In short: Lebenthal's final trade (42:49), picked after Santoli's memo. "Similar to Qualcomm where this has been an unloved stock. All the chip stocks… have been dead money for the last three months. I see them coming back to life." Talkington also names NVIDIA and Micron as still "a huge part of earnings growth" (8:32).
Micron makes memory chips, which AI data centers need in huge amounts. Chip stocks have gone nowhere for three months after a 30% drop. After hearing Mike Santoli describe semiconductors as the group coming back first, Lebenthal picks Micron as an out-of-favor stock he thinks is turning.
In short: Named only as the "more current example" of a moonshot winner in a market where semiconductor hardware drives all the gains; not a view.
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: Subscriber view, passing: "memory suppliers such as Micron and SK Hynix" as the other half of the "sell the knives" trade — lower prices "stimulate usage, which in turn drives greater demand for compute, memory."
13:22In this knife fight, you want to be the one selling the knives. Upstream semiconductor suppliers such as ASML, KLA, LAM Research, and Applied Materials, together with memory suppliers such as Micron and SK Hynix. They benefit regardless of whether the winner is Nvidia, a hyperscaler, an AI lab, or some architecture we haven't seen yet.
In short: Used to state the opposing view fairly before rejecting its durability: "the bull is they're all right to say we're growing the earnings — of a Micron is booming so I want to own it. That is not wrong." His own position follows immediately — "the bear would say well this is not sustainable" — and the shorts he actually holds are the SOXX index and S&P/Nvidia puts rather than a single memory name.
9:33In fact, 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. So the bull is they're all right to say we're growing the earnings — of a Micron is booming so I want to own it. That is not wrong.
In short: The memory bottleneck, tightening on two sides at once. Supply: "China's CXMT reportedly stalled small-scale production of HBM3E, the advanced memory used in AI processors, so that could be positive for Micron and SK Hynix and Samsung," while "Micron accelerates its HBM growth by about 60k wafers per month." Pricing: "Samsung reportedly allocated 70% of its memory production capacity through 2031 to long-term contracts, similar to SK Hynix and Micron, which was the first one to do it," with contract customers now including Nvidia, Microsoft and Google — and spot running far above contract: "HBM3E 36 gigabyte spot is now 2100, roughly four to five times LTA pricing. HBM4 16 stack spot is at 3500. DRAM export unit prices are up 36.6% from May to July, even as unit volumes fell 13%." Samsung "is considering converting its Pyeongtaek S5 foundry line to memory production as early as next year because of demand, which is a positive sign."
Micron is one of only three companies in the world that make the high-bandwidth memory chips that sit next to an AI processor and feed it data. Memory has historically been a brutal commodity business — everyone builds capacity at once, prices crash — and the argument here is that this cycle is behaving differently.
Two things happened this week. On supply, China's CXMT reportedly halted its small-scale attempt to make the same advanced memory, removing a potential fourth competitor. On pricing, Samsung has now locked roughly 70% of its production capacity through 2031 into long-term contracts, following SK Hynix and Micron, which did it first. When almost all the output is pre-sold years ahead, the leftover spot market becomes very tight — and it shows: the spot price of one advanced memory part is running four to five times the contract price.
Micron is meanwhile expanding hard, adding roughly 60,000 wafers a month of high-bandwidth memory capacity. The reason that does not break the price is that the demand for it is contracted rather than speculative. This is the "own the bottleneck" idea Singh returns to repeatedly: in a boom, the money accrues to whoever supplies the scarce input, not to whoever assembles the finished product.
Full passage: premium transcript (PDF).
In short: Terranova's named signal for the whole September call, rather than a stance on the stock. Having argued that bearish sentiment near a factor bottom is the moment not to get bearish, he says what would confirm it: "watch the semis. The semis are showing a little bit of strength here intraday. Micron I believe is higher as we speak. That's going to be my indicator." The most crowded, highest-beta expression of the momentum factor is being used as the tell for whether the factor has stopped bleeding — "I would not get too bearish right now on the fact that momentum has had this ugly quarter."
Micron appears here as a signal rather than as a recommendation. Terranova's argument is that the momentum trade has been unwinding all quarter and that the bearishness now showing up in sentiment is typically what you see near the end of such an unwind, not the start.
To avoid making that untestable, he names what would confirm it: watch the semiconductors, and specifically watch Micron, which was higher intraday. Micron is the most crowded, most volatile expression of the momentum factor, so if it stops falling while everything else is red, the selling pressure in the factor is probably exhausting. "That's going to be my indicator."
The technique generalises: whenever you make a call about a broad factor or theme, pick one liquid, high-beta name as the tell, so you find out quickly whether you are right.
In short: The third-best contributor to S&P returns this year and 14% of expected 2026 index earnings growth — "you add Micron… that gets you to 32%," i.e. NVDA + MU are a third of all S&P EPS growth.
Micron makes memory chips — the high-bandwidth memory that sits beside AI processors — and it has been one of the year's biggest winners. On her preferred measure, contribution to return (a stock's price move multiplied by how big the company is), Micron is the third-largest contributor to the S&P 500's whole return this year.
