biggest stocks; in 1999-2000 the whole index was on 30 to 35 times earnings. The largest company in the market today trades at roughly 17.5 times. Wapner's fair objection is that the index as a whole is dearer than that, so the single stock is not proof of anything about the market.
In short: An illustration inside his answer to Munger, with no view on the company: "Netflix has had a couple of 75% drawdowns. Nvidia has had a couple of 75% drawdowns… If I take a 75% drawdown in something, I'm waving the white flag." The point is that even the era's best businesses hand their holders losses most people cannot sit through.
Nvidia makes the chips almost all AI training runs on, and it is one of the best-performing large companies of the last decade.
That is exactly why it appears here, and it is not a view on the stock. Dillian is rebutting Charlie Munger's claim that an investor who cannot stomach 50% declines deserves mediocre returns. His counter-evidence is that the biggest winners inflict the biggest falls along the way: "Netflix has had a couple of 75% drawdowns. Nvidia has had a couple of 75% drawdowns."
A drawdown is the peak-to-trough fall in an investment's value. Losing 75% means you need a fourfold gain just to get back to where you started, and Dillian's honest admission is that he would not sit through it: "if I take a 75% drawdown in something, I'm waving the white flag. I surrender." The point is about what a real person can hold, not about what the company is worth.
20:06Nvidia has had a couple of 75% drawdowns. I just disagree with Munger fundamentally. Drawdowns are the enemy. If I take a 75% drawdown in something, I'm waving the white flag. I surrender. That's it. I'm not buying more, down 75%. So I just don't understand that mentality at all.
In short: Dawson's lead example of a live accounting shenanigan — a supplier financing its own customers' optics. "We see for example Nvidia extended its accounts receivables in order to help its customers. Well, its customers are the hyperscalers who had extended accounts payables, which flatters their free cash flow. So it made Nvidia's free cash flow look worse, but it made the hyperscalers' free cash flow look better. And if you then normalize that, we would have probably seen even more negative free cash flow numbers coming out of the hyperscalers." Not a call on the chips — a warning that the cash-flow statements on both sides of the AI trade are being managed, and that "if things were as great as you say that they are, you should not have to be playing the games that you are playing."
Accounts receivable is money a company is owed but has not yet collected; accounts payable is money it owes but has not yet paid. Cameron Dawson's observation is that Nvidia has been letting its biggest customers — the cloud giants buying the chips — take longer to pay. That is not illegal or hidden, but it moves cash around in a way that makes both sides look different than they are: the customers get to show more free cash flow (they kept the cash longer), and Nvidia shows less.
The reason this matters is not Nvidia's cash position, which is enormous. It is that the cleanest evidence anyone has for "the AI build-out is genuinely profitable" is the hyperscalers' cash flow — and part of that cash flow is on loan from their supplier. Dawson's argument is that if you undo the effect, the hyperscalers' free cash flow looks worse than reported, not better.
Treat this as a quality-of-earnings flag rather than a call on the chips. Nobody on the episode says the demand is fake; Dan Niles argues the opposite in the same show. The claim is narrower and harder to dismiss: a supplier that has to finance its own customers' optics is behaving like a company near a peak, not one in the early innings.
20:47You know, we see for example, Nvidia extended its accounts receivables in order to help its customers. Well, its customers are the hyperscalers who had extended accounts payables, which flatters their free cash flow. So it made Nvidia's free cash flow look worse, but it made the hyperscalers' free cash flow look better.
In short: Cheap on the guide, uncomfortable on the buyer. The fundamental line is constructive: "Nvidia is guiding to 70% growth. AVGO is guiding to 100% growth. Both are supply constrained and saying real demand is higher, but the market is not giving them credit," and it "has agreed to acquire Hugging Face for 12.93 billion, giving it one of the largest open AI development platforms in the world." The positioning line is the warning: with earnings season over and systematic exposure rebuilt, retail is the incremental buyer — "retail investors bought 250 million of Nvidia shares on Wednesday, their third largest daily purchase since mid-May. The retail buying streak has gone to 15 consecutive trading sessions with investors purchasing more than 2.5 billion in Nvidia stock over this period, which is bananas." Sitting under both is the JPMorgan/Citadel advice to reduce risk for three weeks.
Nvidia makes the processors nearly all AI systems are trained on. On the business itself Singh is positive: the company is guiding to 70% growth, says it cannot make enough chips to meet demand, and has agreed to buy Hugging Face — the main public repository of open AI models — for $12.93 billion. "Both are supply constrained and saying real demand is higher, but the market is not giving them credit."
His hesitation is about who is buying the shares, not about the company. Institutional investors have finished rebuilding their positions, the quantitative and momentum funds that were buying have finished buying, and with earnings season over there is no upcoming event to move the price. What is left is retail: individuals bought $250 million of Nvidia in one day and more than $2.5 billion across fifteen consecutive sessions — "which is bananas."
An asset whose only remaining incremental buyer is the general public is a fragile one, because that buyer has no obligation to keep buying and tends to sell together. That is why both JPMorgan and Citadel are advising clients to reduce risk for the next three weeks, and why a good company can still be a neutral position for a month.
Full passage: premium transcript (PDF).
In short: On a Bloomberg/JP Morgan chart of hyperscaler and Nvidia debt issuance as a share of Treasury issuance: they've "consumed all their cash flow… They're going into debt… I've seen this movie before in the telecom bubble… this is far and beyond what that was." Technology is "in the too hard pile for me" — he won't own it.
Nvidia sells the chips that power AI data centers, and the "hyperscalers" (the big cloud companies) are its biggest customers. Polomny shows a chart where the debt these companies are issuing is climbing fast relative to US Treasury issuance — meaning the AI build-out has outgrown the cash the businesses generate and is now being funded with borrowed money.
He says technology is outside his competence and he won't invest in it, but the pattern looks familiar: the late-1990s telecom build-out was also debt-funded and ended badly, and "this is far and beyond what that was." His stance is a warning, not a short.
12:22move on with the slideshow, but that's what I'm talking about. So, this is interesting. This is from Bloomberg and JP Morgan. Evidently hyperscaler, we're talking about the data centers and Nvidia debt issuance as a percent of treasury bond issuance. And so you can see the acceleration in debt issuance by these data center
In short: A phase-1 "build" winner (with AMD, Micron, SanDisk and power names) — chasing them isn't his strategy, and phase 4 (2028–32 commoditization) is "the most dangerous" time to still hold the bottleneck sellers; "maybe a company like Nvidia will build a good enough moat."
In short: The stock the whole episode is built on, and Kevin Simpson's final trade. Lebenthal supplies the causal claim: "you know where I look in the Mag 7, Scott, is particularly at NVIDIA. I really think that's where the turnaround in the Mag Seven came… we liked the stock response, but it hadn't flowed through, at least last week. I think that was just a delayed reaction. And I'm also happy to say it's not just chips— it's things like Microsoft, it's things like Meta. To me, it was really NVIDIA, the clearing event… that tells us that the AI CapEx is going to continue, that it's profitable." The tape: "Nvidia's up seven — well, now it's 6%… it was up as much as 7½% this week. They did the Hugging Face deal." Simpson's final trade is the one-liner: "NVIDIA. We like the Hugging Face acquisition."
Nvidia makes the chips that AI systems run on. Its results a week earlier were strong, but the rest of technology did not move on them straight away. Lebenthal's argument is that the delay was just a lag, and that the earnings functioned as a "clearing event" — a result definitive enough to settle an open question for everyone in the sector, not only for the company reporting.
The question it settled was whether the enormous spending on AI data centres is going to continue and whether it earns its keep. His evidence is other people's numbers, not Nvidia's: Amazon's chief executive saying a new data centre pays for itself in under three years, Alphabet's cloud business growing 80%, Microsoft's Azure 40%. If the customers are making money on the equipment, the orders keep coming.
Simpson makes it his final trade on the back of the Hugging Face acquisition. Worth noticing what nobody claims: neither of them argues the stock is cheap. The argument is that the spending is durable.
In short: Referenced only as the chips China was cut off from — "they couldn't get the latest and greatest chips from Nvidia. So, they had to get smart… necessity is the mother of invention" — which is why Chinese labs lead on low-cost models. No stock view in this appearance.
25:56So now I have to go from supporting it to putting it on ice or being against it." [clears throat] So on the energy front, that's clearly true. I think on the model front, the Chinese focused on, hey, they couldn't get the latest and greatest chips from Nvidia. So, they had to get smart and go, I always say necessity is the mother of invention.
In short: Explicitly exempted from the semiconductor problem and defended on valuation against the Cisco analogue: "Nvidia is trading at 15 times. Not that I think Nvidia is where the problem is"; "trading at a 15 times PE off of calendar 27 numbers with the S&P up at 19 times, you go, this isn't Cisco in 2000." The print itself was constructive — "the Nvidia report made me feel good in that okay, yeah, I do think there's another year for this to run" — with one criticism, the new long-term guidance: "Nvidia for its entire history has guided out one quarter in advance." Counterweight: its credit default swaps are now wider than the 5-year North American IG average.
Nvidia makes the chips the AI build runs on, and Niles deliberately exempts it from his own semiconductor warning: "not that I think Nvidia is where the problem is." His defence is valuation. The obvious historical parallel is Cisco, the networking company that was the most valuable business in the world in 2000 at over 100 times earnings. Nvidia trades near 15 times next-year earnings against an S&P at 19 — a very different starting point, so "even if they're wrong, it's not Cisco going 20,000 in terms of the downside."
The quarter itself left him constructive: the report "made me feel good in that okay, yeah, I do think there's another year for this to run," even if he won't commit to the magnitude.
His one criticism is about behaviour, not numbers. Nvidia has always guided one quarter ahead and has now given multi-year revenue guidance instead. He understands why — management is frustrated that the stock's multiple lags peers — but he distrusts the instrument, and his example is again Cisco, which promised 30–50% sustainable growth in 2000 and then reported two straight years of falling revenue.
The one genuinely new caution is in the debt market rather than the equity market: the cost of insuring Nvidia's bonds is now higher than the average for large US investment-grade companies. That is an early-warning signal about the financing of the whole AI complex, not a judgement on the chips.
29:42I mean Cisco Systems in 2000 was saying hey, we think the internet can, we think our revenues can grow 30 to 50% sustainably in the future, and then two years after they made that proclamation their fiscal year revenues were down for two consecutive years in a row. So things change. And the Nvidia report made me feel good in that okay, yeah, I do think there's another year for this to run, but the magnitude we'll have to wait and see how that shakes out. But trading at a 15 times PE off of calendar 27 numbers with the S&P up at 19 times you go, this isn't Cisco
In short: Genuinely two-sided. The chart is in his bottoming group ("Nvidia, AMD, couple of other semi names — I'm seeing some charts that are bottoming"), but he is explicitly waiting for the top signal and it is not fundamental: "I don't really think of things in terms of fundamentals… the leather jacket guy said they were growing at 70% and the stock ripped… what I've been waiting for for the last six months is for that second derivative of growth to change… you see the growth rate start to come down to 60 or 50%. And that's when the stocks are going to top." Ownership is the other flag — a student's entire portfolio is "50% Nvidia and 50% Broadcom," which he suspects is typical of US retail, and "they're probably not going to sell at the highs."
Nvidia makes the chips that nearly all AI training and inference runs on, and has been the single largest beneficiary of the AI build-out.
Dillian's view is genuinely split, which is why it is graded Neutral rather than forced either way. The chart is in his bottoming group. But he is explicit that he does not value the company on fundamentals — "the leather jacket guy said they were growing at 70% and the stock ripped" — and that what he is watching for is a specific, mechanical signal: the second derivative of growth. That means the rate at which growth itself is changing. Revenue growing 70% is fine; revenue growth decelerating from 70% to 60% to 50% is the tell. "That's when the stocks are going to top."
The second half of his caution is about who owns it. A student of his holds a portfolio that is 50% Nvidia and 50% Broadcom and nothing else, and he suspects that is common across US retail. Crowded ownership does not make a company worse, but it changes what happens on the way down — "they're probably not going to sell at the highs."
10:58I mean, for sure, Nvidia is growing at 70%. What I've been waiting for for the last six months is for that second derivative of growth to change. And you see the growth rate start to come down to 60 or 50%. And that's when the stocks are going to top. So — yeah, Nvidia — I think earnings are accelerating.
In short: Defended by Jim Lebenthal on two grounds and used by everyone else as the industry's reference point. On the supply-vs-competition question raised by Broadcom's shortfall he sides with NVIDIA: "what NVIDIA is saying is that the whole space is supply constrained… they could do more than 70% if supply constraints come out of the way, particularly on memory. If there is competition, say from AMD versus Broadcom, particularly with the Alphabet partnership, then it really doesn't matter… there's enough of this pie and it's growing big enough to satisfy everyone." On the Taiwan-Semi comparison he adds the durable edge: "NVIDIA has a special secret something as well with the whole ecosystem… I'm talking about CUDA and the software that supports the chips… that really provides a moat." And the near-term view: "the momentum right now to me in NVIDIA is really quite clear… even with September being the month that it is, there's a chance this takes out a new high in this month." Wapner's counterweight: NVIDIA's own blowout quarter "didn't really lift a lot of other boats."
The useful thing here is not the stance but the question Lebenthal poses, because it resolves most of the argument about AI semiconductors in one line: NVIDIA says the industry cannot make enough chips; Broadcom's results imply it is losing business to rivals. Both cannot be the main story.
If NVIDIA is right and the constraint is supply — particularly memory — then competition is close to irrelevant: everything anyone can build gets sold, and a competitor winning a design does not cost the incumbent a sale it could have filled anyway. If Broadcom is right and the constraint is demand, then every rival win is a real loss. Lebenthal sides firmly with supply, which is why he does not treat AMD's or Broadcom's Alphabet partnership as a threat.
His second point is the durable one: CUDA, NVIDIA's software layer. Chips get faster every year, but the code customers have already written for a particular chip is expensive to rewrite — that switching cost is the moat, and it is not something a rival fixes with better silicon. Wapner's caution stands alongside all of it: NVIDIA's own outstanding quarter did not lift the rest of the AI complex, so a good print no longer guarantees a good tape.
In short: Referenced twice, both times as evidence rather than as a position. Terranova uses it as the cause of the current tape: "last week's disappointment surrounding the inability, in particular for high beta momentum, to get that shot of adrenaline off of Nvidia's earnings, the derivative trade, and off of Marvell's earnings — I think that's leading to a significant amount of bearishness as it relates to sentiment." Link uses the guidance as her rebuttal to the data-center moratorium fear: "how does NVIDIA guide from 45% revenue growth by fiscal 28 to 70% if they think that no data centers are going to be built?" No member gave the stock a view of its own this episode.
Nvidia is not given a stance this episode; it appears twice as evidence in other people's arguments, which is itself notable a week after its results.
Terranova cites it as the cause of the current mood: last week's earnings, and Marvell's, failed to give the high-momentum AI names the lift everyone expected, and that failure is what turned sentiment bearish. Link cites the same results as her rebuttal to the data-centre pessimists: Nvidia has guided revenue growth up from 45% to 70% for its fiscal 2028 — a company does not do that if it believes the data centres it sells into will not get built.
Both are using the guidance as a fact about the world rather than as a reason to buy or sell the stock.
In short: "Is the AI build-out a scam?" — "definitely not": a scam means selling something that doesn't work, while every business is adopting AI because it's useful. Nvidia's revenue is GPUs; the proof of end demand is OpenAI/Anthropic revenue from customers "willingly paying."
Asked whether the AI data-center boom is a scam that companies like Nvidia are exploiting, Carlson says no. A scam sells something that doesn't work; AI is being adopted by businesses everywhere because it saves time and money, and companies keep increasing how much they use it after trying it. The clearest proof is not Nvidia's chip sales but the revenue of AI companies like OpenAI and Anthropic, which comes from customers choosing to pay. He compares it to the arrival of the internet and calls AI the most useful technology he has seen since.
