| Ticker | Name | Research | View | What he said | At |
|---|---|---|---|---|---|
| MDB | MongoDB | QT · SA · STK · FA | Positive | His infra example (and he flags the bias — he used to work there): AI is causing "a proliferation of software," vibe-coded apps and faster enterprise development, and "every software application has a database line underneath it" — an explosion in the end market should be an explosion in database software. Open question: model-selected databases could shift the mix. | 6:53 |
| NVDA | Nvidia | QT · SA · STK · FA | Positive | 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." | 11:02 |
| MU | Micron Technology | QT · SA · STK · FA | Positive | Named with Nvidia as the well-known AI name whose numbers embed peak, not extrapolation — the opposite of the behind-the-meter / services / "hidden AI winner" cohort "pricing in a rosier future." He'd rather own the low multiple on the same factor. | 11:02 |
| EWY | iShares MSCI South Korea ETF | QT · SA · STK | Positive | July's pain concentrated in Korea — a notable share of the retail population margin-called on top of hedge-fund deleveraging — inside "a secular trend that I actually think is still going pretty well for basically two large companies in that country." Once all the pieces lined up: "I want to be a buyer when people are getting margin called and liquidated." (Realized vol ~73 on a one-month look-back.) | 53:09 |
| CRWD | CrowdStrike | QT · SA · STK · FA | Neutral | "I'd argue that the real AI winners are CrowdStrike and some like that rather than maybe some of the semis" — cyber numbers will be good because everyone is racing to defend against new AI threats. But the multiple is the problem: ~25–30× revenue for large companies, "the success rate on those is paying up there is low," and the labs themselves think cyber is something they will disrupt. "Lofty territory… more so than AI, frankly." | 6:25 |
| FSLY | Fastly | QT · SA · STK · FA | Neutral | His CDN example of usage-based uplift — "anything that can deliver computing resources… more compute flows through to their numbers," so the short-term spike can be quick, "the long-term still more under question." Also cited as a 2020–21 retail-mania exemplar (paid newsletters written about MongoDB, Elastic and Fastly). | 7:34 |
| CRM | Salesforce | QT · SA · STK · FA | Neutral | A former employer. The de-rating was "somewhat rational" — a recalibration from a decade of double-digit growth to high-single/low-double digits, "not actually pricing that AI was going to disrupt the entire business." At 5–15% growth "it's really hard to dream the dream"; low-to-moderate grower with some left-tail risk, and less likely than infra to blow up numbers near term. | 4:50 |
| WDAY | Workday | QT · SA · STK · FA | Neutral | Grouped with Salesforce in the application-software recalibration — "obviously we've seen an acquisition potentially of Workday," i.e. what happens to low-growth application software once the multiple resets. | 4:50 |
| NOW | ServiceNow | QT · SA · STK · FA | Neutral | Exhibit for the technical read on the software bounce: "some bad news for AI leads into ServiceNow and Salesforce stocks go up several percent" — an odd dynamic that says the rally is partly the unwind of paired long-AI / short-software books. | 4:03 |
| GOOGL | Alphabet (Google) | QT · SA · STK · FA | Neutral | Search still growing double digits and the stock "reasonably cheap" (he owns some megacaps) — but it just printed its first negative free-cash-flow quarter and the sell-side's snap-back to positive FCF in 2028 presupposes ROIC visibility nobody has. Sergey and Larry coming back to work on AI is his evidence the spend is existential, not financial. | 27:28 |
| AMZN | Amazon | QT · SA · STK · FA | Neutral | "Amazon… is probably up something like 10x since 2016 or 2017" — you made extreme returns owning the most well-known company in the world, and "that profile of return… is probably behind us." AWS reaccelerated from a hoped-for 18% to the 40s and the stock barely moved, which he reads as queasiness about ROI and customer concentration. Still 15–20%/yr possible; just a different profile. | 28:02 |
| MSFT | Microsoft | QT · SA · STK · FA | Neutral | His illustration that capex is an option, not concrete: "Microsoft talked about this in the latest call… you're buying powered shells, you're not filling it with the chips yet. You're essentially buying a couple years of optionality" — capacity you can ramp fast if you need it. Rational even at a low-but-positive ROI. | 30:46 |