It is also the second half of the concentration problem: Micron supplies 14% of the index's expected 2026 earnings growth on top of Nvidia's 18%, so those two companies alone are a third of it. That is not a criticism of Micron — it is the reason a single disappointment from a chip company would be an index-level event.
32:46Nvidia just that company alone is 18% of that expected earnings growth. You add Micron which is another 14% that gets you to 32%. So you're talking about one third of all S&P expected earnings growth in 2026 is a function of two companies. If you go out to the top 10 in terms of top 10 earnings growth rate companies and that brings in Chevron and Exxon as I think number nine and 10 on that list.
In short: 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.
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.
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: Two-sided this week, and the two sides are in the same folder. Against: "Apple is evaluating DRAM from CXMT and NAND from YMTC, two China-based memory companies, which pressured traditional memory players like SanDisk down 6% during the week, Micron down 5%… in addition to US potential chip sanctions and tariffs, which is why memory underperformed AI overall" — and the report's own "Potential Reductions" list says to reduce exposure to traditional commodity memory suppliers (Micron, SanDisk) on the Chinese capacity ramp and hardware-tariff friction. For: the same deck carries TrendForce's server DRAM +270% year-over-year, enterprise SSD +235%, server DDR5 +13-18% quarter-over-quarter, HBM +70-140% by 2027 — "structurally bullish for the entire memory complex" — and notes that Nvidia's commitments jumping $119B → $279B is "primarily related to procurement of memory (for the next few years). So memory goes brrr." Plus the policy tailwind: Trump praised Micron's $10 billion of new US AI research labs on top of its $250 billion US commitment. Also one of the ten most active options names on both Tuesday and Wednesday. Net: the cycle data is unchanged; the marginal buyer just found a China-supply and tariff reason to sell it.
Micron makes memory chips — the components that store data next to a processor. It has been one of Singh's clearest longs, because AI accelerators need enormous quantities of expensive memory and only three companies in the world make it.
This week the evidence points both ways, which is why the stance is neutral rather than positive. Against: Apple is testing memory from two Chinese manufacturers, CXMT and YMTC, for phones sold in China. A state-backed entrant flooding the commodity end of the market is the oldest way a memory cycle ends, and the shares fell 5% on the report. The report's own list of positions to reduce names Micron explicitly, on Chinese capacity and possible hardware tariffs.
For: the same research pack contains data that says the opposite. Server memory prices are up 270% on a year ago; enterprise storage up 235%; the high-bandwidth memory used in AI accelerators is forecast up 70-140% by 2027. And Nvidia's $160 billion increase in purchase commitments was described as primarily memory procurement for the next few years. Trump publicly praised Micron's new $10 billion American research investment.
The distinction that reconciles them is which memory. The contracted, specified, high-bandwidth end is sold out and repricing upward. The commodity end — ordinary phone and consumer memory — is what a Chinese entrant can flood. Micron sits in both, which is why one headline about Apple's supplier list can take 5% off a company whose contracted order book is growing.
Full passage: premium transcript (PDF).
In short: A new, deliberately small position — and the one AI-complex equity his framework lets him own. "I actually bought some Micron on the dip, but very small position… I think memory is still gonna send it." The reason is first-hand: he priced out a personal on-premise LLM rig at "like 4 million U.S. dollars and most of the cost is memory" — so the demand is not only hyperscalers but "independents or small businesses that will want their own LLM." He frames it as Jevons paradox, and as memory becoming a commodity: "memory and compute are evolving into commodities now… memory has got another leg to its cycle that will probably break people who aren't long memory when they watch it go off."
Micron makes memory chips — DRAM and the high-bandwidth memory that sits next to AI accelerators. Memory is the part of a computer that holds data the processor is actively working on, as opposed to the processor that does the arithmetic.
McCracken arrives at Micron from first principles rather than from the AI narrative. His argument earlier in the episode is that digital computers are inherently inefficient — they win on generality, and pay for it by having to shuffle enormous numbers of bits around. That shuffling is a memory problem, not a compute problem, which is why he says "you end up needing all this memory and Micron stock's going up."
The demand evidence is personal and checkable: he priced a machine to run his own large language model locally and the quote was about $4 million, "and most of the cost is memory." If one individual faces that bill, so does every small business or independent lab that wants a private model. That is Jevons paradox — the observation that when a resource becomes cheaper or easier to use, total consumption of it usually goes up, not down, because far more people start using it.
The framing that makes it fit his book: he thinks memory and compute are "evolving into commodities." He does not want to pick which model wins (Gemini, GPT, Claude are "all doing the same thing"); he wants to own the standardised input they all consume. His warning is that this leg is not over — it will "break people who aren't long memory when they watch it go off." Note the sizing though: he calls it a "very small position," bought on a dip.
In short: His stand-in for how the semiconductor super-cycle ends: "in 10 years people look back and say, 'Oh, remember when Micron was at such and such a price,' and here's what went wrong" — with ad hoc explanations that miss the real one, that it was simply very expensive into an overdetermined system. Unprecedented cash inflows into Taiwanese and Chinese chip makers point to "a tsunami of supply in early 2028," and in a boom-bust industry "once you lock in supply, my friend, prices are going to zero" because fixed costs must be covered.