In short: "They do keep delivering" and remain the bellwether for the whole AI movement — but it is alone 18% of expected 2026 S&P earnings growth, and "it's hard to extrapolate that with an infinity sign. At some point there's going to be some sort of miss."
Nvidia makes the chips that AI runs on, and its quarterly report has become the health check for the whole AI boom. Her verdict on the latest one is straightforwardly positive: "they do keep delivering," and the capex numbers the market wants keep showing up.
The worry is a maths problem, not a business problem. Nvidia by itself accounts for 18% of the entire S&P 500's expected 2026 earnings growth. When one company carries that much of the index's profit growth, "it's hard to extrapolate that with an infinity sign" — eventually growth slows simply because last year's number was so big (what she calls base effects), and a single high-profile miss would force analysts to cut estimates across the whole index. Her governing rule applies: better-or-worse matters more than good-or-bad.
27:19We just had Nvidia late last week with their numbers. I mean, they do keep delivering in a quite remarkable way. They do. And clearly that's a bellwether for the whole AI movement. And spelled a different way, the weather forecast is pretty good per Nvidia. You've got Broadcom coming up tomorrow as we're taping this, less of a bellwether than Nvidia, but I think the combination of the Nvidias, the Microns, the Broadcoms of the world, they do continue to
In short: The subject of the show and the desk is genuinely split, which is why the net read is neutral. The print was fine — "they validated the AI story as much as they had to" — and the guidance was extraordinary: "they guided to like 70% revenue growth… they looked out to '28, which they never do, and the street was at 44," with Melius today arguing the case to go higher has strengthened and a buyback outstanding. Terranova even sees it re-taking leadership: "NVIDIA is stepping forward into leadership and you're seeing some of the other Mag 7 names participate" — which he says kills the broadening-out narrative he is positioned for. Against that, Talkington raises the show's sharpest objection twice: "NVIDIA should have rallied but didn't. You have to ask the hard questions — why aren't these stocks rallying?" and, on quality of earnings, "I did not like that 60% of their revenues this last quarter were accounts receivable. Why is that? Why can't these people just pay their bills? Why does that need to be such a big number?" Saccocia's version is that demand was never in question and the print changed nothing because "this has been essentially priced in." Harrington uses it as the reason to expect nothing from Broadcom.
Nvidia reported last week and the results were, by any normal standard, excellent — management guided to roughly 70% revenue growth and, unusually, gave a view all the way out to 2028 when Wall Street analysts had only pencilled in about 44%. And the stock did nothing. That is the entire puzzle this show is trying to solve.
Joe Terranova's reading is that this is about who owns the stock rather than how the business is doing. Funds that buy whatever has been going up (the "momentum" strategy) have been selling AI names for two months; when the single biggest possible piece of good news lands and the price still will not rise, that tells those funds their exit was correct, so they sell more. Awkwardly for him, he also thinks Nvidia is quietly re-taking market leadership — which would kill the "the rally is broadening out to other stocks" idea he is positioned for.
Bryn Talkington raises the one genuinely company-specific worry of the hour. About 60% of last quarter's sales had not actually been collected in cash — they sat as money customers still owe. In a business selling to the richest companies in the world, she wants to know why: "why can't these people just pay their bills?" It is not an accusation, it is a flag to watch, because a growing gap between reported sales and cash received is the classic early warning that demand is being pulled forward. Nobody on the desk is bearish on the technology; the disagreement is over whether the stock can be led by news any more.
In short: Everything in the AI complex trades on the same two drivers (hyperscaler capex + lab ARR) at wildly different multiples — the Gavin Baker point. Nvidia and Micron "are really not extrapolating. In fact, they're telling you we're at peak," so "the low multiple names across the AI trade in general are going to do better."
His point about Nvidia isn't about chips at all — it's about what's already in the price. Almost everything in the AI complex, from semiconductors to power equipment to AI-adjacent service companies, ultimately rises and falls on the same two things: how much the big cloud companies spend, and how fast the AI labs' revenue grows. Yet those names trade at wildly different valuations for the same underlying bet.
Nvidia and Micron, he says, are the ones "really not extrapolating" — their share prices imply this is roughly as good as it gets. The expensive end of the complex — behind-the-meter power plays, services businesses, "hidden AI winners" that retail has piled into — is priced for a much rosier future. So his mildly contrarian conclusion is that the low-multiple names should do better from here, because you're being paid the same exposure at a much lower price.
11:02the same underlying drivers and they trade at hugely different multiples and I think Gavin Baker's made this point which makes sense. You've got a lot of behind the meter, you've got some services companies that are trading at pretty high multiples on outyear numbers and then frankly you have things like Nvidia and Micron and the well-known names that are really not extrapolating.
In short: The best quarter a public company has ever printed, and a stock that could not hold it. The print: "96.2 billion in revenue, up 106% year-over-year, driven by 89 billion in data center revenue and a surge in backlog commitments to 279 billion, up 119 billion quarter-over-quarter," roughly $60 billion of net income, adjusted EPS $2.22, gross margin 75%, Q3 guided to $108B vs $104.2B. CFO Colette Kress projected 70% revenue growth through fiscal 2028 — against consensus of 47% — with Vera Rubin in full production, and "they're calling it a supply constrained market." The result: "Nvidia added 453 billion in market cap in a single session, the largest one-day market cap gain in US history." Then it fell. His three reasons: Warsh's hawkish speech; the White House chip tariffs, which "came in at such a bad time… the stock index basically gave up most of its gains and ended up down 2.3%"; and the structural one — "a lot of Nvidia's growth is actually being financed by itself, right? Through these circular capital arrangements… people are not as confident, even though the number is spectacular, that if the credit markets unravel whether that number can actually be hit. And that's an uncertainty at the moment. It's not the base case, but it's still a big uncertainty." Note the dependency he names out loud: "some of that revenue growth is contingent on financing for Anthropic doing an IPO in October and OpenAI securing financing," and Nvidia's own CFO put AI-lab demand supported by its balance sheet at ~25% of next year's business. Mizuho: "results were okay, but not enough." Cheapness marker from Raymond James (target to $352, then $515): "Nvidia is trading at less than 15 times 2027 earnings, below the S&P 500 at 18.6 times."
Nvidia reported what is, on the ordinary reading, the best quarter any listed company has ever produced: $96.2 billion of revenue, double last year's, roughly $60 billion of profit, and — the number Singh flags as the real one — customer commitments jumping from $119 billion to $279 billion in a single quarter. The chief financial officer then guided to about 70% revenue growth through the 2028 financial year, against Wall Street's expectation of 47%. The shares added $453 billion of market value in one session, the largest one-day gain in American history.
And then handed it back. Three reasons. Warsh's hawkish speech pulled every high-multiple stock down. The White House chose that moment to float tariffs on semiconductors extending beyond chips to servers, laptops and games consoles — "horrible timing." And the third is structural.
That third reason is worth understanding properly. Nvidia has invested nearly $50 billion into the AI laboratories that then buy its chips, and told investors that demand supported by its own balance sheet will be roughly a quarter of next year's business. Money leaves Nvidia, arrives at a customer, and comes back as revenue. That is not fraudulent — the chips are real and so is the demand — but it means the growth forecast depends on the credit markets continuing to fund those customers. Singh says it plainly: part of the 70% guide "is contingent on financing for Anthropic doing an IPO in October and OpenAI securing financing," and if credit tightens, investors doubt "whether that number can actually be hit." He is careful to add: "It's not the base case, but it's still a big uncertainty."
The valuation, for context, is not the problem. Raymond James notes Nvidia trades at under 15 times its expected 2027 earnings, cheaper than the S&P 500 at 18.6 times. What is being repriced is not the earnings but the confidence in who is paying for them.
Full passage: premium transcript (PDF).
In short: "Can't build fast enough." Q2 FY27 revenue accelerated 106% Y/Y to $96.2B ($4.1B beat) — Data Center +117% to $89.0B, Edge Computing +27% to $7.2B — with gross margin 75% (+3pp), operating margin 66% (+5pp), non-GAAP EPS $2.22 ($0.13 beat), operating cash flow +57% to $24.1B and free cash flow +59% to $21.3B against $99.4B of cash and $33.4B of debt. Q3 guided to $108.0B (+12% Q/Q, +89% Y/Y, a $3.4B beat) at a 74% gross margin, assuming no Data Center compute revenue from China. Growth accelerated (85%→106% total, 92%→117% Data Center), adding ~$15B sequentially with another ~$12B guided; demand broadened — Hyperscale $48.7B while AI Clouds, Industrial and Enterprise grew faster to $40.3B (neoclouds, enterprises, sovereign AI alongside Big Tech). Rubin is arriving without a digestion pause: Blackwell Ultra still ramping while Rubin ships with purchase orders "from every major hyperscaler, AI cloud, and system OEM," and it raises the take per AI factory — ~$18B/GW (Hopper) → ~$25B/GW (Blackwell) → ~$40B/GW (Vera Rubin), delivering 30x throughput per megawatt and 35x lower token costs than Grace Blackwell Ultra. The costs: gross margin bends to 74% in Q3 and ~71%–72% in Q4 on higher memory costs ("supply-driven rather than a sign of weaker demand or pricing"), and supply and capacity commitments jumped from $119B to $279B in three months, largely to secure memory — "a huge vote of confidence in future demand—and a much larger commitment if the cycle eventually slows." NVIDIA as financier: nearly $50B invested in frontier AI labs, partnerships with major financial institutions helping raise >$500B of third-party capital, plus credit support and take-or-pay commitments — "critics call this circular financing. NVIDIA argues it is simply removing a capital bottleneck," with the strategic benefit of locking in deployments "before competing silicon reaches scale"; NVIDIA says labs receiving balance-sheet support could be roughly one-quarter of its business next year. Jensen Huang's answer to custom silicon: "whereas many of these XPUs are inference-specific chips for one cloud or one service, NVIDIA is a platform, an entire AI factory platform that spans the entire AI life cycle that you can use in any cloud" — and on demand: "when the world goes to agentic, fully agentic systems, you are going to have agents running all the time, working with other agents running all the time," which would decouple inference demand from human usage. Outlook: ~70% revenue growth in FY28, "far above prior Wall Street expectations," adding more than $200B of annual revenue. Valuation: up ~20% YTD, still only ~20x forward earnings, "well below the multiple it commanded earlier in the AI boom and below the rest of US Big Tech"; Q2 13Fs showed hedge funds not accumulating as they used to, many still underexposed against its 8% S&P weight. "In semiconductors, a low P/E can sometimes signal peak earnings rather than a bargain… So far, that skepticism has been repeatedly proven wrong." Bear: spending outruns the profits it generates while custom silicon takes inference share. Bull: "agentic AI keeps expanding compute demand faster than efficiency gains and competition can reduce it." "So far, NVIDIA is still winning that race." A disclosed author holding.
NVIDIA sold $96.2 billion of chips and systems in three months — more than double the same quarter last year. What makes that remarkable is not the size but the direction: growth sped up, from 85% to 106%. Big companies almost never do this. The bigger you are, the harder each extra percentage point becomes, because you have to add more absolute dollars to move the same rate. NVIDIA added roughly $15 billion of revenue in a single quarter and expects to add another $12 billion next. And management says next fiscal year should grow about another 70% — more than $200 billion of new annual revenue — while still being limited by how much it can physically build, not by how much customers want.
The most useful new framing in this issue is dollars per gigawatt. AI data centres are ultimately constrained by electricity, so the industry measures itself in gigawatts of power. NVIDIA's revenue from each gigawatt of AI capacity has gone from about $18 billion with its Hopper generation, to $25 billion with Blackwell, to about $40 billion with the new Vera Rubin systems. That matters because it means NVIDIA can grow even when the world can't build power plants any faster — it simply extracts more from each one, by selling not just the chip but the CPUs, networking and software around it.
Two complications are worth understanding properly. The first is what critics call circular financing. NVIDIA has invested about $50 billion into AI labs and helped arrange more than $500 billion of outside capital for AI infrastructure, sometimes backing customers' borrowing directly. So some of the money that comes back as NVIDIA revenue was, in a sense, put there by NVIDIA. The company's defence is that its customers' demand is growing faster than their balance sheets, and it is removing a financing bottleneck rather than fabricating demand. Both can be true. The number that keeps it honest is NVIDIA's own: labs receiving balance-sheet support could be about one-quarter of next year's business. That is the share of revenue that depends on those customers eventually making money from what they're building.
The second is customers building their own chips. OpenAI just published results for its own inference chip, Jalapeño, claiming 1.5 to 1.9 times more work per watt of electricity than the NVIDIA systems it tested. Inference — running a trained model to answer questions — is repetitive and predictable, which makes it the easiest workload to design a specialised chip for. Jensen Huang's answer is not that customers won't do this, but that their chips are narrow: built for one workload in one cloud, while NVIDIA sells a general-purpose platform that handles training, inference and networking anywhere. The real question is whether that flexibility is worth NVIDIA's much higher prices as the workloads become more standardised.
Two more things to hold in view. Profit margins are set to slip from 75% to about 71–72% by the fourth quarter — but because memory chips have become expensive and scarce, not because customers are pushing back on price. And NVIDIA has committed to $279 billion of future supply purchases, up from $119 billion three months ago, mostly to lock up that memory. That is an enormous vote of confidence — and an enormous obligation if demand ever cools.
Finally the valuation, which is the counterintuitive part. After all this, NVIDIA trades at roughly 20 times expected earnings — cheaper than Apple, Microsoft or Google. That sounds like a bargain, and the author flags exactly why it might not be: "in semiconductors, a low P/E can sometimes signal peak earnings rather than a bargain." Chip companies are cyclical; the market often prices them cheaply right before profits fall, because the "E" in the ratio is about to shrink. A low multiple can be a warning dressed as a discount. The author's honest summary is that the debate is no longer whether the stock looks cheap, but how durable these extraordinary earnings are — with the bear case (spending outruns the profits it generates while custom chips take inference) and the bull case (always-on AI agents expand demand faster than efficiency reduces it) laid side by side. "So far, NVIDIA is still winning that race." The author owns the stock. Analysis, not a recommendation.
In short: The scarcity story doesn't square with the utilization data: one GPU warehouse study shows usage "only sitting at around 35–40%… even at peak loads," while used A100 / Nvidia GPU prices are bid up on scarcity. His read is hoarding — "double and triple ordering going on because everyone's terrified… they'll be caught short" — which "represents a kind of potential flux of product into the market at some point in the future." Demand that is inventory, not consumption.
Nvidia sells the chips that data centers run AI on. The bull case has always rested on scarcity — everyone wants them, there aren't enough, so prices stay high even for second-hand ones.
Kedrosky points at a number that doesn't fit that story: in at least one big GPU rental warehouse, the chips are only busy 35–40% of the time, even at the busiest hours. If they were genuinely scarce, they'd be running flat out. His explanation is hoarding — buyers are ordering two and three times what they need because they're terrified of being caught short if demand suddenly explodes. That's stockpiling, not consumption.
Why it matters: stockpiles eventually come back the other way. All those idle chips are "a potential flux of product into the market" — a wave of supply that shows up exactly when buyers stop panic-ordering. Demand built on fear of shortage unwinds fast.
17:08I just find that really interesting, the way these processes get divorced from the underlying economic realities. I mean there was a great study came out the other day showing that in one warehouse — sort of think of it like an AWS of GPUs — that the GPU usage is only sitting at around 35-40%, which should be eye-popping in terms of utilization rates. Because you ask yourself, wait a minute, I was told a story about scarcity and rising prices and that A100 prices in the used market, Nvidia GPU prices, were
In short: Peak-date evidence, not a stance: "NVDA and GOOG back in May" — the most recent of the leadership highs he charts, and the datapoint that sits underneath the concentration warning ("chips near 20% at their highs earlier this year"). No fresh view here on his prior NVDA positioning.