| META | Meta Platforms | QT · SA · STK · FA | Neutral | Growing in the 20s and the biggest relative capex spender; the "we could sublease the excess compute" message is signaling to buy rope from investors, not a real profit lever. But Mallaby's book has Zuckerberg equally enthusiastic about AI, crypto and NFTs — he "has a tendency to go big on tech trends" the core business doesn't need. "That's a more binary question as a shareholder. It's not one I can underwrite." Meta's troubles aren't AI's troubles. | 37:17 |
| AAPL | Apple | QT · SA · STK · FA | Neutral | Cited as the bull's durability data point: "Apple services is still growing double digits today" — alongside Google search and Meta in the 20s, the basis for a possible 10–15 years of double-digit growth that "lifts all boats." Also excluded (with Meta) from the hyperscaler-cloud engine that drove Mag-7 performance. | 2:41 |
| ORCL | Oracle | QT · SA · STK · FA | Neutral | Named in the roll-call of scaled compute providers — Oracle, SpaceX, the labs procuring directly, the neoclouds — that took the cloud market "from basically three vendors with thousands of customers" to nine or ten providers with fewer customers. Structurally why hyperscaler multiples don't re-rate on growth. | 29:01 |
| IGV | iShares Expanded Tech-Software ETF | QT · SA · STK | Neutral | "It's been entertaining to see socks [SOX] and IGV basically have a negative one correlation" — the cleanest evidence that the software rally is substantially the unwind of long-AI / short-software pairs rather than a verdict that software isn't disruptable. | 4:03 |
| SOXX | iShares Semiconductor ETF (the SOX index) | QT · SA · STK · FA | Neutral | Q2 delivered a 100% gain in the SOX with heavy retail involvement, so chopping around afterwards is "to be expected" — but he doesn't think late June was the end of the mega trend. Realized vol ~55 on a one-month look-back (peak ~185): "how much leverage do you need for sectors that are realizing that level of volatility?" | 4:03 |
| TSLA | Tesla | QT · SA · STK · FA | Neutral | 2021-analogy reference: the SPACs and GameStops cratered in Q1 2021, but "stocks like Tesla, software eventually kind of hit their peak in late 21" — the real businesses held up until inflation and rates did the damage in 2022. | 22:01 |
| Anthropic | Anthropic (private) | — | Neutral | The S-1 lands "next week or the week after"; he expects little genuinely new information (the numbers are broadly leaked) but says having the labs public replaces third-hand ARR leaks with metrics and dampens narrative volatility. On the listing itself: "I think it's going to trade at a crazy price. I'm not saying I'll buy it, but there's going to be a lot of enthusiasm for it." | 1:00:50 |
| OpenAI | OpenAI (private) | — | Neutral | With Anthropic, the ARR the whole complex trades off: "people are latching on to every ARR leak of Anthropic or OpenAI" in an information vacuum. Both are also building their own power and data centers — part of why the profit pool's landing spot in the compute chain is unclear. | 19:46 |
| SpaceX | SpaceX (private) | — | Neutral | The capital-vacuum precedent — the moment people could buy SpaceX they sold the space also-rans to fund it. He isn't very worried about a repeat for AI: equity issuance into real businesses at realistic prices "is less capital sucking than we need to put $10 billion into Nikola." "I would certainly quibble with the SpaceX valuation." Also counted in the ~$500B the labs + SpaceX + Google raised year-to-date, which markets absorbed. | 1:00:11 |
| xAI | xAI / Elon Musk's compute buildout (private) | — | Neutral | "Elon is entering that space" and is "pivoting all of his attention to building data centers" — new supply that could "move us quicker to the glut" and shorten the window in which today's extremely high price of compute holds. | 10:34 |
| CRWV | CoreWeave | QT · SA · STK · FA | Negative | Asked where to be in AI he "started with a negative": the risk of extrapolating a very short-term dynamic — "the pricing of compute, certainly maybe with neoclouds" — while Elon's entry may hasten the glut. Named among the nine or ten scaled compute providers that turned a three-vendor market into a crowded one with fewer customers. | 29:01 |
| NBIS | Nebius Group | QT · SA · STK · FA | Negative | The other named neocloud in the same caution: today's very high compute pricing is the short-term dynamic he'd be most careful extrapolating, and the supply set (Oracle, SpaceX, the labs, the neoclouds) keeps widening. | 29:01 |
"View" is Jeff Keller's stance in this conversation (Positive / Neutral / Negative), not a price rating. Several names are analytical references rather than positions — he declined to discuss individual book positions. Research links: QT Qualtrim · SA Seeking Alpha · STK Stock Analysis. Private companies have no ticker.