Micron makes memory chips. Memory is the textbook boom-bust business: when prices are high everyone builds new factories, the new capacity all arrives at once, and prices collapse — because a chip fab has enormous fixed costs, so once it's built you have to keep it running and dump the output at whatever price clears.
The current story says this cycle is different — a permanent "super cycle," memory prices never falling back, GPUs a permanent duopoly. Kedrosky's counter is that record cash is pouring into Taiwanese and Chinese chip makers right now, which points to "a tsunami of supply in early 2028." Once that supply is locked in, "prices are going to zero."
He uses Micron as the name people will point to afterwards — "remember when Micron was at such and such a price" — with a pile of after-the-fact explanations that all miss the real one: it was expensive going into a cycle where lots of different things could go wrong.
30:39And the semiconductor industry has a history of exactly this. If you look back, it's probably the most capital intensive boom-bust industry on earth. But from a system standpoint and a scale standpoint, it's a classically overdetermined system where in 10 years people look back and say, "Oh, remember when Micron was at such and such a price, and here's what went wrong.
In short: Peak-date evidence, not a stance: paired with Broadcom as chips that "peaked different weeks a few months ago" — a marker on the dispersion chart, with no revisit of the May-15 memory/circularity argument.
In short: The other side of Nvidia's price hike, and the week's clearest statement of where pricing power sits: Nvidia's inability "to hold the line on prices or absorb the growing costs shows how much leverage makers of memory chips like Samsung, SK Hynix and Micron have," with NAND and DRAM prices continuing "to go parabolic." Corroboration through the week: "UBS remains bullish on Micron with a buy rating given price increases and future capacity additions," and Micron committed $10 billion over the next decade to new research labs in Boise on top of $250 billion of planned US R&D and capex. Also one of the ten most active options names of the week.
Micron makes memory chips — the components that store data while an AI processor works on it. An Nvidia chip is useless without a lot of memory attached, and only three companies in the world make it at scale: Samsung, SK hynix and Micron.
This week produced the clearest possible proof of where the power sits. Nvidia, the most dominant company in technology, told its largest customers that AI server prices are rising more than 15% because it cannot absorb the rising cost of memory. When a business with 75% profit margins passes the cost through rather than eating it, the supplier is the one setting the price. Memory prices for both types of chip continue to rise steeply.
Two supporting facts from the same week: UBS keeps a buy rating on Micron precisely because of price increases and new capacity, and Micron is committing $10 billion over ten years to new research labs in Boise on top of $250 billion of planned US investment — a company spending like one that expects the shortage to last.
Full passage: premium transcript (PDF).
In short: A sell-side call and a market-cap reference, with no committee stance this episode. BMO takes it to outperform with a $1,300 target — "33% higher than where it opened today" — inside a four-name semi initiation (NVDA 340, MU 1300, AVGO 455, AMD 550). Brown lists it in the group whose valuations are already common knowledge — "we know Micron, we know AMD, etcetera" — while making the point that the unfamiliar AI names (Arista, Vertiv, GE Vernova) are where the guidance surprise lands. Semis were down 6% on the week per Santoli.
In short: Named twice, both times as a caution rather than a call. Talkington on the earnings math: technology is the biggest contributor to growth "and within tech, it's really NVIDIA and Micron skewing that even more — the earnings are outsized." Terranova uses the name as a positioning analogue for today's crowded refiner trade: the refiners "almost are beginning to look like the Micron type of memory trade at the end of June… everyone's there already." And in his opening flow read, memory is one of the three places capital is leaving: "capital is coming out of the momentum factor, it's coming out of memory, it's coming out of semiconductors."
In short: "Memory and storage moved into focus." A top-five holding in the AI-infrastructure cluster and a top-five buy for Coatue, Altimeter and Sands (three funds, level with AMAT and AMD). The framing: "as AI infrastructure scales, funds are increasingly looking beyond compute to the components that feed and store all that data." One dissenting datapoint — Druckenmiller exited MU alongside AVGO and INTC.
Micron makes memory chips — including the high-bandwidth memory that is stacked next to AI accelerators and feeds them data. An AI chip without enough fast memory is a fast engine with a narrow fuel line.
It was a top-five buy for Coatue, Altimeter and Sands, part of what App Economy calls memory and storage "moving into focus": "as AI infrastructure scales, funds are increasingly looking beyond compute to the components that feed and store all that data."
One dissent to hold alongside it — Druckenmiller exited Micron in the same quarter. Memory has historically been a brutally cyclical business, so reasonable investors disagree about whether the AI build-out has genuinely changed that. Three buyers against one seller is a lean, not a verdict.
In short: One of the ~5 S&P names that did 20% earnings growth and drove the index's hyper-concentration; the watch item is margin durability — "like Micron they're worried about it peaking out and the memory guys are 85% gross margins which is insane."