In short: Positioning stated flatly against Michael Burry's short: "I am still long Nvidia and several other tech stocks. I'm less long cuz I've gotten nervous. But I am not short." Hedging is "possible" but undisclosed. The concern is the customer list, not the product — "revenue was up more than 100%. However, 70% of its accounts receivable was from five companies." On the $500B third-party capital facility: "that's not exactly" real — per Bloomberg it was still being negotiated and "Jensen went out and basically just announced it anyway… right now there is no $500 billion."
Nvidia makes the chips that AI runs on. Eisman is still a holder and says so plainly, in direct contrast to Michael Burry, who is short the stock: "I am still long Nvidia and several other tech stocks. I'm less long cuz I've gotten nervous. But I am not short."
His worry is not the product but the customer list. Accounts receivable is money already billed but not yet collected, and "70% of its accounts receivable was from five companies." A supplier whose unpaid invoices sit with a handful of buyers inherits those buyers' financial problems — and those buyers, in turn, get much of their own AI revenue from two loss-making labs.
He also throws cold water on a headline: the $500bn financing facility announced with KKR, Apollo, Blackstone and BlackRock is, per Bloomberg, still unsigned. Negotiations were dragging and "Jensen went out and basically just announced it anyway… right now there is no $500 billion." Worth remembering that an announced deal and a completed one are different things.
20:13I'm less long cuz I've gotten nervous. But I am not short. — Would you consider hedging against your long positions given these mounting risks? — I might. I might. It's possible. — And when we talk about hedges, what kind of hedging do you believe in? — I'd rather keep that to myself for now.
In short: Passing mention — cited as the read-out on AI capex, not as a stock call. The entire reference is one parenthesis: "(Nvidia's second-quarter results reported yesterday indicate this AI spending boom will continue to run white hot.)" Its role in the argument is as a coincident indicator of the second stimulant — the note's case is that "AI-related spending" is "another enormous economic accelerant" whose "$700 billion plus annual expenditures on the great data center build-out" are being funded by "a deluge of hyperscaler debt raises," producing "a reverse 'crowding out'" that drives up the government's own borrowing costs. Nvidia's quarter is the evidence that the spending has not yet slowed. The surrounding view of the capex itself is two-sided but ultimately cautionary — "this pace, too, is highly unlikely to be maintained longer term," against the bullish rebuttal that "the productivity boost from AI will be so mammoth that these outlays will easily pay for themselves," which "may well turn out to be the case," though "there is a substantial risk that returns on both the federal and AI spending will turn out to be disappointing." No valuation, target, rating or recommendation is given on the shares.
In short: Q2 revenue $96B (>2× YoY), Q3 guide $108B, and fiscal-2028 growth guided ~70% while customers would support ~100% — "supply constrained." "I far underestimated the continued demand of this company," yet he stays out: his focus is durability of growth, and Google/Amazon are "a bit more durable."
In short: Three red flags at once: rumours that "a lot of Nvidia chips are sitting in warehouses" for data centres yet to be built (shadow inventory), chips "being used as collateral in some of these deals, which is insanity to me," and the headline $500B Blackstone financing that was only an MOU, "not a contract."
Nvidia sells the chips that AI data centres run on, so its order book is treated as proof the boom is real. Dowd's argument is that the order book is being inflated three different ways at once.
First, double ordering: when buyers can only get half of what they ask for, they start asking for twice what they need — so the backlog counts the same demand more than once. Second, shadow inventory: chips are reportedly sitting in warehouses waiting for data centres that haven't been built, which means they were sold but aren't generating anything. Third, chips are being pledged as collateral for loans — borrowing against hardware whose value depends on the boom continuing, which Dowd calls "insanity."
He adds a fourth caution about headlines: the announced $500 billion of Blackstone financing was a memorandum of understanding, not a signed contract, so it "can just vaporize overnight."
6:53So, there's probably shadow inventory of these chips all over the place. They've been using chips as collateral in some of these deals, which is insanity to me. There was recently about a week or two ago, Nvidia announced 500 billion in financing from Blackstone, some of the investment banks.
In short: Pricing power that reads as a warning as much as a strength. Nvidia is "hiking prices by 15% plus to hyperscalers and other customers" on systems with the Vera Rubin and Grace Blackwell chips, effective on systems shipped early next year, "with memory chip costs soaring" — but "the inability of the industry's most dominant company to hold the line on prices or absorb the growing costs shows how much leverage makers of memory chips… have." Hence the margin call: "the company has a gross margin of 75%, but that's likely a peak margin." He also flags the second-order effect — the hikes "add complexity to the industry's massive AI data center build out ambitions" alongside power delays, project delays and labour shortages. Into the print: UBS expects a $3-4B beat ($94-95B) with $107-108B guided, Rubin ramping toward 500,000 GPUs a month; Morgan Stanley says the stock is underowned by active funds. Elsewhere in the week Nvidia put $1.5B into SB Energy for 4.25 GW behind OpenAI's Ohio campus and discussed investing in Mercor at a $20B valuation.
Nvidia designs the processors that train and run AI. This week it told its biggest customers — Microsoft, Amazon, Google, Meta — that the price of its AI servers is going up more than 15%, on the systems shipping early next year.
A price rise usually reads as strength. Singh reads it both ways. The company can raise prices because demand exceeds supply, which is genuinely good. But the stated reason is that memory chip costs are soaring and Nvidia cannot absorb them — and a business with 75% gross margins that still has to pass costs on is telling you that its suppliers, not it, hold the whip hand. Hence his conclusion that 75% is "likely a peak margin": the most profitable quarter is probably behind rather than ahead.
There is a second-order cost too. The customers building data centres are already fighting power shortages, construction delays and labour scarcity; a 15% price increase on the most expensive component makes those projects harder to justify. That is the seed of a slowdown in the very spending that has driven the shares. The stance is neutral, not negative: demand is real, analysts expect a $3-4 billion revenue beat, and big funds are still underweight the stock.
Full passage: premium transcript (PDF).
In short: Carlson analogy: the stock investors "missed" whose overflow demand made AMD a catch-up trade — the frame for Zeta as a second bite at the Palantir apple.
In short: The subject of the show, and nobody is bearish — the fight is over whether the print matters. Setup: NVIDIA is having its best month since April, +16%, leading the mega caps this year, and reports Wednesday. Josh Brown has the day's framing: "Jensen Huang is the AI Fed chair. He is to the earnings growth of the S&P what Kevin Warsh is to interest rates." The numbers he cares about: "expected revenue of 92 billion… this is the largest company in the world. That would be a 97% year over year increase — not in earnings, in sales. It's insanity. I cannot believe this stock sells at a near market multiple. The earnings will also be up 99%." But the trade isn't NVDA: "there are two to 300 large cap stocks where what NVIDIA has to say about 2027 and 2028 on the call will absolutely have an impact… they will trade up or down based on what the AI Fed tells us." Harrington is the other side: "I don't think anything that NVIDIA says next week is going to matter to NVIDIA. We already know what they're going to say… it's predetermined right now." Sechan: "NVIDIA is always important — they've continued to deliver again and again… I think NVIDIA is a safe bet." Raskin (owner since 2014): "the numbers are going to be great… I'm more interested in what the stock does than what the numbers are, because that will tell you how optimistic people are." Her ask of the call: stop talking backlog and "prove out the end case — what are the killer apps besides coding?" BMO initiates at outperform, $340 target (57% upside).
Nvidia reports next Wednesday and nobody on the desk expects bad news. Wall Street is looking for about $92 billion of revenue in the quarter — for the largest company in the world, that would be sales roughly double what they were a year ago, with profits up a similar amount. Josh Brown's reaction is that a business growing like that trading at roughly the same multiple of earnings as the average S&P company makes no sense to him.
The interesting argument is about whether the report still moves anything. Jenny Harrington says no: everyone already knows Nvidia will say demand is enormous and the order book is full, so the stock usually drifts either way afterwards. Brown's answer is that the report is not really about Nvidia — it is about everyone else. He calls Jensen Huang "the AI Fed chair": in the same way the Fed chair's words reset the price of money, Nvidia's commentary on 2027 and 2028 resets the assumed demand for hundreds of other companies that build, power, connect and cool the machines. Those companies are not small — he lists $100 to $300 billion businesses most people cannot name.
The useful nuance is what a good print now requires. Scott Wapner's framing is that the market no longer needs a revelation, only confirmation — and Brown agrees, because multiples have actually contracted this year while prices rose, meaning investors are not extrapolating. Amy Raskin, who has owned the stock since 2014 but is not overweight, wants one more thing: proof that customers are getting real productivity out of AI beyond writing software. That evidence, she says, has still not arrived.
In short: Referenced as evidence of the credit-creation phase of the bubble, not as a company call: "you got NVIDIA, Jensen bragging about creating a compute security derivative, whatever. That's fine. No judgment. I know what that is. That's making it easier to get more credit to them. So the bubble's not over yet." The host separately argues the track is laid for $300 billion of operating profit.
1:08:55goes over the South Sea Bubble and John Law's Mississippi bubble. And what you take from that is when these things happen, you have to keep applying new credit. You have to keep supplying new credit to keep the thing going. And so when I see headlines last week where the SEC is changing the rules around securitizations for anything related to AI, where you're getting all the big private equity guys in a room to securitize, and you got NVIDIA, Jensen bragging about creating a compute security
In short: The scale illustration for the passive-flow argument, not a call on the stock: "Nvidia's market cap is how big? 5.7 trillion… that means Nvidia's market cap equals 57,000 little 100 million-dollar businesses" — against only ~20,000 listed companies in the world. It sits inside his tongue-in-cheek observation that buying any $600bn+ Nasdaq/S&P name has "a 100% success rate" — the easiest money of the last 17 years, and the thing draining the small caps he buys.
Summerton isn't making a call on Nvidia — he's using it as a measuring stick. At a $5.7 trillion market value, Nvidia alone is worth the same as 57,000 companies of $100 million each, in a world that only has about 20,000 listed companies in total. That single comparison is his whole argument about where money has gone.
His explanation is passive investing. When someone sells shares in a small company and puts the money into an index fund, that money is automatically distributed by size — so it flows overwhelmingly into the largest names and none of it reaches the small ones. Repeat for seventeen years and you get his tongue-in-cheek "100% success rate" observation: every company that has crossed a $600bn market value on the Nasdaq or the S&P has gone on to make investors money, averaging 90%, with no exceptions — while companies that reached the same size outside those indices did not. The index membership, not the business, is doing much of the work.
He is explicit that this has worked and is the easiest money of the era. The point is the mirror image: the same flow has left thousands of small, profitable, capital-starved companies unloved and demotivated — which is where he goes hunting.
2:11And this big tick is called passive investing. And passive investing is sucking the lifeblood out of all these little companies in the rest of the world. So, just take this. Nvidia's market cap is how big? 5.7 trillion. Okay. 5.7 trillion. That means that Nvidia's market cap equals 57,000 little 100 million-dollar businesses.
In short: Not traded today, but it bookends the show. Talkington's earnings caveat: within tech "it's really NVIDIA and Micron skewing that even more — the earnings are outsized," and the thing to watch a year out is the rate of change of earnings, not the level. It is also one of the three suppliers inside Alphabet's TPU ecosystem in Simpson's Alphabet case (Broadcom, NVIDIA, Marvell). And Santoli sets the calendar: with semis down 5% on the week and "caught up in their own dynamic of liquidation, oversold, bounce, resistance," the next macro event for the group is Nvidia's report — "we'll see a week from today, right? NVIDIA. That's going to be the next big moment."
In short: Third of the named run-up examples — "AMD and Intel and Nvidia, all of those have that same characteristic… these semiconductors can really correct on you in a big big way." (The host separately volunteers that she sold hers at 4x and watched it run; Oakley's reply: "I wouldn't be too hard on myself if I had a 4x.")
Nvidia designs the chips at the centre of the AI build-out and is the third name in his run-up list: "all of those have that same characteristic." The stance follows directly from his hyperscaler argument — if the companies buying these chips have overshot on capacity, the demand that justifies today's price doesn't hold twenty-four months out.
The exchange around it is instructive on sizing rather than timing. The host confesses she sold hers at four times her money and watched it keep running; Oakley refuses to treat that as a mistake — "I wouldn't be too hard on myself if I had a 4x." Taking a large gain and being early is, in his framework, a good outcome, not a failure.
9:16But any particular stocks in that space, any particular names? — Well, all of the ones in the group itself. If you look at AMD and Intel and Nvidia, all of those have that same characteristic. They've been run up, a lot of money going into them pushed them much higher than they have been, and really Intel hadn't gone up much at all until this last run.
In short: Still core, no longer the marginal buy — the headline of the whole piece is "the AI trade broadened beyond NVIDIA." NVIDIA remains a top-five holding at five of the 20 funds (behind AMZN/TSM at nine and GOOGL at eight), but "NVDA itself was a top buy for only one fund." Held rather than added: the incremental dollar went to TSM, Cerebras, memory, storage and semi equipment.
NVIDIA sells the accelerator chips that train and run most large AI models, and has been the default way to own the AI theme since it began.
The Q2 filings show it moving from the trade to a holding. It is still a top-five position at five of the 20 funds — but it was a top buy for only one, while Taiwan Semiconductor and Cerebras were top buys at four each, and Applied Materials, Micron and AMD at three each.
That gap between "widely held" and "actively bought" is the whole headline: the funds have not turned against NVIDIA, they have decided the next incremental dollar is better spent on the suppliers, the memory and the alternatives. Nothing here says the business is deteriorating; it says the crowd has already arrived.
In short: The framing anecdote, told by Slegers about the guest. "He even made a bet with me that Nvidia was undervalued… 3 years ago. Everybody was already convinced that Nvidia was overvalued. Kris wasn't. Nvidia is up almost +400% since then." Also listed among David Gardner's hundred-baggers. Note the tension with Part I, where Heyndrikx's own worst mistake was skipping Nvidia as an official pick for being "too big at around $60 billion."
Nvidia appears here as a story rather than an analysis. Three years ago, when the consensus was that the shares were expensive, Heyndrikx bet Slegers that they were undervalued; the stock is up roughly 400% since. It is offered as evidence that a growth investor who focuses on how fast the business is expanding can be right when a valuation-first investor would have passed.
Worth reading alongside the first half of the interview, where the same investor's biggest admitted mistake was also Nvidia — he declined to make it an official pick years earlier because at around $60 billion he judged it too big to multiply. It went on to rise 8,720%. The two anecdotes together say something useful: the framework's misses and its hits came from the same place, a judgement about how much room a business still has.
In short: The August 26 print is the sector's read-through and he lays out the checklist: data-center revenue (85-90% of the top line) and its sequential growth, the Blackwell → Vera Rubin ramp, forward guidance for the next two-to-four quarters, GAAP gross margin against the 73-75% range (TSMC packaging and HBM costs are the pressure), hyperscaler capex cross-reference, the Singapore shipment investigation, Spectrum-X/InfiniBand and CUDA licensing. Separately: it "is raising 500 billion alone to finance its own GPU purchases" — circular financing that "works in the short term… eventually people will question it."
Nvidia's August 26 results are, in his framing, the read-through for the whole AI complex — so he lays out exactly what to check: the data-center segment (85-90% of revenue) and whether its quarter-on-quarter growth is holding, the transition from Blackwell chips to the next generation, forward guidance covering the next two to four quarters, and whether gross margins stay in the 73-75% range against rising packaging and memory costs.
The bigger reservation is financial engineering. Nvidia is raising $500 billion to help finance purchases of its own chips — money that comes back as revenue. That kind of loop "works in the short term," and he expects Nvidia to keep beating for a couple of quarters. "At some point next year, the market will say — how long is this sustainable?"
29:40I would look to see how the compute business does for Nvidia. They call it a data center revenue and product architecture ramp coming up as well. The data center revenue accounts for like 85 to 90% of Nvidia's top line. Because it's been growing so much faster.