A jargon-free summary of the thesis behind each name — what it actually is and why he holds that view. (Plain-language companion to the table above; renders on each ticker's consolidated page.)
MongoDB sells the database — the filing cabinet every piece of software stores its information in. Keller used to work there, and says so up front. His argument is simple arithmetic: AI is making it far cheaper and faster to create software, both for amateurs writing "vibe-coded" apps and for engineers inside big companies. "Every software application has a database line underneath it," so if the number of applications in the world explodes, the amount of database software sold should explode too.
The honest caveat he adds: databases used to be chosen by human developers, and increasingly they're chosen by a large language model writing the code. Nobody knows yet which vendors the models will favour, so the market-share picture is genuinely open. His bet is on the rising tide rather than on the share shift.
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.
Micron makes memory chips, one of the most cyclical businesses in technology — its profits swing violently between shortage and glut. That cyclicality is exactly why its valuation stays low: the market assumes today's earnings are the peak.
Keller treats that as the attraction rather than the flaw. If you want exposure to AI capex and lab growth, you can buy it here at a multiple that already assumes a downturn, or you can buy it in a name priced for years of uninterrupted growth. Same driver, very different price of admission — and he'd rather own the cheap version.
EWY is the simplest way for an outsider to own the Korean stock market, which is dominated by a couple of enormous technology companies. In July it was hit by forced selling, not bad news: leveraged hedge funds were unwinding, and a striking share of Korean retail investors got margin calls — meaning their brokers made them sell whatever they held, at whatever price, to cover borrowed money.
That distinction is his entire trade. When holders are selling because they have to, the price stops reflecting the business — and he thinks the underlying secular story for "basically two large companies in that country" is still going pretty well. His discipline is not to guess at it, but to wait until all the pieces are visible (the deleveraging, the margin calls, a 3–4 standard-deviation move) and then be the buyer on the other side of the liquidation. Note the volatility involved: the ETF was realizing about 73% annualized over the month he spoke.
CrowdStrike sells cybersecurity software. Keller's fundamental view is genuinely bullish: "the real AI winners are CrowdStrike and some like that rather than maybe some of the semis," because AI creates new kinds of attacks and every organisation is now rushing to upgrade its defences. In a market where nobody can see three years out, a business whose next two quarters are almost knowable gets paid a huge premium — and that's what cyber has been.
The other half of his view is the price. Large cyber names trade around 25–30 times revenue — not profit — and "the success rate on those is paying up there is low." He also notes that people at the AI labs consider cybersecurity something they intend to disrupt too. So: right theme, very demanding entry price. He won't call himself bearish, but he thinks it's "lofty territory… more so than AI, frankly."
Fastly runs a content-delivery network — the plumbing that moves data and computing closer to whoever is using an app. It gets paid by usage, so when overall computing activity rises, its revenue rises automatically without it having to sell anything new. That's why it and similar names bounced: more AI activity mechanically flows into the numbers.
Keller's caution is that a usage spike is easy to see in the short term and much harder to underwrite long term. He also uses Fastly as a cautionary marker from the last cycle — in 2020–21 people were launching paid newsletters about it, which in his experience is the sign that a trade's clock has started ticking.
Salesforce (another former employer) is the standard-bearer for "application software" — the finished business programs companies buy, as opposed to the plumbing underneath. Its collapse was widely read as the market pricing in AI destroying its business. Keller thinks that's mostly wrong.
What actually happened, in his telling, is a recalibration. For a decade these companies grew at double digits and were valued at a premium to everything else. Now they grow at maybe 5–15%, and at that rate the old valuation tricks — multiples of revenue, adjusted earnings — stop working and "it's really hard to dream the dream." So the de-rating was "somewhat rational": these are now low-to-moderate growers with a bit of tail risk, and the bounce off the bottom makes sense too. That's a stance about the multiple, not a verdict that AI leaves the business intact.
Google's core search business is still growing double digits and Keller thinks the stock is reasonably cheap — but it just reported its first quarter ever where the business consumed more cash than it produced, because of the sheer scale of AI spending. Wall Street's forecasts show that reversing and free cash flow turning strongly positive again by 2028, which quietly assumes everyone knows what the return on this spending will be. Nobody does.