30:17Other things that I looked at for like Micron as well and SanDisk and SK Hynix like gross margin resilience is something people focus on on Wall Street. So, the target range again analysts probably expect GAAP gross margins for Nvidia around the 73 and 75% range. Like Micron they're worried about it peaking out and the memory guys are 85% gross margins which is insane.
In short: The memory-cartel thesis restated with new numbers: "Micron's AI driven profits are projected to hit 400 billion over two years. Morgan Stanley projects a combined operating income of about 400 billion for Micron in 2026 and 2027, driven by high bandwidth memory for AI that boosts gross margins to 80 to 89% and operating margins of 81 to 87%. Bank of America sees even more upside modeling $236 of EPS by 2030 with steady margins and long-term contracts locking in 100 billion of revenue through then." The pricing-power tell from the same week: "Micron said that customers remain largely insensitive to memory pricing in data centers." Deck page 43.
Micron makes memory chips, and the specific kind AI systems need — high-bandwidth memory stacked next to the processor — is in genuine short supply. Morgan Stanley now models roughly $400 billion of combined operating profit across 2026 and 2027, on gross margins of 80-89%. Those are software margins on a commodity manufacturer, which tells you how tight supply is.
The single most telling detail is smaller than any forecast: Micron said its data-centre customers are "largely insensitive to memory pricing." When your buyers stop arguing about price, you have a shortage rather than a cycle. Bank of America takes it further, modelling $236 of earnings per share by 2030 with $100 billion of revenue already locked into long-term contracts.
Full passage: premium transcript (PDF).
In short: Named as a deliberate omission from Pershing Square — "he doesn't own any Micron or Nvidia, he doesn't own any Sandisk, he's not in these AI companies," because Ackman "believes the outperformance of this subset of companies is likely temporary." Carlson relays it approvingly; consistent with his own refusal to chase the memory complex.
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: Named with TSMC/CoreWeave/Nebius as the "known" good news powering the bounce. Memory is where the crowding is worst, and he thinks semis/memory weaken again in coming months — none of them has made a new high on this rip.
Micron makes memory chips, and it's in the same bucket as TSMC here: the good news is real but already reflected in the price. Hayes uses it as an example of the "recency bias" pulling retail investors back in — they see a strong print and assume the rally continues.
What he's watching instead is positioning: memory is where the crowding is most extreme, and across the roughly 50 semiconductor charts he sent the host, not one has made a new high on this bounce. A rally that can't make new highs after that much good news is a rally being sold into, not a new uptrend.
12:19seller of this move when you just saw CoreWeave's earnings and you saw Nebius's earnings and you saw Taiwan semiconductor saying that demand is insatiable and Micron and all these guys and the answer is because that's known. So what you had, you have not seen revenue growth were earnings expectations at levels this high, 15% revenue growth since Q4 of 2021.
In short: Lebenthal's conviction buy: two tranches at the last earnings report and just after, average price ~$1,000 with the stock at 877 — "I got a lot of clients saying, hey, why did you buy it at that price?" His answer: if the cycle is "elongated by years," you're buying six times forward earnings with earnings "growing through the roof" and the company about to anniversary the CHIPS Act, after which it can buy back stock at six times. "It's a great set up." Adam Parker: "Micron's probably going to double from here"; Mizuho target 1375, UBS 1625. He rejects the bubble/bear case outright given hyperscaler reports showing "nothing but growth and growth increasing."
Micron makes memory chips, which AI servers consume in enormous quantities. Jim Lebenthal bought it in two chunks — at the last earnings report and just after — at an average price of about $1,000, and the stock is currently $877. Clients have asked why he paid up.
His answer is about the length of the cycle. Memory has historically been boom-and-bust, so investors pay a low price for peak earnings, assuming they will collapse. Lebenthal thinks this boom lasts years, which makes six times next year's earnings a bargain rather than a trap — especially since Micron is about to pass the anniversary of its CHIPS Act funding, after which it becomes free to buy back its own shares at that cheap price. Adam Parker thinks the stock doubles; Mizuho and UBS carry targets of $1,375 and $1,625. Lebenthal dismisses the bear case (that AI is a bubble and data-center construction stops) by pointing at two weeks of hyperscaler results showing "nothing but growth and growth increasing."
In short: The single largest real driver of index earnings: Micron's $25 of EPS contribution is ~25% of the entire S&P's Q2 earnings growth, on 850% expected 2026 EPS growth and another doubling penciled for 2027. The supply backdrop got better again — Micron, Samsung and SK Hynix have sold out all 2027 memory capacity, customers are allocated only 60-70% of requested volume, and DigiTimes' insiders say 2027 is "the most severe moment of memory shortage." The caveat he keeps attached: memory is highly cyclical, so any deceleration in AI capex produces "amplified downside in these earnings, which would push the Nasdaq a lot lower."
Micron makes memory chips, and this year it is single-handedly holding up the S&P 500's earnings. Its contribution was $25 of index earnings per share — roughly a quarter of the entire index's earnings growth for the quarter — on profits up 850% in 2026, with another doubling expected next year.