In short: He walks Jensen Huang's op-ed page by page and lands on it: "he talks about how this financing is from external parties, it's not circular financing, GPU rental rates are going higher. And I think in the short term, he is right." The mechanism: financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion of third-party capital, up to $125B backstopped by Nvidia, with residual-value support capped at 25% of an opportunity — "by creating this $500 billion financing, he's effectively going to guarantee GPU purchases." The durability evidence he accepts: a 2020-vintage A100 still in commercial use six years later with rising lease rates, one-year H100 rentals $1.70 → $2.35/GPU-hour (Oct-25 → Mar-26), cross-provider on-demand medians $2 → $2.70, and B200s at $5.30-7/GPU-hour — "their value is not fixed at installation… the installed base remains productive well beyond its initial depreciation period, which is one of the bear's main theses." His own caveat on the ROI answer: the usefulness of AI "needs to become revenue." Deck pages 19-20.
The bear case on Nvidia is "circular financing": Nvidia funds the companies that buy Nvidia chips, so the revenue is really its own money coming back. Jensen Huang answered it directly, and Singh spends fifteen minutes on the answer because he finds most of it convincing for now.
The structure: six of the largest capital allocators in the world — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — are setting up platforms to lend over $500 billion to companies that want to build AI data centres. Each lender underwrites its own deals. Nvidia's role is limited to guaranteeing what the used equipment will be worth, for at most a quarter of any project. Nvidia gets guaranteed chip demand; the money is genuinely other people's.
The evidence Singh finds hardest to argue with attacks the other bear thesis — that AI chips become worthless quickly. They are getting more expensive to rent, not less: the price to rent an H100 for a year rose from $1.70 to $2.35 per hour between October and March, on-demand rates went from $2 to $2.70, and a chip Nvidia launched in 2020 is still earning money six years later. If the installed base keeps producing revenue long after it is written off, the depreciation argument fails.
His caveat is the honest one and he states it plainly. Huang says the return on all this investment "is in the usefulness of AI." Singh's reply: usefulness "needs to become revenue." He is bullish on the short term and unresolved on the rest.
Full passage: premium transcript (PDF).
In short: Asked what a millennium of financial history says about pricing power: "it tends to erode." He worries about the market's assumption "that nobody else can create GPUs at scale and quality of an Nvidia ever" — that is what is being priced. "Monopolist-like pricing power tends to not last very long. Sometimes it can last for a few years. But it never lasts forever as far as I know." Separately, take its MOU with the big alternative managers as a press release until it settles into actual structures.
NVIDIA is the "picks and shovels" of the AI build-out — it sells the tool everyone digging needs, which is historically the best place to stand in a boom. Wigglesworth does not dispute that. His caution is about what the share price is currently assuming.
Asked what a thousand years of financial history says about pricing power, his answer is flat: "it tends to erode." Extraordinary margins attract competition, because in a capitalist system people respond to incentives. What is being priced into NVIDIA today, he argues, is the assumption "that nobody else can create GPUs at scale and quality of an Nvidia ever" — and NVIDIA itself depends on a supply chain it does not own.
He is not calling a top and gives no timing: "sometimes it can last for a few years. But it never lasts forever as far as I know." Treat it as a reminder that a monopoly premium is a wasting asset, and that the loss of it, whenever it comes, will be a de-rating rather than an accounting event.
14:59Throughout your many many centuries of reading of history, what tends to happen to that pricing power? — It tends to erode. There are so many examples of this. I do worry for example that right now Nvidia is the picks and the shovels distributed to the entire AI revolution. But it obviously depends on its own supply chain, and I do wonder about the assumption that nobody else can create GPUs at scale and quality of an Nvidia ever.
In short: Host-framed, not an Eisman clip — the narrator places Nvidia at the top of the gold-rush stack ("the ultimate pick and shovel merchant… supplying the hardware to the hyperscalers"), one rung above the hyperscalers Eisman defends. Eisman does not name it here; his own NVDA stance is on the Aug 28 page.
2:39Now, why does this context matter? Well, it goes back to the gold rush analogy. During the gold rush, the people that made all the money were those selling the picks and the shovels, not the gold miners themselves. Now, in an AI context, obviously, the ultimate pick and shovel merchant is Nvidia, who is supplying the hardware to the hyperscalers, but next in line are the hyperscalers that are building the data centers for future AI use.
In short: The dividing line between the three models. CRWV is defined as the "purest version of the neocloud model" precisely because "it buys NVIDIA GPUs, installs them in data centers, and rents the compute" — and those GPUs are the collateral behind the debt costing $640M of quarterly interest. CBRS is defined by the opposite choice: "instead of buying NVIDIA GPUs, it designs its own wafer-scale processor." Referenced as the ecosystem's supply anchor; not a stance call. (Also a disclosed author holding.)
In short: Named only as a Helix Platform partner: Vistra is "committing up to $1 billion to a strategic infrastructure partnership with KKR, NVIDIA, and the Kuwait Investment Authority (KIA) to serve as the preferred power partner for at-scale data-center development," in which "NVIDIA provides compute expertise, KKR provides capital and deal sourcing, and KIA provides a long-duration investment mandate." The point being made is about Vistra's contribution, not NVIDIA's: Vistra "supplies the one thing none of them can build quickly: 24/7 dispatchable, carbon-friendly nuclear and gas generation in the markets where AI data centers are being built." No view is taken on NVDA's shares here.
In short: "The big news was Nvidia created a new financing technique for AI data centers" — partnering with six large asset managers on a $500 billion financing push that treats compute infrastructure "much like the commercial real estate toll roads or other assets you can borrow against." His read is constructive, not alarmed: "this is not a brand new concept. It's a tried-and-true way to create a financial infrastructure to support a capex hungry growth industry." It should involve securitizations, means "hyperscaler free cash flow could — I emphasize could — improve," boosts revenue for participating banks and private-credit firms, and shows "there is as yet no slowdown in AI capex."
Nvidia announced a $500 billion arrangement with six large asset managers to finance AI data centres. Instead of the tech giants paying for buildings and chips out of their own cash, the data centre itself becomes collateral you can borrow against — the same way office towers and toll roads are financed, usually by bundling the loans into bonds sold to investors ("securitization").
Eisman's reaction is notable because it is the opposite of alarm. He calls it "a tried-and-true way to create a financial infrastructure to support a capex hungry growth industry" — boring, well-understood plumbing arriving because a real industry needs funding. He flags three effects: the tech giants' free cash flow (cash left after spending) could improve, banks and private-credit firms earn fees for arranging it, and — most importantly — the deal proves there is no slowdown in AI spending yet.
Worth noting this is his counterpoint to commentators who read the same deal as circular, late-cycle financial engineering. His discipline is the same one he applies all episode: don't turn an unfamiliar structure into a disaster narrative, because "predicting total disaster is just too emotionally tempting."
6:21And with respect to AI, hyperscalers continue to increase their capex budgets. And this week, the big news was Nvidia created a new financing technique for AI data centers. Nvidia is partnering with six large asset managers on a $500 billion financing push designed to treat compute infrastructure much like the commercial real estate toll roads or other assets you can borrow against.
In short: The other named absence from Ackman's book — he owns "the hyperscalers, the capex spenders, these digital network companies," but none of the picks-and-shovels AI names whose outperformance he expects to prove temporary.
22:15We know that Bill Ackman has invested in all of these companies, the hyperscalers, the capex spenders, these digital network companies, but he doesn't own any Micron or Nvidia, he doesn't own any Sandisk, he's not in these AI companies. And he believes the outperformance of this subset of companies is likely temporary. He says, while the S&P 500 index has increased by approximately 10% through the first 6 months of the year, nearly the entirety of gains has come from two sectors that provide the picks and shovels for AI infrastructure. Just two of the S&P
In short: Negative on the financing scheme, not the chips: the $500B third-party compute platform only works if the rating agencies bless securitized GPUs as investment grade, which is why Jensen must argue a >5-year useful life. "You're going to take a semiconductor chip that is 5 years outdated… if you don't get paid back, you're basically not secured by anything useful." A mañana problem people only figure out in 3-5 years.
His problem isn't Nvidia's chips — it's the new way the buildout is being paid for. Nvidia just announced a plan with six giant investment firms to raise over $500 billion from outside investors to finance AI data centers. In plain terms: instead of customers buying chips with their own money, Wall Street will bundle the loans against those chips into bonds and sell them to insurance companies and ordinary investors. That's "securitization," the same machinery used for mortgages before 2008.
The test of any secured loan is simple: if the borrower stops paying, what do you get to take back? With a house, you get a house — it's still standing in fifteen years and probably worth something. Here, you get a five-year-old graphics chip that's been run hard and may be obsolete. "If you don't get paid back, you're basically not secured by anything useful." That's why Jensen Huang keeps publicly arguing these chips last more than five years — the entire $500 billion depends on convincing credit-rating agencies to stamp the bonds investment grade so they can be sold widely, and the rating agencies are paid handsomely to agree.
Hayes doesn't think this blows up soon — "people aren't going to figure this out until 5 years from now." He was explicit that the clip would be useless three months out and very useful in three to five years. He also reads the urgency as a signal: five senior principals left the Hamptons in mid-August for this meeting, which tells you the old sources of AI financing have run dry.
36:28In the case of lending on effectively data centers, let's just get to the brass tax of it, semiconductors, Jensen is strongly trying to make the case that the useful life of these assets is more than 5 years. And the reason he has to make that case is because the success of this $500 billion is going to be solely predicated on convincing the rating agencies to put investment grade ratings on this so they can stuff all the insurance companies and all the retail investors full of this product. Okay? And the problem is
In short: The consortium is confirmed: NVIDIA will partner with financial firms to raise $500 billion of third-party capital, having signed MOUs with Goldman, Apollo, Blackstone, Brookfield, KKR and BlackRock, to help its customers finance the cost of compute; terms undisclosed, and the street is "more positive than not," with notes reiterating overweight on revenue sharing "alleviating circularity concerns." Brown: "of all the people in this world that you could choose to contra and bet against, you want to bet against Jensen Huang? I don't." He narrates the logic from Huang's seat: the stock went nowhere for a year on the Cisco-1999 fear — "they think you're giving money to all the competitive local exchange carriers in 1999 to buy Cisco routers with your money, and then you get to book it as revenue." The answer: "we know the world is structurally short compute and probably will be for at least the next five years. What if we can socialize that risk and that upside a little bit and bring in third-party people who just want to bet on compute demand itself having a value?" With trillions of dollars of dry powder in private credit and private equity and a wealth-management world hunting for yield, "let's build a bridge between what Wall Street is looking for — more stuff to invest in with a yield we feel good about — and what Silicon Valley needs, which is spreading out the risk and not having to have Amazon and Google do a debt offering every month. This is the best of both worlds… the ecosystem might have another 10 million new check writers. I think it's brilliant." Sechan: it moves exposure off the balance sheet — "they kept their own exposure at 25%. How does that not de-risk the name?" — and "they've kind of said to the market, compute is now an investable infrastructure asset," which also broadens the customer base to buyers who had financing concerns. Terranova: "for NVIDIA you no longer have to question if they are going to have to cut pricing on GPUs — that's no longer an issue. And for the consortium it's a home run: the compute is the collateral for the debt. Why wouldn't you want that?" His forecast: "in the next 5 years one of the largest futures markets in the world will be oil, and it will be compute" — and when Wapner raises commoditization risk, he argues transparency and hedging ability make it better for everyone, "ultimately a good thing."
The deal flagged as a rumour yesterday is now confirmed: Nvidia has signed preliminary agreements with Goldman Sachs, Apollo, Blackstone, Brookfield, KKR and BlackRock to raise $500 billion of outside money that will help Nvidia's customers pay for the computing power they buy.
The problem it solves is a reputational one. Investors had been worried about "circular financing" — the fear that Nvidia lends customers money which comes straight back as Nvidia revenue, the way Cisco was accused of doing in 1999. Josh Brown's framing of the fix is the clearest on the show: the world "is structurally short compute and probably will be for at least the next five years," so rather than funding customers off its own balance sheet, Nvidia invites outsiders to take that risk and that reward. There are trillions of dollars sitting in private credit and private equity funds looking for investments that pay a yield, and this creates one — "a bridge between what Wall Street is looking for and what Silicon Valley needs, which is spreading out the risk and not having to have Amazon and Google do a debt offering every month."
Rob Sechan's numbers make the de-risking concrete: Nvidia keeps only 25% of the exposure itself. And the deeper implication he draws is that computing power has just been declared "an investable infrastructure asset" — something a pension fund can lend against, like a toll road. Joe Terranova adds two consequences: worries about Nvidia having to cut chip prices are now off the table, and the lenders are protected because the compute itself is the collateral. He goes further — within five years, he thinks, compute will be traded on one of the world's largest futures markets, alongside oil.
In short: "Nvidia lined up $500 billion in financing with a consortium of the biggest private-capital firms, compute itself pledged as collateral and the company backstopping up to 25% of it." With Jensen Huang now calling compute an investable infrastructure asset — "a tidy way of saying the AI buildout has become a credit product" — Mart's point is not the chips but the structure: it makes "the whole edifice more sensitive to credit conditions and stacks another layer of circularity onto a set of deals already drawing hard questions."
Nvidia makes the chips that AI data centers run on. The news Mart picks up is not about the chips — it is about how the buildout is now being paid for. Nvidia has arranged $500 billion of financing from a group of the largest private-capital firms, and the security behind that borrowing is the computing capacity itself: the lenders' claim is on the machines and the revenue they generate. Nvidia has also agreed to stand behind up to a quarter of it, meaning if the borrowers can't pay, Nvidia absorbs some of the loss. Jensen Huang describing compute as "an investable infrastructure asset" is, in Mart's phrasing, "a tidy way of saying the AI buildout has become a credit product."
Why that matters more than the headline number: equity-funded capex only has to answer to shareholders, but debt-funded capex has to be refinanced on whatever terms the credit market offers at the time. So the more of this buildout that runs on borrowed money, the more the whole complex becomes hostage to credit conditions — spreads, lender appetite, the cost of rolling the paper — rather than to AI demand alone. And because Nvidia is simultaneously the equipment supplier and a backstop to the buyers' financing, the arrangement stacks "another layer of circularity" onto deals that were already drawing hard questions: money flows out to customers and comes back as revenue.
This is an observation inside a macro letter rather than a call on the stock — Mart takes no position on Nvidia's price here. His interest is in what it says about the market's load-bearing wall: the same buildout that has driven the index all year is now leaning on the credit cycle, and that is a different, more fragile kind of support than it looked like a year ago.
Full passage: premium transcript (PDF).
In short: Bought 2017 at a split-adjusted ~$2.50 — a +8,720% return — but "never made it an official pick." The stated reason for the omission is the most instructive line in the interview: "I thought it was already too big at around $60 billion, while Shopify only had a market cap of $4.5 billion at the time." The two reasons for buying, in his order: "its founder and CEO Jensen Huang and the fact that they started talking about AI. In 2016 already, the company opened its press releases with 'NVIDIA is the AI computing company.' That really convinced me already back then. Usually, there are signs like that, but you have to follow a company closely to see them."
Nvidia designs the chips that AI systems are trained and run on. The guest bought it in 2017 at the equivalent of about $2.50 a share — a gain since of roughly 8,720% — for two reasons, in his order: he rated the founder, Jensen Huang, and he had noticed that the company had started describing itself, in its own press releases as early as 2016, as "the AI computing company."
That second point is the transferable one. The signal was not in the financial statements; it was in the language the company chose for itself, which only someone reading its releases closely would have seen. He calls it a sign you can only spot by following a company properly.
And then the confession. He never made Nvidia an official recommendation to his subscribers because at a $60 billion valuation he thought it was already too big to multiply — while Shopify, at $4.5 billion, looked like the better shape. He was right about Shopify and wrong about the reasoning, and the wrongness cost his readers the single best investment of the era. Any rule that screens out companies for being large should be read against that.
In short: "Really tap[ping] the capital markets" as AI build-out spending eats free cash flow — the swings around AI concerns favour uncorrelated dividend strategies. Context, no stock view.