His read is that the spending continues regardless. Google's founders came back to work on this personally; the leadership treats it as existential and generational, not as a return-on-capital exercise. That's good for the technology and ambiguous for shareholders: the money keeps going out the door whether or not the payback maths works, and the era when simply owning the biggest tech companies produced extraordinary returns is probably over.
Amazon is Keller's illustration of how the megacap trade has changed. The stock is up roughly ten times since 2016–17 — you got extraordinary returns for owning one of the most famous companies on earth. He thinks 15–20% a year is still achievable from here, but that the multiple-bagger profile people mentally attach to these names is behind us.
The telling detail is AWS, Amazon's cloud business. Two years ago investors were hoping it could grow 18%; it's now growing in the 40s — and the stock is only incrementally higher. That gap between spectacular growth and a muted share-price response is, to him, the market quietly expressing unease about how much of that revenue converts into durable profit, and about how few customers it now depends on.
Microsoft supplies Keller's clearest example of how to think about AI capital spending: it isn't concrete, it's an option. On its latest call the company described buying "powered shells" — data-center buildings with electricity connected but no expensive chips installed yet. That is deliberately buying a couple of years of the ability to scale up fast, without committing the full cost.
Read that way, the arithmetic everyone argues about ("what return does this $1,000 in the ground earn?") is the wrong question. You spend because not spending could cost you the whole business, and because the upside if you win is open-ended. A modest positive return on that option is still a rational purchase — which is why he expects the spending to continue no matter what the share price does.
Meta is the biggest AI spender relative to its size and therefore takes the most criticism. Its message that it could rent out any excess computing power is, Keller argues, pure signalling — a few billion dollars from short contracts is meaningless to a company this size. The real purpose is to tell investors "we hear you, and we'll stop if we have to," so they'll extend the company some rope. He thinks that's rational, and the underlying spending is rational too, for the same existential-option reason as its peers.
His hesitation is about the person, not the plan. Quoting Sebastian Mallaby's book on DeepMind's Demis Hassabis, he notes Zuckerberg was equally enthusiastic about AI, crypto and NFTs — a pattern of going big on tech trends the core advertising business doesn't need. Owning Meta therefore means underwriting Zuckerberg's judgement on a binary bet: "it's not one I can underwrite." Importantly, he separates that from the broader AI trade — Meta's stumbles are Meta's, and there are always winners and losers among the AI model builders.
CoreWeave is a "neocloud" — a company that buys enormous quantities of AI chips and rents the computing power out by the hour. Its economics depend heavily on the price of compute, which is extremely high right now because demand far exceeds supply.
Asked where investors should be in AI, Keller deliberately opened with a negative, and this is it: the danger of assuming today's very high compute prices persist. Elon Musk is entering the same business at scale, which he thinks could "move us quicker to the glut." And the wider structural point — the market for renting compute has gone from three serious providers to nine or ten (the hyperscalers, Oracle, SpaceX, the labs building their own, plus the neoclouds), while the pool of customers is small and concentrated. More sellers and fewer buyers is not a combination that protects pricing.
Nebius is the other neocloud he names, and it falls under exactly the same caution: its value rests on renting out AI computing power at today's very high prices, which is precisely the short-term dynamic he'd be most careful about extrapolating. New supply keeps arriving — from Musk, from the labs building their own data centers, from Oracle and SpaceX — into a market with relatively few, very large customers.
Anthropic is one of the two leading AI labs and is about to file the paperwork to go public. Keller doesn't expect the filing to reveal much — the revenue figures have largely leaked already — but he thinks having a lab publicly listed is genuinely important for everyone else. Right now the entire AI complex trades on third-hand rumours about lab revenue; once there are audited numbers, the whole sector gets a real gauge instead of guesswork, and the narrative swings should calm down.
On the listing itself he's deliberately non-committal: "I think it's going to trade at a crazy price. I'm not saying I'll buy it, but there's going to be a lot of enthusiasm for it." The market wants direct exposure to an AI lab and currently can't get it — you can only buy it wrapped inside Google, or indirectly through chipmakers. He's also relaxed about the money it will absorb: the labs, SpaceX and Google will have raised roughly $500 billion this year and markets took it in stride.
Summary & timestamps derived from the public YouTube video (transcript in transcript.txt) for personal study. Not investment advice. © Monetary Matters Network / Other People's Money for source material.