The supply picture got tighter again: Micron, Samsung and SK Hynix have sold every chip they can make in 2027, and customers are being allocated only 60-70% of what they asked for. Industry watchers now expect 2027 to be the worst of the shortage.
The thing to keep in mind is that this is still a commodity business. Because so much of the index's earnings growth now rests on it, any slowdown in AI data-centre spending shows up amplified — "which would push the Nasdaq a lot lower."
Full passage: premium transcript (PDF).
In short: Up 5% week-to-date. Link: "Micron is a fairly new position. It's a volatile one for sure, so it's a smaller position. But I just think the visibility is phenomenal — they have $100 billion in bookings between now and the end of the year, take-or-pay contracts. That's a lot of visibility, and they're really in the sweet spot in terms of this whole AI memory shortage. We're short everything in this world and they're right there." Sized deliberately small for the volatility despite the conviction.
Micron makes the memory chips AI servers need, and Stephanie Link has recently started a position — deliberately a small one, because the stock is volatile. What convinces her is the contracted order book: "$100 billion in bookings between now and the end of the year, take-or-pay contracts." A take-or-pay contract obliges the customer to pay whether or not they take delivery, which converts a notoriously boom-and-bust business into something close to guaranteed revenue. "We're short everything in this world and they're right there."
In short: Named (with NVIDIA, Marvell and Corning) in Terranova's concentration warning: "where the narrative falls apart is if the 493 and the broadening out falls apart… if the market concentrates back to NVIDIA, if the market concentrates back to Micron and Marvell and Corning, the momentum factor is not there. That's where the market has a challenge." Market color, not a call on the name.
In short: The structural beneficiary of Nomura's forecast that DRAM revenue goes from ~$80B (2022) to >$2.06 trillion by 2030 (ASPs $1.90/GB → $13.70 in 2026, peaking $18.60 in 2027, settling $16-17) — "the implications are huge for the big three players, Samsung, SK Hynix, and Micron… memory chips could become one of the biggest profit pools in the whole semiconductor industry." At the capitulation lows Micron traded at a 5× forward PE (SK Hynix 4×, SanDisk 5×) and is now a top-10 US company by market cap; China "cannot flood the market with HBM," leaving a supply-demand imbalance for another two or three years.
Micron is one of only three companies in the world (with Samsung and SK Hynix) that make the memory chips AI servers depend on. Nomura published a forecast this week that Singh calls shocking: the memory market growing from roughly $80 billion in 2022 to over $2 trillion by 2030 — five to seven times what mainstream forecasters assume — driven by chip prices rising from $1.90 per gigabyte to a peak near $18.60 and staying there.
The reason China can't spoil it: its national champion CXMT makes only ordinary memory, not the high-bandwidth kind AI needs, holds under 10% of the market, and costs 30% more per chip to produce. Meanwhile Micron briefly traded at five times earnings during the July panic. If even part of Nomura's forecast is right, memory becomes one of the largest profit pools in the entire chip industry.
Full passage: premium transcript (PDF).
In short: The pickaxe seller in the AI gold rush, and a genuine moat — Micron's CFO told him that even with every patent, engineer and process person, "if one of our fabs went down… we're not sure we can get the same throughput out of it… there's a part of this business that's black magic." The three memory makers can't keep up, are on allocation and are "jacking up their prices." But "even better than pickaxe makers is put the whole thing in the too hard pile": where is it in three or five years, and which of the three pulls ahead? Unknowable.
Micron makes memory chips — one of only three companies in the world that do, alongside Samsung and SK Hynix. In the AI boom they are the classic "selling pickaxes in a gold rush" business: the hyperscalers must buy from them, the three can't produce enough, customers are put on allocation (told to take a number), and prices are being raised aggressively. That price inflation is a big part of why headline AI capex numbers look so enormous.
Pabrai thinks the barrier to a fourth competitor is genuine, and he has it first-hand: Micron's own finance chief told him that even with every patent, every engineer, every process expert and the original build team, if a factory went down "we're not sure we can get the same throughput out of it" — part of the manufacturing is, in his word, "black magic" that can't be documented or copied.
Even so, he won't own it. "Even better than pickaxe makers is put the whole thing in the too hard pile." The question he can't answer is where the business stands in three or five years, and whether one of the three pulls decisively ahead of the others — and an unanswerable question is an automatic pass, however attractive the present looks.
11:19There's a lot of barriers to entry for a fourth player to come in. All the patents and scale and everything. And I remember a few years back I was talking to the CFO of Micron. — Right. — And he said that we have all the patents, we have all the engineers, we have all the process people, we have everybody. He said that if one of our fabs went down and we tried to replicate that fab and we have everything, we have all the factors of production, everyone who actually built it the first time and all that.
In short: The exhibit for "cheap doesn't save you in a bust." "Their fans were quick to point out how cheap stocks like Micron (MU) were even at their peaks. Nonetheless, MU has lost one-third of its market cap since June 24th… despite reporting blow-out earnings." Haymaker's conclusion is a process rule rather than a price call: this "once again demonstrates the prudence of systemically selling into these hockey stick-like moves, irrespective of how exciting the story sounds."