22:37So, you know, that benefits folks like us, dividend strategies, when we start to see these more wild swings around concerns over the AI build out, and you're seeing Nvidia and others really tap the capital markets, Meta raising a lot more debt. That's not something they used to have to deal with.
In short: Coming off a monster week (+11%, best since May '25) and reiterated as top sector pick at BofA. Terranova: "I think the stock is in the midst of a breakout" — tracing out the same pattern he identified in Apple several months ago, and the clearest case of new momentum building while Apple's and Alphabet's wanes. It gave back ~2.6–3% intraday on the FT report of a $500B AI-financing partnership with Apollo, Blackstone, BlackRock's GIP, Brookfield, Goldman and KKR; Santoli: "good to have NVIDIA waking up."
Nvidia designs the chips that train and run AI. It just had its best week since May 2025 (+11%) and Bank of America reiterated it as its top pick in the sector. Joe Terranova says the chart is "in the midst of a breakout" — and importantly, it's tracing the same pattern he spotted in Apple months ago, before Apple's big run. His whole framework this episode is watching where new momentum is being built versus where it is draining away; right now it is building in Nvidia and draining from Apple and Alphabet.
The stock still gave back about 3% on the day, on a Financial Times report that six of the biggest names in finance — Apollo, Blackstone, BlackRock's infrastructure arm, Brookfield, Goldman Sachs and KKR — are lining up a partnership to supply Nvidia with roughly $500 billion of financing for AI projects. Investors read a number that large partly as confirmation of demand and partly as a reminder of how much borrowed money the buildout now needs.
In short: Paulo — a tactical trade, his "trade after the trade, like later in the fall maybe": take passive "and turn it up to 11" so the blow-off narrows from Mag 7 to "there can be only one." NVDA is 8% of the S&P and the largest name again, yet "kind of lost its cache… part of the semi-crowd," missed "the big whoosh of Korea and Samsung and Hynex and Micron" — "people have kind of forgotten about it." A pain trade for value, growth (can't overweight 8% of bench) and long/short pods alike; with so little volume moving these inelastic passive-owned names, "we could see like that kind of last 50% move in a month… and then it's over." Candidate for the "world's first $10 trillion company" headline; catalyst = a Jensen reveal (quantum-on-a-chip, robotics) and Le Shrub's predicted Elon deal.
NVIDIA designs the chips that AI models are trained and run on. Everyone knows the company; the interesting part of this conversation is a claim about flows rather than the business.
Paulo's argument runs like this. A huge share of American savings now sits in index funds, which buy the biggest companies automatically, in proportion to size, with no opinion about price. That machine has been getting more extreme, not less. So when a market makes its final speculative push, he expects it to narrow to a single name rather than lift everything — "you go from the Magnificent Seven to there can be only one." His candidate is NVIDIA, precisely because it has gone quiet: it is lumped in with unloved chip stocks, it sat out the recent Korean memory rally, and "people have kind of forgotten about it," even though it is roughly 8% of the entire S&P 500 and the largest company again.
The reason that could move violently is a piece of market plumbing: relatively little stock actually changes hands in these giants day to day, because index funds buy and never sell. When demand meets a very small floating supply, the price has to move a long way to find sellers — "inelastic," in the jargon. Paulo thinks that could deliver "the last 50% move in a month, and then it's over," probably with a splashy catalyst (a Jensen Huang product reveal, or the Elon deal Le Shrub predicts) and a headline about "the world's first $10 trillion company."
Read the label carefully: this is explicitly a trade — "the trade after the trade, like later in the fall maybe" — and it is a description of a blow-off top, i.e. the last leg up before it ends. Both speakers are simultaneously reducing risk into September.
In short: The hub of the circular-financing loop: "Nvidia is considering a 250 billion dollar backstop of OpenAI's infrastructure spending, which they're spending mostly on Nvidia GPUs. OpenAI makes up 40% of Microsoft's cloud backlog and 70% of Microsoft's AI revenue, while Microsoft is 22% of Nvidia's revenue and Nvidia is 17% of Micron's revenue. How circular is that?" Also the winner of Musk's commitment to build SpaceX compute exclusively on Nvidia (Vera called "the best AI computer"), and of Bernstein's 2028 server forecast — but the whole loop rests on hyperscaler capex staying funded by debt markets.
Nvidia sits at the centre of a money loop that Singh keeps returning to. It is considering guaranteeing $250 billion of OpenAI's data-centre spending — money OpenAI would then largely spend buying Nvidia chips. OpenAI in turn accounts for 40% of Microsoft's committed future cloud revenue. Microsoft accounts for 22% of Nvidia's revenue. And Nvidia accounts for 17% of Micron's. "How circular is that?"
Nothing here says the demand is fake — the shortages are real and Elon Musk just committed SpaceX to building its computing exclusively on Nvidia hardware. The point is that each company's revenue is increasingly another company's capital spending, funded by borrowing. If credit markets tighten, the whole ring contracts at once.
Full passage: premium transcript (PDF).
In short: Two committee buys in one show, up more than 11% on the week. Kevin Simpson bought more at 218: "this is the OG, and it is the one stock that has continued to just make more and more and more money every single day, but not have the love from a share-price perspective… in the old days everyone needed electricity, you needed the internet — now the question is how much AI does every company need? They are the primary player in this space. This is not a trade. This is an investment. We're going to own it for a long time." He deliberately bought before the August 26 print "because they tend to report so late that you get a lot of the action from all of the other players beforehand… I think the report's going to be awesome, but I think the move's going to happen beforehand." (It now pays a dividend, "so I think it's worth a consideration.") Stephanie Link bought it for the first time ever, on valuation: "trading at a valuation they haven't seen since 2019, at 18 times — it's actually cheaper than Colgate or Hershey or any of the staples companies that are getting 3% organic growth. This company just posted 85% total revenue growth year over year, and the visibility is outstanding given the backlog." Not hope, she says: "look at their backlog, look at their customer base, and listen to Elon Musk — they haven't even started." Plus mid-70s gross margins and free cash flow set to double next year, "which is why they're financing all these other companies." Talkington: when Elon says SpaceX's massive CapEx is going all in on NVIDIA, "if you don't own NVIDIA after that, then you're missing the boat."
Two committee members bought Nvidia in the same show, for two different reasons.
Kevin Simpson added at $218 and framed it as ownership, not a trade: "this is the OG… the one stock that has continued to just make more and more and more money every single day, but not have the love from a share-price perspective." His analogy is infrastructural — every company once needed electricity, then the internet, and now the question is simply how much AI each one needs. His timing choice is worth copying: Nvidia reports last among the big technology companies (August 26), so much of the information is already public from its customers and suppliers by then. He expects a good report but thinks "the move's going to happen beforehand," so he bought ahead of it.
Stephanie Link bought it for the first time ever, purely on price. Nvidia trades at about 18 times next year's earnings, its cheapest since 2019 — "actually cheaper than Colgate or Hershey," companies growing sales 3% a year, while Nvidia just grew revenue 85%. Asked whether that's hope, she points at the order backlog, the customer list and Elon Musk's commitments: "they haven't even started." Add gross margins in the mid-70s and free cash flow set to double next year — which is also how it can afford to help finance its own customers.
In short: Up 12% over five sessions on Musk's SpaceX-call praise, and the only big-cap tech name yet to report (August 26). Ethridge: he expected exactly this rotation back "into the one that really should matter, because NVIDIA is trading now below the rest of the S&P 500 on a forward multiple — which is absolutely absurd when you consider it's the largest company in the world by market cap" and roughly 90% of AI spend flows through it. "It didn't make sense to me that the stock got so unloved for as long as it did," and it historically trades up into the print — "the surprise will be if it continues to go up right after the print." Brown answers the circular-financing knock: "if you question that, you have to question the entire thing — do we still feel really great about SK Hynix and Samsung or do we not? If you believe in one, you have to believe in the other… it's like pulling the wrong Jenga piece." His framing of NVIDIA's choice: with ten neoclouds building data centers without their own cash flow, "if you're Jensen, do you want NVIDIA chips in there, or Trainium chips?" And it's "the last of the mega-cap techs that's still massively free-cash-flow positive" — $48.5B — "so they still have the ability to lend out dollars in the form of structured loans." Barron's says it finally has momentum and can keep going.
Nvidia has risen 12% in a week and is the last big technology company still to report, on August 26. Malcolm Ethridge's point is a valuation absurdity: Nvidia now trades at a lower multiple of next year's earnings than the average S&P 500 company, despite being the largest company in the world and the one through which roughly 90% of AI spending flows.
The main bear argument is "circular financing" — the worry that Nvidia lends money to customers who then buy Nvidia chips, letting it book the loan back as revenue, the way Cisco was accused of doing in 1999. Josh Brown's answer is a consistency test: the same AI demand that justifies owning memory makers like SK Hynix and Samsung is what Nvidia is financing. "If you believe in one, you have to believe in the other… it's like pulling the wrong Jenga piece out."
He also explains why Nvidia does it. A wave of smaller "neocloud" data-center operators is building capacity without the cash flow of an Apple or an Amazon, so they must borrow — from firms like Blackstone. If Jensen Huang doesn't help, those centers fill with rival chips instead. And Nvidia can afford to: it generates $48.5 billion of spare cash, "the last of the mega-cap techs that's still massively free-cash-flow positive."
In short: The only mega-cap tech name green, up nearly 4% after Elon Musk endorsed it on the SpaceX earnings call ("they have the best AI computer"); Melius loved the endorsement. Lebenthal: "you can buy it right here, just to cut to the chase… new money that comes into accounts at our firm, we're putting in NVIDIA, we're not hesitating" — ~5–7% from an all-time high, roughly 19–20× forward against a 30–35% EPS growth rate, and the endorsement "just continues to show that NVIDIA will have more demand than it can supply for the foreseeable future." Terranova: "it is literally a hanging curveball down the middle" — he bought at 199, got stopped out at 193 while on vacation, called it "the next Apple," and says "I handled the position wrong. It's going to a new all-time high."
Nvidia was the only mega-cap up on the day, gaining nearly 4% because Elon Musk said on SpaceX's earnings call that Nvidia has "the best AI computer" — a customer endorsement from one of the biggest new spenders on AI hardware.
Jim Lebenthal's case is simply that the price no longer matches the growth: at roughly 19–20 times next year's earnings while profits grow 30–35% a year, "new money that comes into accounts at our firm, we're putting in Nvidia, we're not hesitating." Joe Terranova agrees and uses himself as the cautionary tale: he bought at $199, set a stop just below, and got knocked out at $193 while on holiday — right idea, wrong position management. "It is literally a hanging curveball down the middle… it's going to a new all-time high."
In short: The benchmark AMD is now measured against at a new level: with Helios, "AMD is increasingly competing with NVIDIA at the system level rather than selling a cheaper accelerator" — the rack, not the chip, is the unit of competition. (Referenced; not a stance call.)
In short: The balance sheet is no longer clean: CDS on Nvidia debt doubled from 40 bps in June to ~80 bps on reports of >$750B of AI-infrastructure financing — a $250B guarantee backing OpenAI's 10-GW, $500B Ohio hub (a SoftBank subsidiary), talks on $350B of vendor financing for OpenAI's own GPU purchases, and a >$500B SK Group procurement/HBM partnership. "This is a very big circular deal." Offsetting: it traded down to a 15× forward PE (from 25) at the lows, and Jensen Huang is meeting Lutnick as Commerce probes Blackwell leakage to China via Singapore.
The cost of insuring Nvidia's debt against default has doubled in two months. The trigger is what Singh calls "a very big circular deal": Nvidia has agreed to guarantee up to $250 billion so OpenAI can lease data-center capacity, is in talks to lend another $350 billion so OpenAI can buy Nvidia's own chips, and signed a $500 billion-plus partnership with Korea's SK Group.
In plain terms, Nvidia is increasingly financing its own customers' purchases — which flatters today's revenue but means Nvidia carries the risk if those customers can't pay. A company that used to be a debt-free cash machine is now underwriting hundreds of billions of other people's commitments. That's what the credit market is repricing, even though the stock itself got cheap (15× forward earnings, down from 25×) in the sell-off.
Full passage: premium transcript (PDF).
In short: Mentioned once, only as the relative-performance denominator now that the constraint has moved off GPUs: Intel "up over 200% versus an Nvidia that's up like 20% or so." No view on the business, no valuation, and no price target — the "$1420" in the compilation's title is the channel's, not his.
0:00into infrastructure which goes right back to chips and as you know Sarah pointed out with some of those other charts earlier there's shortages in there that are going to continue for the next couple of years. So I think now the bottlenecks have switched. You've got CPU bottlenecks now which is why you've seen Intel up over 200% versus an Nvidia that's up like 20% or so.
In short: The third company named in the index-concentration challenge — the S&P's 49% Information Technology weight is presented as a misallocation of investment capital that cannot represent U.S. economic output, and the Mag 7's "cheaper than 12 months ago" 24.8x P/E is dismissed as a planted axiom once free cash flow and the $108 billion stock-comp charge are used instead.
Full passage: premium transcript (PDF).
In short: The credit market has started pricing the vendor-financing loop: "Credit is bouncing some today on the Risk On, but worth noting that NVDA's announcement a few days ago of the $250bln latest in Great Circularity financing has not been taken well and its CDS has joined the party…" Continues his standing NVDA funding-short framing — the tell is now in default-protection pricing on the balance-sheet side, not the P&L.
Nvidia makes the chips that AI data centres run on. The worry here is not about demand for those chips — it is about who is paying for them. Nvidia has increasingly been helping to finance its own customers: putting money into, or arranging debt for, the companies that then turn around and buy Nvidia hardware. Paulo has called this loop "the Great Circularity" for over a year. The newest instalment is a $250 billion financing announcement, and his point in this note is that the market reaction to it happened in the credit market, not the stock market.
The specific signal is Nvidia's CDS — a credit default swap, essentially an insurance contract that pays out if the company can't service its debt. When the price of that insurance rises ("widens"), professional lenders are quietly saying the company's balance sheet has gotten riskier. Nvidia's CDS widening on the announcement means the people who lend money read a quarter-trillion-dollar financing commitment as an obligation, not a triumph — "has not been taken well and its CDS has joined the party."
Why that matters more than a down day in the shares: equity investors can stay enthusiastic for a long time, but the credit market decides whether the whole build-out keeps getting funded. Paulo's standing position is that Nvidia is best shorted as a funding story rather than a demand story — and this is the tell moving from the income statement (revenues versus receivables) onto the balance sheet, at exactly the moment his credit contacts are describing data-centre bond deals being pulled because the arithmetic no longer works.
In short: Engine 1 illustrated, not recommended. "Nvidia's stock surged nearly 900% between 2023 and 2025. That's despite its P/FCF multiple falling from over 100x to 58x. The reason? Free cash flow grew by almost 25x (+2,500%)." The point being taught is that a share can rise enormously while getting cheaper — no view on the stock is offered.
In short: The campaigner. Jensen Huang used his first-ever X post to enter Washington's open-weight fight, sharing an open letter — initially 25 signatories including Microsoft, Meta and Palantir — urging policymakers to preserve the development and deployment of open-weight models. NVIDIA then launched the Open Secure AI Alliance, a coalition of nearly 40 companies building defences against AI-powered cyberattacks, with Microsoft, SpaceX and IBM as founding members (Anthropic, OpenAI and Meta notably absent). NVIDIA is also one of the model providers listed in Microsoft Foundry. The commercial logic is unstated but obvious: open weights broaden who buys GPUs. A disclosed author holding. (Referenced; not a stance call.)
Nvidia makes the chips nearly all AI runs on, so anything that widens the number of companies training and running their own models is good for it. That's the subtext here: CEO Jensen Huang used his first-ever post on X to back an industry letter — signed by Microsoft, Meta, Palantir and 22 others — asking Washington not to restrict "open-weight" AI models, the kind any business can download and run on hardware it owns.