Micron makes memory chips — the DRAM and flash storage that goes into phones, PCs and, lately, AI data centres. It was the bulls' favourite rebuttal to bubble talk, because on paper it never looked expensive: even at its highest price it traded at a modest multiple of its earnings, so buying it felt like buying value rather than hype.
Hay uses it to make the opposite point. Since June 24th, Micron has lost a third of its stock-market value — and it did that after reporting blow-out results. In other words, both of the things investors were relying on (a low valuation and great earnings) were present the entire way down and neither one helped. That happens when a stock's price was driven by the flow of money into a theme rather than by its fundamentals: when the flow reverses, the fundamentals are irrelevant to the fall.
The lesson he draws is a rule for next time, not a forecast for Micron: when any position enters its "hockey stick" phase — the near-vertical part of the chart — sell some of it systematically on the way up, "irrespective of how exciting the story sounds." You will feel too cautious while it is still going up, and you will be very glad of it afterward.
In short: Luria's cleanest dislocation: "Micron sells like it's six times earnings… as if the cycle is over," while Intel at 100× and Cerebras are priced "as if this cycle is continuing through 2030." "That is inconsistent." And the ranking has flipped: "historically the CPU market's been a little better than memory… as we sit here today, I can make an argument that the memory chip market is much better than the CPU market."
Micron makes memory chips — the DRAM and high-bandwidth memory that sits next to an AI accelerator and feeds it data. Gil Luria's argument isn't about the company at all; it's about a contradiction in how the market is pricing the chip sector.
Two groups of chip stocks imply two opposite futures. Intel (about 100× earnings) and Cerebras are only worth their prices if the AI spending cycle keeps running through 2030. Micron at roughly 6× earnings is priced as if the cycle is already finished — as if next year's profits fall off. Both cannot be true, and that gap is what Luria calls a "dislocation." On top of that the usual hierarchy has flipped: CPUs (Intel, AMD) were historically the better business than memory, but "as we sit here today, I can make an argument that the memory chip market is much better than the CPU market." Cheap price on the better business, expensive price on the worse one.
34:14Okay? If you look at Micron and Nvidia to a certain extent at their valuation, their valuation implies that the cycle is already over, right? That next year is down, right? That is inconsistent. That's why Micron sells like it's six times earnings. Yeah. Six times earnings. AMD 50 times earnings. Intel 100 times earnings.
In short: The cheap-looking semi that isn't: "their pees are very low, single digits. Their price to sales though are extremely high because their margins are just off the charts" — and those margins come from scarcity, not innovation ("they haven't changed the product really at all"). With "a lot of overordering going on in the semiconductor world," if data-center growth merely slows "those boom earnings can turn to bust… those stocks just hit air pockets" — semis being "about as cyclical as an industry can get… especially RAMs."
Micron makes memory chips, and it looks cheap on the most common measure: single-digit price-to-earnings. Hay's warning is that this cheapness is a trap, and the giveaway is a different ratio — price-to-sales, which is "extremely high." Put together, those two facts mean profit margins are at extraordinary highs, and the "E" in the P/E is temporarily inflated.
Crucially, those margins come from shortage, not skill: "they haven't changed the product really at all." Prices are high because buyers are desperate. That reverses fast, and it reverses violently, because customers over-order during a shortage — "a lot of overordering going on in the semiconductor world. That's always been the case in that industry."
So if data-centre construction merely slows (his expectation — still growing, just less), "those boom earnings can turn to bust" and the stocks "hit air pockets." Memory chips are the most cyclical corner of the most cyclical industry: "especially RAMs."
41:28Their price to sales though are extremely high because their margins are just off, off the charts. And that's where your whole point about if there are physical constraints on the construction or political constraints on the construction of AI data centers and therefore a data center growth slows way down which I think is going to happen all go negative.
In short: The core of the semis-on-dips leg: Morgan Stanley forecasts 25% memory price rises and Micron's HBM is sold out through 2027. Micron +8% on Jul 21 — "we took that opportunity to trim a little bit and then we added back at the end of this past week." Corroborating evidence: a Micron/Meta white paper showing a 38× slowdown once shuffle data exceeds DRAM and spills to SSD, Elon Musk publicly thanking Micron for reasonable pricing on the Tesla call, and CXMT pricing that leaves "no alternative to SK Hynix, Samsung and Micron."
Micron makes memory chips — the DRAM and high-bandwidth memory (HBM) that AI servers consume in enormous quantities. Three things this week say the shortage is still getting worse, not better: Morgan Stanley forecasts 25% price increases, Micron's HBM output is sold out through 2027, and a joint Micron/Meta paper showed that when a workload outgrows available memory and spills onto disk, it runs 38× slower — so buying more memory isn't optional.
There's a nice tell too: Elon Musk paused a Tesla earnings call to thank Micron for selling him chips at a reasonable price. When your customer publicly flatters you, you have the pricing power. Singh trimmed into the July 21 spike and added the position back at the end of the week — this is an actively traded position, not a buy-and-forget.