Nvidia followed up by launching the Open Secure AI Alliance, a roughly 40-company group building defences against AI-powered cyberattacks, with Microsoft, SpaceX and IBM as founding members. Notably absent: Anthropic, OpenAI and Meta. The line-up is a decent map of who benefits from open models spreading and who prefers them contained. Referenced in passing, not a stance call.
In short: On the latest circular deal — Nvidia backing the financing so OpenAI can buy Nvidia chips: "it's so incestuous… it's totally convoluted. You don't know if there's any real profitability in any of the machinations." Asked whether there is profit for Nvidia itself: "Well, there is. Yes, for Nvidia. They're not doing it for free."
Nvidia makes the chips every AI data centre is built from. The thing Eisman flags is not demand but the shape of the newest deal: Nvidia helps fund OpenAI, and OpenAI then uses money to buy Nvidia's chips. Money goes out one door and comes back in another as revenue — a circular arrangement. "It's so incestuous… it's totally convoluted. You don't know if there's any real profitability in any of the machinations."
His caveat is important and keeps this a neutral rather than a negative: pressed on whether anyone is actually making money in the loop, he agrees Nvidia is — "there probably is for Nvidia. They're not doing it for free." The doubt is about whether real, outside-funded profit exists anywhere else in the chain, and about how confidently anyone can read the accounts of a structure this tangled.
4:24And it's so incestuous. — It's convoluted. — It's totally convoluted. You don't know if there's any real profitability in any of the machinations that eventually — — there probably is for Nvidia. Like he was saying. — Well, there is. — Yes, for Nvidia. They're not doing it for free.
In short: Ives' #1: "there's one chip in the world fueling the AI revolution… I don't even think there's a debate. A third-rate Nvidia chip is a year and a half to two years ahead of Huawei" — "it's really like their world, everyone else paying rent." Plus the multiplier: "for every dollar spent on an Nvidia chip, we estimate there's $8 to $10 multiplier across the rest of tech." Luria adds Nvidia is now building its own free open-source model (Nemotron) because "open source models use just as much compute as closed source models."
Nvidia designs the chips almost every AI data centre is built around. Dan Ives' claim is that there simply isn't a contest: "a third-rate Nvidia chip is a year and a half to two years ahead of Huawei," so even Chinese buyers want Nvidia — "it's really like their world, everyone else paying rent."
His second point is about how far the money travels. For every dollar spent on an Nvidia chip, he estimates "$8 to $10 multiplier across the rest of tech" — CPUs, memory, networking and telecom gear, data-centre cooling and power. That is why he treats a strong Nvidia order book as an indicator for the whole complex rather than one company's results.
Luria adds a wrinkle that cuts the same way: Nvidia is now funding a free open-source model (Nemotron), because open models consume just as many chips as closed ones and Nvidia would rather the world's free model be American than Chinese. Giving software away to protect hardware demand.
36:45It just speaks to the reality of you could talk to anyone in the supply chain, the reality is that any big Chinese tech company would wants an Nvidia chip over Huawei. And I think as that plays out globally, especially when it comes to physical AI and how Jensen's building it, it's really like their world, everyone else paying rent when it comes to the chip perspective.
In short: The call's opening news: Nvidia is in talks to guarantee ~$250B of financing for an OpenAI data center — a 10-gigawatt SoftBank project in Southern Ohio — "one of the biggest guarantees I've ever seen… very shocking because they were creating their own open-source AI, but I guess they need someone to buy their GPUs." Nvidia sits on the receiving side of the capex transfer, which is why semis are the buy-on-dips leg while hyperscalers wait: "when the market realizes all the Max7 guys are raising capex spending, you'll see another bid for the semiconductor names." Jensen Huang's open-weights essay argues open models raise demand.
The news that opened the call: Nvidia is in talks to guarantee about $250 billion of financing for an OpenAI data center — a 10-gigawatt site SoftBank is developing in southern Ohio. A "guarantee" means Nvidia would stand behind the loans, so lenders will fund a project whose whole purpose is to buy Nvidia chips. Singh's dry read: "very shocking, because they were creating their own open-source AI — but I guess they need someone to buy their GPUs."
The bigger point is where Nvidia sits in the money flow. The cloud giants are burning their cash on AI; the chipmakers are receiving that cash. That's why his call is buy semiconductors on dips while waiting on the cloud companies — and why he expects that once the market digests everyone raising capex this week, "you'll see another bid for the semiconductor names."
Full passage: premium transcript (PDF).
In short: "It's obvious the AI winners are Nvidia, the picks-and-shovels company. We all know that ship sailed a few years ago" — the premise for asking his team for the non-obvious idea (Zeta). Also the customer whose single-supplier call on high-bandwidth memory made SK Hynix an "awesome business."
54:33So, it's obvious the AI winners are Nvidia, the Pix and Shovel company. So, we all know that that ship sailed a few years ago. So I always try and encourage my team, tell me something I don't know. And so, based on my experience, if you can find companies with proprietary databases, right, that they themselves can use that proprietary database with AI technology to grow their business or help other companies grow.
In short: The valuation ceiling of the complex: "not one company in the hardware space should trade at higher valuations than Nvidia." Gross margins "seem solid the next couple years," then "anyone's guess."
Nvidia makes the chips at the heart of the AI boom. Chanos isn't attacking Nvidia itself — he uses it as the ruler for everyone else. His rule: nothing that only exists because it buys Nvidia chips should be worth more (per dollar of profit) than Nvidia. Many data-center and "neo-cloud" stocks break that rule, which is his signal they're overvalued.
On Nvidia's own profits: the fat ~75% margins look safe for a year or two, but "tech cycles are short" and after that "it's anyone's guess" as real competition arrives. So he's cautious, not short.
20:21I would turn the argument around from the other side of the telescope and say, okay, there are a handful of companies that are gatekeepers and price makers in this boom. Nvidia's one of them. Not one company in the hardware space should trade at higher valuations than Nvidia. That's one of the things we hold dear.
In short: Terranova: down today, but after a strong afternoon yesterday "I think NVIDIA is going to follow" and "participate as well in the coming weeks." Baruch "likes it a lot here" — consolidating, and "the free cash flow there is becoming amazing relative to the other Mag Sevens"; jokes that if it weren't a $5T company he'd think a take-private was possible.
Nvidia makes the chips that train and run AI. It was down on the day, but Joe Terranova expects it to "follow" the mega-cap breakout and participate in the coming weeks after a strong afternoon the day before. Bill Baruch likes it "a lot here" while it consolidates, and highlights that its free cash flow (the cash left after running the business) is "becoming amazing relative to the other Mag Sevens" — i.e., it's throwing off more spare cash than its trillion-dollar peers.
In short: Poster child of the earnings bubble: its numbers soar as hyperscalers place GPU orders recognized as revenue immediately, while the buyers' costs (depreciated 5-6 yrs) lag — an inflated-earnings gap that collapses when spending slows. Efficiency breakthroughs (a model on an iPhone) threaten chip demand.
Nvidia is the clearest example of his "earnings bubble." When a hyperscaler orders GPUs, Nvidia books that as revenue and profit almost immediately, so its numbers look spectacular. But the customer only spreads the cost over 5-6 years (depreciation), so the buyer's pain is hidden while Nvidia's boom is front-loaded. When the spending slows — or efficiency breakthroughs (a model running on a phone) mean fewer chips are needed — that inflated-earnings gap snaps shut. In a 2000-style bust these supplier stocks fall hardest.
6:20And so what's happening is all these orders are going in and Nvidia's numbers are soaring and Micron's numbers are soaring and Western Digital's numbers are soaring, and so as their earnings soar you're seeing Micron stock go up hundreds of percentage, up 700% this year. But the hyperscaler costs don't show up right away.
In short: Link's first-ever NVIDIA buy (last week's noteworthy move): 18× forward, revenue growing ~80% with mid-70s gross margins, FCF set to double, 97% GPU share and proprietary software hard to replicate; "the cheapest it's been since 2019," out of favor, finding its footing. Terranova: a "return to accumulation" after sitting sideways for a multi-month period; benefiting from the rotation out of memory. She'll keep adding on ugly days.
Nvidia makes the chips that train and run AI. The news here is that Stephanie Link bought it for the first time ever, because the price finally got cheap enough: about 18 times next year's expected earnings, while the company is still growing sales roughly 80% a year at fat (mid-70s%) profit margins, with free cash flow set to double. She notes it controls about 97% of the market for AI chips and has software that's very hard for rivals to copy.
Her clincher: on valuation it's the cheapest it's been since 2019, and it's badly out of favor even though it has quietly stopped falling and "found its footing." Joe Terranova frames the same thing in trading terms — after going sideways for months, buyers are stepping back in (a "return to accumulation"). Link plans to keep buying it on ugly down days.
In short: Giroux: "Nvidia's monopoly in chips is ending" — Anthropic built the world's best LLM on low-cost Google/Amazon ASICs; the shift to inference makes ASICs/CPUs competitive; neocloud GPU-rental demand fades as hyperscalers in-source; GPU rents accrue to hyperscalers instead. Ellenbogen: already significantly underperforming other AI-buildout stocks as LLMs lessened their Nvidia ties. (Black's counterpoint: 22x this year's earnings — reasonably priced.)
The panel's most consequential negative. Giroux's argument: Nvidia's chip monopoly is ending — Anthropic built the world's best AI model on cheap custom chips from Google and Amazon, and as AI shifts from training models to running them (inference), those cheaper chips compete head-on with Nvidia's GPUs. The "neocloud" companies whose whole business was renting out Nvidia GPUs lose their reason to exist as big customers bring the work in-house. Ellenbogen notes the stock has already lagged other AI names for a year as the model builders diversified. Black's lone counterpoint: at 22x this year's earnings it isn't expensive — the debate is about what happens to the "22x" part.
In short: Baruch "loves what we're seeing — NVIDIA waking up right here," flat/digesting on the 200-day, putting new client cash to work (buying the target NVIDIA + Broadcom, "buys right here and it's going to run"). Simpson's final trade: Blackwell keeps outpacing, demand "unbelievable," finally approaching $100B/quarter.
Nvidia makes the chips that train and run AI. After going nowhere for a month (flat/slightly up on the year), Bill Baruch likes that it's "waking up" while holding a key trend line, and is putting new client money to work in it (alongside Broadcom). Kevin Simpson made it his final-trade pick, noting demand for the new Blackwell chips is "unbelievable" and the company is finally approaching $100 billion of revenue per quarter.
In short: The bull leg still stands — Nvidia "reported 70% revenue growth. So clearly, the AI capex story was not slowing." The mega-caps were all "racing to buy Nvidia chips" to build out AI.
Nvidia sells the chips every AI data center is built around. In this episode it's the piece of the bull case that still holds up: it reported 70% revenue growth, which tells Eisman the AI build-out spending "was not slowing." All the mega-caps were "racing to buy Nvidia chips."
That matters because his whole bear case is about the buyers of those chips (the hyperscalers) turning capital-intensive and moat-less — not about demand for the chips themselves drying up. As long as Nvidia's growth stays high, "this story is not over."
10:33Today, Oracle is even lower at 140. But that's just Oracle. The next crack in the argument showed up when the tech companies reported fourth quarter numbers. Nvidia, of course, reported 70% revenue growth. So clearly, the AI capex story was not slowing. What got people to start to freak out was the capex budgets for 2026.
In short: Link initiates ("never owned it, regretfully"): best-in-breed on sale — 18× PE, 15× EBITDA, cheapest since 2019, 97% server-GPU share, earnings +130%/rev +85% last quarter, FCF doubling to $206B; "in the penalty box" but fundamentals only better. Brown long-bullish — no longer a momentum stock (the crowd left for Micron), better than one, and you don't pay a 40–50× multiple. Bryn owns/loves it, sells calls, questions whether a $5T cap goes to $10T. Jason's final trade (87% earnings growth).
Nvidia makes the chips that train and run AI. The news here is that Stephanie Link — a value-minded investor who says she "never owned it, regretfully" — finally bought it, purely on price. Even though Nvidia's business is booming (97% share of AI server chips, earnings up 130% and revenue up 85% last quarter, free cash flow on track to double to $206 billion), the stock has gone nowhere and is actually down on the year, so it now trades at just 18× earnings — the cheapest since 2019, and far below the 40–50× you'd pay for hotter chip names. Her rule is "buy low, sell high," and this is a rare chance to buy the best company in the industry on sale.
Josh Brown adds that Nvidia isn't a "momentum stock" anymore — the fast-money crowd left for Micron — which he thinks makes it safer, not weaker: lower hype, solid outlook. Bryn Talkington owns it and loves it, but raises the ceiling question: it's already worth almost $5 trillion, and she's not sure the market will let any company double to $10 trillion soon — so she keeps "selling calls" against it (collecting income by agreeing to sell at a higher price) rather than just holding and hoping.
In short: Reference only — the counterparty to Valar's atomic-power milestone: Valar "signed a contract with Nvidia to provide 30 megawatts (MW) of atomic power to one of NVDA's data centers." Cited to illustrate nuclear entering the data-center power stack, not an argued call on NVDA here.
In short: Named as a reference customer for ServiceNow's new ITSM Level-1 automation use case (with the <100-day ROI guarantee) — a proof point that GS says addresses system-integrator uncertainty over which AI use cases to lead with.
Full passage: premium transcript (PDF).
In short: The anchor customer — buys roughly two-thirds of its HBM4 from SK Hynix for the Vera Rubin platform, so SK Hynix "sits inside the most important hardware roadmap in tech." That's also a concentration risk: NVIDIA is actively building second sources (pulling in Samsung and Micron) and holds pricing leverage over its lead supplier. A disclosed author holding. (Recap, not a stance call.)
In short: Two-sided: the memory-cycle beneficiary whose Rubin generation keeps lifting HBM demand (and whose chips are ~42% of a data-center's cost) — but sold off with the group when Anthropic said it's talking to Samsung for a custom ASIC, and Burry disclosed a new short, calling the semi index "a pure form of overvaluation rarely seen."
Full passage: premium transcript (PDF).
In short: The GPU benchmark Cerebras is positioned against: its flagship Blackwell packs ~200B transistors vs Cerebras's ~4T-transistor wafer-scale chip, and a GPU cluster shuffles data between thousands of chips where Cerebras keeps a model on one. Cerebras's pitch is "genuine non-Nvidia demand" — the rarest thing in AI hardware. Feldman also eyes selling Cerebras's fast "decode" to companies that already own NVIDIA GPUs. Referenced/neutral.
NVIDIA is the company whose GPU chips power almost all AI today, and it's the benchmark Cerebras is measured against. NVIDIA's top chip (Blackwell) has about 200 billion transistors; Cerebras's single wafer-scale chip has about 4 trillion — roughly twenty times more — because it's an entire wafer rather than a chip cut out of one. The pitch is that for the "answering" part of AI (generating text fast), one big chip beats a crowd of smaller ones wired together.
Why this matters: nearly every AI hardware story is really a story about buying more NVIDIA. Cerebras is rare because it has genuine demand for a non-NVIDIA alternative. Interestingly, Cerebras even sees a way to sell its speed advantage to companies that already own NVIDIA GPUs — handling just the slow part of the job. A recap reference, not a call on either stock.
In short: The capex beneficiary (not a spender), down ~18% in the Mag7 tally; raised ~$25B of debt >3× oversubscribed (~$85B demand) and its Vera Rubin platform needs >5× the memory, pulling in Micron's HBM4.
Full passage: premium transcript (PDF).
In short: Referenced — a key counterparty deciding HBM allocation; Jensen Huang confirmed NVIDIA will source HBM4 for its next-gen Vera Rubin platform from all three makers. "Margins even NVIDIA would envy."
In short: Simpson added more at ~$194 (under $200) and final-trades it: no longer the doubles-every-five-minutes Nvidia (Micron took that lead), but it gets cheaper on valuation as it generates revenue/FCF; a dividend bump to $0.25/qtr + an ~$85B buyback make it "investable, not just a trade." Blackwell ramping faster than any product in company history; demand outpaces supply. (Worst week since April '25.)
Nvidia makes the chips that train and run AI. Simpson bought more under $200 and argues the stock has actually gotten cheaper in a meaningful sense: the company keeps growing its sales and its free cash flow (the cash left after running and investing in the business), so each dollar of stock now buys more earnings than before. It also just raised its dividend (to 25 cents a quarter) and announced an $85 billion buyback — both signs management thinks the shares are good value and a way to return cash to owners.
His framing — "investable, not just a trade" — means he now sees it as a long-term hold rather than a quick momentum bet. The fundamental support: "Blackwell," its newest chip line, is ramping up (selling into customers) faster than any product in the company's history, and demand still outstrips what they can supply.
In short: Named (with AAPL/SpaceX) as a crowded, margin-funded mega-cap. An AI-sector crash could force those holders to sell gold for margin liquidity — a temporary metals down-leg — but "in the end, no… it always reverts back much higher."
NVIDIA is the poster-child of the AI stock boom. Schectman isn't analyzing its business — he's using it (with Apple and SpaceX) as an example of a stock so widely owned, often on borrowed money ("margin"), that a sharp drop forces those investors to sell other things to cover their loans. One of the things they might be forced to sell is gold, which is why an AI-sector crash could briefly knock the gold price down even though gold had nothing to do with the crash. His key caveat: that kind of forced selling is temporary — like in March 2020, gold dipped then "came right back up" — so he'd treat any such dip as noise, not a change in the long-term case.
53:53for precious metals beyond what they've already fallen so far? Well, I think that's what we're supposed to believe, that they throw out the baby with the bathwater. To me, it's always been MOPE, management of perception economics, as Jim Sinclair used to say. In reality, how many of the people that are on margin with NVIDIA and Apple and SpaceX have a whole big position in gold that they have to sell in order to come up with margin liquidity? When the market collapsed in 2020, yes, gold sold off, but it came right back up.
In short: His one-decision framing: technology is now ~40% of the S&P, energy just ~3–3.5%. "Would you rather own Nvidia at its current price, or all of the energy stocks in the S&P, for the next 10 years? I know which one I'd rather own" — i.e. avoid the most-crowded mega-cap, own the under-owned molecule layer.
He doesn't argue Nvidia is a bad company — he argues it's the most crowded trade on Earth, and crowding is the risk. His framing is a single thought experiment: technology is now about 40% of the whole S&P 500 while energy is barely 3%. So if you could make one buy-and-hold decision for the next decade — all of Nvidia at today's price, or every energy stock in the S&P — he'd take the entire under-owned energy sector over the single most-loved megacap. It's a relative-value/crowding call, not a company call.
22:01So that's how I'm focusing on this. And I think you look for an example — I'll stop after this. I think technology is now almost 40% of the S&P; energy is still 3% of the S&P, 3 and a half. So I ask people, would you rather own all of the — would you rather own Nvidia for the next — you can make one investment decision for the next 10 years.
In short: Explicitly no view — asked what he thought of Nvidia's valuation, he said "I can barely pronounce it," so opining on the price-to-value relationship "is a non-starter." Money is made on the delta between price and value, and without a view on value the price is useless; outside his circle of competence.
29:26Somebody was asking me the other day, what I thought of Nvidia's valuation. And I said, I can barely pronounce it. So, the idea that I could opine on the relationship between price and value is a non-starter. I can't say that SpaceX is undervalued or overvalued because I don't understand the net present value of settling Mars.
In short: "The forward earnings on NVIDIA are complete baloney" — concentrated in a handful of counterparties. Doesn't own it; being long is "the dumbest trade" — at ~$4.5T it must hit $8T to double, while uranium gives far better risk/reward. (Peter Lynch: run from the hottest stock in the hottest sector with 50 buys.)
He doesn't own Nvidia and thinks being long it is "the dumbest trade." His core objection: its "forward earnings are baloney" because they depend on a handful of big customers, not the millions of small ones Microsoft or Google have. At ~$4.5 trillion it would have to reach $8 trillion just to double — poor odds. Channeling Peter Lynch ("run from the hottest stock in the hottest sector"), he says you can't make money on a stock with 50 analyst buys; uranium offers far better risk/reward.
10:27You can't say that about NVIDIA, 23 times earnings? Oh, the forward earnings on NVIDIA are complete baloney. Tell me why. Because they're concentrated with a handful of companies. Think of the forward earnings of Microsoft. Those earnings, or Google, they have like millions of counterparties that give them that revenue.
In short: Explicitly no view — "I can price natural resource companies… conventional financial services businesses. But I don't know how to price technology companies… and to be honest with you, I don't care." Outside his circle of competence (cites Cundill/Buffett's "no FOMO").
6:36— Yep. — As do hyperbolic declines. — Yep. — And that's the only kind of trading I can do. You have to give me something very very obvious for me to make money as a trader. What I've learned to do is stick to what I do. So when people ask me to comment on whether, say, Nvidia is overpriced, I have to say I don't know.
In short: The bull case's bellwether: 1Q26 revenue +85%, accelerating from ~65% a few quarters ago. "As long as Nvidia's revenue growth remains elevated, this story is not over." No sign AI capex is weakening.
Nvidia makes the chips that power AI, so Eisman treats it as the single best gauge of whether the AI boom is still going. The bull case rests on one fact: its revenue grew 85% last quarter — and that's accelerating from ~65% a few quarters earlier. As long as that growth stays high, "this story is not over," because it means companies are still pouring money into AI hardware.
It's the one clearly positive leg of his otherwise increasingly skeptical AI snapshot — the evidence that the spending hasn't cracked yet.
6:39Here is my snapshot where we stand on the bull-bear curve. On the positive side, there is no sign that AI capex is weakening. Quite the opposite actually. Every major company that reports still ups its AI capex and Nvidia is the bellwether of the group. Its 1Q26 revenue growth was an incredible 85% which is an acceleration from the 65% growth a few quarters ago.
In short: Passing reference — the custom Gemini powering Apple's Siri runs on NVIDIA chips inside Google's data centers.
In short: "Look at Nvidia today, a 20 billion-dollar bond offering. The Mag 7 has done almost 300 billion of bond offerings." His evidence that the old cash cows are now capital-intensive businesses funding AI with debt and equity.
He doesn't argue the business here — he uses that day's $20 billion Nvidia bond deal as a symptom. The Mag 7 has now issued roughly $300 billion of bonds to fund AI. Companies that were famous for generating cash without needing outside money are now borrowing at scale, which is exactly the transition from "cash cow" to "capital-intensive business" that changes how you should value them.
14:38Look at Nvidia today, a 20 billion-dollar bond offering. The Mag 7 has done almost 300 billion of bond offerings. So, they're selling stock. Google just sold 80 billion of stock. They're selling bonds. They're doing everything to fund artificial intelligence. And so, that creates a dynamic where these companies are the most expensive ever.
In short: The ultimate beneficiary of the circular machine: all $35B of "Big Sky" proceeds buy Nvidia chips ("Jensen loves this shit"). Round-tripping inflates demand; over-ordering of GPUs/TPUs risks a multi-year glut and write-down wave.
Full passage: premium transcript (PDF).
In short: Hub of the round-tripping: invests billions in OpenAI/neoclouds who then buy Nvidia GPUs; Jensen "consistently goes and pumps the market" (the $2B Marvell call). In the cascade unwind, GPU names sell off after opto/memory — "this whole supply chain is just one trade."
Singh's worry isn't Nvidia's chips — it's the financing loop around them. Nvidia invests billions in customers (OpenAI, cloud startups), who then use that money to buy Nvidia's GPUs. That "round-tripping" inflates apparent demand. He also points at the promotion machine: Jensen Huang put $2B into Marvell, then went on TV calling it a future trillion-dollar company — the stock jumped 25%.
His scenario if it cracks: a cascade where memory and optical-component makers sell off first, then the GPU names like Nvidia, with power providers holding up last. The trigger would be a hyperscaler (Meta, Microsoft) announcing capex cuts. With semis ~20% of the index, he expects 10–15% market-level "hiccups" within two years — not a prediction of immediate collapse, but a warning that the whole supply chain has become one crowded trade.
40:45for your PC. It's incredibly expensive. look at Sandis for example as well. So what's happening is they've accelerated the spend., for example, Nvidia is investing billions into companies like OpenAI or other Neocloud providers, and then they're using that money to basically buy Nvidia GPUs.
In short: Named in the same largest-constituent list as the epicenter of the forced selling; also implicated in "the chips are probably going to have to" fall in price as the technology proliferates.
0:19You have ample opportunity to evaluate businesses. So, what one of the things we've been thinking a lot about is Newton's third law uh uh physics, which is for every action there's an equal and an opposite reaction. So, yes, SpaceX is going public, but where is the capital going to come from? And your point on the passive indexes, they're going to have to sell the largest constituents, which make up uh you know, are made up by names like Google and Meta and Microsoft and Amazon and Nvidia.
In short: Referenced as the archetypal "bottleneck" AI stock momentum is chasing — the kind of name pulling the narrow rally (not a Carlson pick here).
15:20Well, I do and I still rate it as a buy. One thing you'll notice with all of these companies that have moved down over the past year is they're not in the new exciting thing. The new shiny exciting thing is artificial intelligence and especially the bottleneck stocks. The ones like ASML, Nvidia, TSM, any company that has a bottleneck. You have memory stocks.
In short: Mentioned only in passing — speculation about whether NVIDIA's CEO would attend the Trump–Xi summit, cited as a read on US–China diplomacy. No stock view.
27:53you know, they there was speculation whether Nvidia CEO would come over there and all this fanfare and the like and all at the time of the same time that there's this back channeling with Iran to try to figure out some sort of deal or ceasefire and you know people are putting two and two together that perhaps Xiinping will be the mediator there and China to try to get some sort of resolution to occur here because you know if there's a US blockade I assume there's no oil going to China from Iran.
In short: Bought in 2023 at ~4× earnings — the biggest S-curve; the chip "renaissance"/decommoditization.
Nvidia makes the chips that train and run AI. Sacerdote bought it in 2023 at about four times earnings — dirt cheap — because he saw the biggest "S-curve" ever beginning. An S-curve is the typical path a new technology takes: a long flat start, then an explosive vertical takeoff once the barriers fall, then a leveling off; catching the takeoff is where the money is.
He frames Nvidia as the centerpiece of a chip "renaissance," where AI is forcing real innovation back into hardware that had been a sleepy, commoditized business for decades. Every year people called it a bubble; every year it kept compounding because the demand was real.
20:57When we bought Tesla in 2019 for the car scurb, we were paying five times earnings. When we were owning Apple, we were paying four times earnings. When we bought Amazon for AWS, we we were getting it for free. And you know, the world doesn't think exponentially. and they're so focused on the next year, the next quarter.
In short: "I don't know how to value a stock like Nvidia" — understands the growth thesis but won't opine; part of the <20-stock concentration that worries him about S&P breadth.
"I don't know how to value a stock like Nvidia." Rule understands the AI growth story but won't put a number on it. His real worry is concentration: fewer than 20 stocks (some say fewer than 10) are driving the whole S&P 500 — and if he can't value them, he doubts most of their owners can either, which makes the index fragile.
0:55I'm conflicted, mostly because that's not what I do. Money is made on the intersection of price and value and I don't know how to value a stock like Nvidia or Intel. I understand the growth thesis. I'm personally not affected by fear of missing out. So I don't have a way to comment on the valuations. Similarly I don't express an opinion on the S&P 500 except that I'm concerned about the breadth — so much of the valuation concerns less than 20 stocks, some would say less than 10.
In short: Partner in the Munich robotaxi test (with Uber and Autobrains) that went live this week — one data point on the widening AV footprint.
In short: Growth is accelerating (85% vs 65% two quarters ago, ~75% GM, $91B guide vs the famous $11B "big bang" of May 2023) yet the stock lags because investors play the constraint names — a divergence that "has to normalize" with Nvidia coming up. Plus a ~$20B CPU story this year (Grace, standalone CPU racks) as agentic AI explodes CPU demand; supply secured across the value chain.
Nvidia makes the GPUs (graphics processors) that train and run AI — the picks and shovels of the entire boom. Rasgon's headline: growth is speeding up, not slowing down. Revenue grew 65% two quarters ago and 85% last quarter, at roughly 75% gross margins, and the company just told investors to expect $91 billion next quarter — nearly ten times the $11 billion guide that shocked Wall Street in May 2023.
So why has the stock lagged the rest of the chip sector? Three reasons: it already had a huge run; investors prefer to chase whatever part of the supply chain is scarcest (memory, equipment, optics — where earnings estimates have jumped tenfold); and at over $5 trillion — about 8% of the S&P 500 — many funds are simply not allowed to own more of it. Rasgon sees that gap as the opportunity: the bottleneck stocks can't be right while Nvidia is wrong, so "either the constraints come down or Nvidia comes up." Bonus: Nvidia now has a CPU business (~$20 billion this year, about the size of Intel's or AMD's) riding the agentic-AI wave, and it locked up its own supply chain years in advance.
13:17Either the constraints are going to go down or Nvidia I think has to come up. The valuations I think have to normalize — but there's no constraint on GPUs right now. — Well there there is and there isn't. So there there's leading edge logic and and what's called coas the which is the packaging technology and we can talk about that if you want to put the chips together.
In short: A caution to AI bulls: Nvidia's product needs energy and copper — the building-block materials of AI are "grossly underinvested" vs the tech stocks expected to benefit.
Rule doesn't rate Nvidia itself — he uses it to make a point to AI enthusiasts. Nvidia's chips run in data centers that need electricity, and electricity needs copper (to generate it and to wire it everywhere). His argument: the raw materials that make AI physically possible are badly under-invested compared with the popular tech stocks, so even an AI believer should own copper.
28:56If you're the kind of person who as an example is buying Nvidia as a consequence of the consumption of their product in AI, understand their pro their product requires energy and it requires copper. Uh the materials that are the building blocks of AI are grossly underinvested in relative to the technology stocks that are expected to be the direct beneficiaries.
In short: Referenced only — CrowdStrike poached its Bartley Richardson to be Chief AI and Autonomous Systems Officer (the talent-flow detail behind CRWD's new AI-leadership build-out).
In short: The one Mag-7 he'll defend — "not a high multiple stock," but "sales better keep coming through."
Nvidia is the dominant AI-chip maker. In a conversation where Hay is bearish on most of the megacap "bubble," Nvidia is the one he'll defend: he says it's "not a high multiple stock," meaning its price isn't crazy relative to its earnings — unlike the truly speculative names.
The catch is the condition he attaches: "sales better keep coming through." The valuation only holds if Nvidia's revenue keeps growing fast; if AI spending slows, the story changes. So he's constructive but conditional — neutral, not a ringing buy.
32:39It's just sales better keep coming through. So that that's a different issue. Uh so I I think that where we're really seeing this speculation, I mean the the whole idea of these space stocks, I mean it's going to be probably decades before there's really meaningful revenue and profits coming from those or at least close to a decade.
In short: Emblem of an unsustainable AI/semiconductor blowoff — at a ~$5.4T market cap it's bigger than the entire S&P 500 health-care sector; "these relationships don't last forever."
Nvidia makes the chips behind the AI boom and has become the poster child for it. Dowd's eye-popping stat: at about $5.4 trillion, Nvidia alone is worth more than every health-care company in the S&P 500 combined.
He's not analyzing the business — he's saying that kind of lopsided dominance is a classic late-stage bubble signal: "these relationships don't last forever."
19:57And that all the liquidity that's left is focused on AI and semiconductors. And I want to point out something that is pretty pretty alarming. Semiconductors is a percent of the S&P 500 is 17%. That's one industry. Um and uh and and and Nvidia, the stock Nvidia is market cap 5.4 trillion is bigger than the whole healthc care sector of the S&P 500.
In short: The yardstick for stretched US AI value ("a thousand Alibabas fit into Nvidia") and one of history's great IPOs that were each only a fraction of GDP.