Full passage: premium transcript (PDF).
In short: The third leg of the memory oligopoly and the valuation benchmark in his spreadsheet: Micron on a mid-to-high-teens FCF yield vs high-teens/low-20s for Samsung and SK Hynix — the gap he says drew SK Hynix to market its story in North America. Comparator, not the pick.
Micron is the American member of the memory oligopoly, and here it plays the role of yardstick rather than pick. In his spreadsheet Micron's free-cash-flow yield sits in the mid-to-high teens while the two Korean names are in the high teens to low twenties — a gap wide enough, he suggests, that SK Hynix has been coming to North America to tell its story and close it. The takeaway for a listener is that all three are cheap by his measure; he just prefers the cheaper two.
50:27of the valuation gap — so there's only three of them so up till a year ago when Nvidia needed high bandwidth memory there was only one call they could make right so my teammate Devon who flagged SKINX to me a few years ago after he went to head office in Nvidia, he said, "Oh, by the way, there's this company across the street that had office space across the street from Nvidia.
In short: A short-term trade — after Argus cut SanDisk to hold, "short-term, SanDisk and Micron are both buys." Micron guided next-quarter revenue +15% above consensus and sees the memory supply-demand imbalance persisting beyond 2027 (16 take-or-pay SCAs, >$100B committed over 5 yrs at floor prices).
Micron makes the memory chips AI servers and electronics need, and there's a shortage pushing prices up. After a research shop (Argus) cut peer SanDisk to "hold" and spooked the group, Singh flags Micron as a short-term buy — a trade off the washout, not a forever-hold. His backing: Micron guided next-quarter revenue 15% above expectations and expects the memory shortage to last beyond 2027, having locked in over $100 billion of committed customer orders with floor prices.
Full passage: premium transcript (PDF).
In short: Cited via the semiconductor group: the ~12–15 highest-quality semis screen to ~40% downside in a normal bear market, twice his 20% max — the semis "trade down after stellar earnings" because true cyclical investors look 2–4 years out to normalized earnings.
5:52And that's driven almost the entire return in the market this year. And based off the companies we follow within that list, we think by and large they're overvalued. And just taking the semiconductors as an example, we look at the high-quality semiconductors out of all of the industry, about 12 or 15 companies, we would say they're overvalued on average, and their projected downside in just a typical 20-25% bear market for the index would be 40% downside for the semiconductors from today's prices.
In short: Skeptical of the LTA / order-book faith: a company with "negative gross margins 3 years ago" now awarded market cap on 80%-margin promises — future revenue penciled off spot prices. From ~$100 to $1,250 to $950: "egregious."
Micron makes memory chips (DRAM) needed alongside AI GPUs, and its stock has exploded — from about $100 eighteen months ago to a peak of $1,250. Bulls point to "LTAs" (long-term agreements) and a big order book as proof the boom is locked in.
Chanos is skeptical: this is a company that had negative profit margins just three years ago, now being awarded a huge market value on the assumption of 80% margins far into the future — an assumption based on today's high spot prices. "History says you should be very cautious about that view."
18:34— The best example of that is actually DRAM, right? And so this is obviously Micron. And they've owned this market and so now all of a sudden they figured out that you need high-bandwidth memory to attach to your high-end GPUs, right? To train the models, all that stuff, right? So when you think about Micron's a great example where after their last earnings report, some of the big reporters and whoever, big bulls were saying these are LTAs. They
In short: Avoid — too cyclical, not enough moat. A bullish analyst's own numbers: EPS to $250 in 2028 then back to $50 by 2030; at a normal 8× that's ~$400 vs a ~$1,200 peak — a 2/3 drop as the base case.
Micron makes memory chips (DRAM). Finucane avoids it because the business is too cyclical — it lost money in 2022 and is now earning near Apple-level profits, showing how violently its earnings swing, and it lacks a durable competitive moat.
His killer argument uses a bullish analyst's own forecast: that analyst sees earnings per share hitting $250 in 2028 (on a doubled 90% gross margin) but then collapsing to $50 by 2030 as the memory cycle turns. Micron normally trades at about 8× earnings, so $50 × 8 ≈ $400 a share — versus a recent peak near $1,200. That's a two-thirds fall as the base case, using the optimist's numbers. Not a game he'll play.
19:00A good example of that is in 2022, they lost money. And now they're on the verge of having higher net income than Apple, which just shows how wide the swings are for a business like that. So we don't want to own something that's that cyclical. It's not a fit for us. And we don't necessarily think that they have enough competitive advantages.
In short: Split. Weiss exited: "the reason I exited Micron is that we're closer to the end of the AI trade — not AI fundamentals, but AI trade." Baruch holds it as his largest position (cost basis "like $70") — content to see it "come in a little bit" and "not at the highs going into" back-half earnings, after watching memory names post terrific reports then sell off; some leverage unwinding out of South Korea. Net: one out, one holding through.