31:10But but worse than that, this to me is almost like a Pearl Harbor Pearl Harbor situation where the these these engineers are taking code, they're dumping into the Kimmy facilities and they're running it through open source and all of the capex that we're spending on AI al some of it being ripped off and in open source is like feeding it into a pool of piranhas and that can get you really this whole malinvestment thesis that's out there about in any in any kind of boom there's malinvestment right there was malin investment in the shale
In short: Referenced — Nutanix's IdentityAI launched with NVIDIA GPU support (AMD coming). A peer-set mention, not a stance.
In short: Disclosure-only — one of the author's App Economy Portfolio holdings (the SpaceX AI buildout's compute supplier by implication, but not directly discussed in the body).
In short: Constructive but explicitly not buying — the archetypal scarcity seller (rev +80% YoY) "probably will do well" and is "becoming less cyclical over time" via CUDA/ecosystem, but it's "Not Nvidia" among his biggest bets.
8:18ASML still have a little bit of cyclicality. Then, of course, we have TSMC. TSMC is a very durable seller. It's not some highly cyclical stock. We have Nvidia which I believe because of CUDA, because of the infrastructure, because of the ecosystem, it has some cyclicality, but I truly believe that Nvidia is becoming less cyclical over time.
In short: The subject — a blow-out Q1 FY27: revenue +85% Y/Y to $81.6B, Data Center +92% to $75.2B (92% of the business), 75% gross margin held through the Blackwell ramp, FCF +86% to $48.6B, $91B guide (+95% Y/Y). "Demand has gone parabolic"; compute is the raw material for AI revenue. ~27x forward, EPS +140% Y/Y, supply-constrained.
NVIDIA makes the chips ("GPUs") that train and run almost all modern AI. This quarter it sold $81.6 billion worth in three months — up 85% from a year earlier — and almost all of it (92%) was data-center gear. The headline idea is the "token economy": a token is the unit of text/output an AI model produces, and Jensen Huang's pitch is that companies now pay real money to generate those tokens, so the chips that produce them are no longer just a cost — they are the raw material a whole new industry sells. Compute is the new oil.
Why the 75% gross margin is a "lie detector": gross margin is the slice of each sale left after the direct cost of making the product. If rivals were undercutting NVIDIA on price, or if customers were losing their appetite, NVIDIA would have to cut prices and that margin would slip — usually before anything else shows up in the numbers. It didn't slip (held at 75%), which the author reads as demand and pricing power still intact.
"China as a call option": a call option is the right to buy something later at a set price — it costs little and only pays off if things go your way. NVIDIA is currently assuming zero revenue from China (the US cleared H200 chip sales to ~10 Chinese firms, but Beijing hasn't let them ship). So China is upside that isn't in the forecast at all — if it reopens, it's pure bonus on top of an already-booming business.
Vera and "GPUs think, CPUs orchestrate": NVIDIA is also pushing into CPUs (the general-purpose chips that coordinate work) with a new line called Vera. Jensen's mental model: in a world of billions of AI agents, the GPUs do the heavy "thinking" (running the models) while CPUs do the "orchestration" (scheduling, coordinating, moving data). Management says Vera opens a $200 billion addressable market and ~$20 billion of CPU revenue this year — a second growth engine alongside the GPUs.
In short: He was very early (talking to Jensen back in 2000). Competition is coming, but Nvidia "is still going to be the dominant part of the market"; one of only two working switched scale-up networks.
Nvidia makes the GPUs at the center of the AI boom. Baker was extremely early (he was talking to CEO Jensen Huang about chips back in 2000). He acknowledges rivals are coming, but expects Nvidia to stay "the dominant part of the market."
A key technical edge: to run today's biggest AI models efficiently you need a special kind of chip-to-chip plumbing he calls a "switched scale-up network," and Nvidia operates one of only two that actually work in the world (Amazon's Trainium has the other).
16:32I appreciate that. You talked about silicon and what's happening on the chip side. Obviously, you were very early in Nvidia. You were talking about talking to Jensen in 2000, let alone 2023 when the ChatGPT moment happened. Alternatives are coming. Competition is coming.
In short: The original revenue-circularity node (his standing funding short): sells GPUs to neoclouds/labs "funded either by NVDA's equity stakes to pay for the chips, or… massive funding rounds anchored by hyperscalers." Cited as the base layer the circularity has now grown "beyond" — into memory and, worse, into the earnings line.
Nvidia sits at the base of the loop: it sells the AI chips, but many of its customers can only afford them because Nvidia itself, or the big cloud companies, funded them in the first place — so Nvidia is partly buying its own sales. That's the "revenue circularity" everyone now half-knows about. Paulo's new warning is that the game has grown beyond Nvidia into memory chips and, more dangerously, into the earnings statements of the cloud giants. Nvidia is his standing "funding short" — a bet against the company because the cash flowing through this circle is ultimately recycled venture money, wonderful until someone asks for their money back.
In short: The copper-demand swing factor — old Hopper data centers used little copper, but Blackwell/Vera Rubin need ~50–80k tons/GW; ~15 GW/yr of builds ≈ 750k tons vs ~1M tons total demand. Also the China rare-earth bargaining chip.
Nvidia is the AI-chip giant, mentioned as the swing factor for copper demand. Its older "Hopper" data-center systems barely used copper, but the newer "Blackwell" and "Vera Rubin" systems need enormous amounts — roughly 50,000-80,000 tons of copper per gigawatt of computing. At ~15 gigawatts of new builds a year, that's about 750,000 tons against total world demand of roughly a million tons — a coming demand shock.
He also notes Nvidia chips were a bargaining chip with China: when China cut off rare-earth exports, the US quickly agreed to ship Nvidia chips and the rare earths resumed.
42:52Um what was the catalyst for it eventually to catch up in prices? We haven't seen anything yet in terms of the AI demand, really. I mean, the old data center architecture uh with the Nvidia Hoppers didn't actually require that much copper. You know, the grid upgrades are really starting now. But, as we move to the Blackwell architecture, uh as we move to the Vera Rubins, you know, these data centers are going to be consuming inordinate amounts of power as they become more and more, you know, they get the latest generation of chips, which just simply means you
In short: Framed as both partner and rival, and the framing is the thesis: "Although Arista and Nvidia collaborate closely, they're also becoming competitors. Arista connects Nvidia GPUs via Ethernet, while Nvidia promotes its own expensive InfiniBand. As AI grows, Arista's affordable solution becomes more attractive." No view on Nvidia itself.
In short: The reference point for the chip layer he owns — "dramatic undersupply of chips," every Nvidia rack pushing power/PCB/networking suppliers; the few component makers "innovating hand in glove with Google and Nvidia." (TTMI "just won Nvidia.")
Nvidia makes the chips that train and run AI, and here it's the anchor of Sacerdote's "chip layer" — the part of the stack he owns alongside the AI models. His point is that chips are in "dramatic undersupply," and each Nvidia server rack draws so much power and runs so hot that it forces every surrounding supplier (boards, networking, power) to innovate.
The few component makers that can keep up work "hand in glove with Google and Nvidia," which is exactly why he likes picks like TTMI — which "just won Nvidia" as a customer.
15:236 terabit, the next year's 3.2 terabit. So the people selling into that — there's only a few of them who can do that, that are innovating hand in glove with Google and Nvidia. So there's less competition, there's higher margin, there's higher ASPs every year, and you've got tremendous visibility. So all these companies that nobody used to ever pay attention to are now golden, wonderful businesses, and that earnings algorithm is units growing 50%.
In short: 2022 precedent: −60% "on the eve of the AI revolution" (no buybacks then); cited among the overbought "idiotic AI stocks" set to deflate.
Nvidia is the leading AI-chip maker, also used here as a 2022 precedent: it fell about 60% that year — "on the eve of the AI revolution" — back when it wasn't buying back its own stock to support the price.
He lumps it with the overbought "idiotic AI stocks" he expects to deflate. The lesson he's drawing: even the eventual biggest winner can suffer a brutal drop, so the current crowding into chips is dangerous.
15:22He invested $400 billion in the metaverse. The stock dropped 70% in 2022. Same thing with Jensen. Right on the eve of the AI revolution, they didn't buy back any stock. The stock was down 60% during that time. Yeah, but gold gold miners, I mean, gold, silver, copper, oil and gas had the best year in 2022.
In short: Cited as evidence of the AI bid, not as a pick: +20% in April 2026 on "stronger demand for AI chips and data center expansion." Sits inside the concentration warning — IT and Software at 35% of the index, the top ten at 38.5% — and inside the "too hard pile" verdict on picking AI winners.
In short: Down ~19%; a top-2 passive holding unwinding as the AI-capex margin story deteriorates.
Nvidia makes the chips at the heart of the AI boom and was, with Microsoft, one of the two biggest stocks in the S&P 500.
It's down about 19% as the AI-spending story sours and money rotates out of these crowded top-two index holdings. When the most over-owned names start to unwind, he expects them to keep underperforming as cash moves toward hard assets.
6:41The Mag 7's down 15%. Meta is down I think more than 20%, right? Nvidia's down close to 20, 19. And because the street made all these investments, they're expecting these 820 data centers to be built in 5 years because of all these different problems with energy costs, with memory. Think of the memory costs.
In short: The other half of the top-2 S&P concentration; "everyone's in the chips — the dumbest trade on the map."
Nvidia makes the chips that power AI, and it's the other half of that top-two concentration in the S&P 500.
"Everyone's in the chips — the dumbest trade on the map." His point is that the trade is dangerously crowded: when everyone already owns it, there's little new money left to push it higher, and a lot that could rush out. He'd rather own the energy infrastructure that powers AI than the chips themselves.
25:30Well, you're really setting yourself up for failure because the S&P is nothing like it used to be. 10-15 years ago, the top two or three stocks in the S&P were maybe 7% of the index. In recent years, the top two stocks got up to 14-15% of the index, and they're both AI stocks — Microsoft and Nvidia.
In short: The yardstick, not a pick: at a 39.9x forward P/E it is cheaper than both Walmart and Costco. Used only to show how far the flight into "AI-proof" businesses has gone. No view on the company itself.
In short: #10 on the Q4 table and #6 over two quarters, and named in the "a lot of large cap stocks in both lists" summary. No stance.
In short: Bill/Cole — hyperscaler-mania member. Profitable and spitting out cash (Amber's pushback), but the paradigm shift is that the whole group's returns on capital are declining; the top-10 "turns over every decade," and this cohort is the crowded, over-owned one.
23:06— Yeah. So they're I mean they're cyclical. So to your point, our our view is not that their revenue isn't growing — and there might be somebody win. — That's that's not the catch. The catch is that their returns on capital are declining. — Yeah. — And so, you know, think about it like this.
In short: Cautionary example, not a call: Druckenmiller — "one of the smartest, most shrewd investors in the world" — sold his Nvidia "easily a double ago" thinking it had gone too far, too fast. Put the short ticket away; moves run farther than you can imagine.
16:56If you hear yourself saying it's gone too far too fast, and that's fine if you want to sell your long, but if you got out a pink ticket trying to go short, and you say to yourself, it's gone too far too fast, put that pink ticket away and wait another couple weeks because it can always go farther than you can imagine. Like let's not forget Stanley Dreen Miller sold his Nvidia what like 100% ago like it double — a double ago. Easily a double ago. Yeah.
In short: Emblem of an S&P that's ~30% AI — ~34 Rio Tintos fit inside its market cap; the over-owned growth side of the rotation.
Nvidia is the AI chip leader and the symbol of a stock market now roughly 30% tied to AI. His issue isn't the company — it's the crowding: everyone owns it, it's enormous (about 34 Rio Tintos fit inside it), and it's the over-owned growth side of the rotation he expects money to leave.
9:22Right now the S&P is almost 30% AI stocks, which is insane. And the street's expecting 820 data centers over the next 5 years. But there are three big hurdles — NIMBY (not in my backyard), power, and infrastructure equipment. A lot of data centers are probably going to get delayed, and the credit markets are already sniffing this out. Just look at Blue Owl. Look at Jefferies equity versus Goldman Sachs. But the CoreWeave bonds are trading with almost 12% yield.
In short: The face of the Mag-7 / AI trade Cole thinks is "greatly overpriced." Like the oil euphoria of the 2010s — where producers and servicers and industrials all got caught up — the AI mania sweeps in more than just the capex spenders. "We have to kill this fever."
NVIDIA is the poster child of the AI trade, and Smead thinks the whole thing is badly overpriced. His analogy is the late-1990s telecom boom: everyone knew the internet was coming, tons of money poured into building it, and the eventual survivors still made only tiny returns. He argues the same fate awaits the companies pouring capital into AI. NVIDIA sells the chips that fuel that spending, so in his view it's caught up in a "fever" the market eventually has to break — a caution, not a recommendation to own it.
1:47AND SO I JUST SAY THAT BECAUSE I THINK THE MARKET'S TRYING TO SWALLOW THAT AND UNDERSTAND THAT THE LIKELIHOOD IS WE'VE GREATLY OVERPRICED THIS. — SO IF WE'VE GREATLY OVERPRICED THIS, WHAT DO YOU DO. ARE YOU NOT ARE YOU NOT OWNING THESE STOCKS. AND BY THE WAY, IF YOU DON'T OWN THESE STOCKS, I'M NOT SURE WHAT STOCKS TO OWN BECAUSE I ASSUME THAT THAT IT'S LIKE A DOMINO EFFECT.
In short: 31× sales (2–3 std devs above its 20-yr mean) and a top-2 passive concentration; "everyone's in the chips — the dumbest trade in the world."
Nvidia is the AI chip leader. His objection is the price and the crowding. It trades at about 31 times its annual sales — far above its own 20-year norm — and it's one of the top two holdings that index funds automatically buy. He calls piling into chips "the dumbest trade in the world" because everyone already owns them.
Full passage: premium transcript (PDF).
In short: The funding short: Paulo shorted the post-earnings pop and added at the open. The tell is a YoY/QoQ revenue vs accounts-receivable divergence — rising cash consumption despite revenue growth, "a massive strain on working capital." Even a leaked H200-to-China headline couldn't lift it (a "news failure"); "the worm has turned," a long way down in time and price.
NVIDIA makes the chips that power the AI boom and has been the market's biggest winner. Paulo is now short it. His main red flag is technical but simple: NVIDIA's revenue keeps growing, but the money customers owe it but haven't paid yet (accounts receivable) is growing even faster, and its actual cash is getting tighter. In plain terms, sales look great on paper while real cash is under strain — often a warning sign, especially when the company is also helping fund the very customers buying its chips.
He thinks the "story" that drove the stock has peaked — and once a story turns, "it's a long way down." A sign of that: normally a bullish headline (letting NVIDIA sell chips to China) would pop the stock; instead it fell, which he reads as the market no longer rewarding good news.
In short: The face of the "Down-&-Out" semis-bear call. Semis ~37× trailing earnings with "increasingly fragile" technicals; the industry's age-old double-, triple- and quadruple-ordering curse is set to "come back to haunt the semis." "We doubt NVDA will be immune... it might be the ultimate casualty of any demand hiccup," with DeepSeek a possible "convulsion."
NVIDIA is the dominant AI-chip company and the market's biggest star, but Hay puts it at the front of his bearish "Down-&-Out" call on semiconductors. His worry isn't NVIDIA's products — it's the chip industry's recurring boom-bust pattern: when chips are scarce, customers panic-order two, three or four times what they actually need, which inflates sales until the moment demand cools and all those phantom orders get cancelled at once.
With semis trading around 37 times earnings and chart action looking "fragile," Hay thinks that curse is due to return, and "we doubt NVDA will be immune... it might be the ultimate casualty of any demand hiccup." He flags DeepSeek (a cheaper AI model that questions how many chips are really needed) as a possible trigger for a sharp "convulsion." A warning, not a price target.
Nothing matches this filter.
Verbatim excerpts from the public transcripts (auto-pulled at each mention's timestamp, lightly cleaned). Timestamps link into the video; "source page" opens that commentator's full analysis at this row. Click a mention's header line to expand it (one open at a time). For personal study — not investment advice.