Micron makes memory chips that AI servers need in bulk — and it's the flashpoint of the day. Steve Weiss sold out of it, explaining that "we're closer to the end of the AI trade — not the fundamentals, the trade." His point: Micron's business and revenues will keep growing, but the stock trade has run its course as the market rotates away earlier than expected.
Bill Baruch takes the other side: Micron is his largest position, bought around a $70 cost basis, and he's happy to hold it. He actually wants it to cool off ("come in a little bit") rather than go into back-half earnings at its highs, because he's watched memory names post great results and then sell off. So the desk is split — one seller, one committed holder — which nets out to neutral.
In short: "As cyclical as they come" — had negative gross margins in 2023, now 85% and stock up 700% this year on the DRAM order surge. Unsustainable: talking about $250B of capex into a coming memory oversupply.
Micron makes memory chips (DRAM), a famously boom-and-bust business — it had negative gross margins as recently as 2023 (it lost money on every chip) and now enjoys 85% margins, sending the stock up 700% this year on the data-center order surge. Fred's warning is simply that this is the peak of a cycle, not a new normal: Micron is talking about $250 billion of new factory spending just as China's CXMT and YMTC and Samsung all add capacity — the recipe for a memory glut that crushes prices and margins.
5:51as they come. It had negative gross margins back in 2023. Negative gross margins, never mind losing money. They had negative gross margins. So it's a hugely cyclical business, but now they have 85% gross margins. And because of the incredible amount of spending that's going on by the hyperscalers, it's $750 billion by the five biggest US guys and then there's another hundred billion or so from outside.
In short: Lebenthal bought it over the past three weeks (now ~3%, "more than a nibble") and won't flinch on today's ~5% drop — his "stick to your style" fundamental play: the coming Chips-Act anniversary lets Micron start returning capital to shareholders (the catalyst he waits for), and he thinks it does "fine over the intermediate to long term." Taiwan Semi's print at week's end is the first near-term read on semis.
Micron makes memory chips that AI servers need in bulk. Jim Lebenthal has been buying it over the past three weeks (now about 3% of his portfolio) and shrugged off its ~5% drop on the day — his whole point is "stick to your style" and don't panic over daily swings. His fundamental hook: the anniversary of the "Chips Act" is approaching, after which Micron can start returning cash to shareholders (buybacks/dividends) — a concrete catalyst he likes to wait for.
He thinks Micron does "fine over the intermediate to long term" and is watching Taiwan Semiconductor's earnings at the end of the week as the first real read on how the whole chip sector is doing.
In short: Bought in the Tue/Wed memory flush (−26% from its high) and it's since rallied. BofA says the memory cycle is nowhere near a peak; Micron's 2026 profit ≈ 35 years of combined prior profit; DRAM ASPs +13-18% in Q3 (TrendForce) with disciplined supply and shortage risk into 2027.
Micron makes the memory chips that both AI servers and ordinary electronics need, and there's a massive shortage driving prices up sharply. The stock had run up and then got caught in a violent sell-off of everything AI-related — it fell about 26% from its high. Singh used that panic to buy it on Tuesday and Wednesday, and it bounced. His view (backed by Bank of America and the price-tracker TrendForce, which just raised its forecast) is that the shortage is real and lasts into 2027, so the pullback was about nervous positioning, not the business breaking. Micron's 2026 profit alone is expected to roughly equal 35 years of its past earnings combined.
Full passage: premium transcript (PDF).
In short: The worked case against the memory trade. "Memory has been a very cyclical business… Periods of high revenue and profits are almost always followed by periods of low revenue and profits. Why? Because memory is a commodity business. These companies don't have any pricing power. Their profits are completely driven by supply and demand." The 2023-24 losses (with CEO Sumit Sadana quoted on customers "being very aggressive with pricing") shut down industry investment; AI demand now meets that starved supply, so "prices are sky high" and margins exceed even the 2018 peak — "and Micron's stock went down the year thereafter." Grantham on the mechanism: "If you make abnormal profits, you will receive competition. If you make obscene profits, you'll get ferocious competition." Verdict: "The market is clearly not expecting any mean reversion or increased competition. Maybe this time really is different. But that's not a bet that I want to make. And I don't think you should do either."
Micron makes memory chips — the storage inside phones, computers and AI servers. This year it has been one of the market's biggest winners, along with Western Digital and SanDisk, on the back of AI demand. Slegers' argument is not that the demand is fake; it is that memory is a commodity, and commodity profits are borrowed from the future.
The point turns on pricing power. A memory chip from one maker is interchangeable with another's, so no producer can set its own price; the price is whatever supply and demand dictate that month. His test for this is a good one: could you imagine Coca-Cola or Moody's losing money because customers demanded lower prices? No — but that is exactly what happened to Micron in 2023 and 2024, when its own chief executive says customers pushed prices down so hard that the industry stopped building factories.
That is why prices are so high now: AI demand has arrived on top of capacity nobody built during the bad years. Historically, extraordinary profits then attract exactly the investment that ends them. He cites Jeremy Grantham for the rule — "if you make abnormal profits, you will receive competition. If you make obscene profits, you'll get ferocious competition" — and points to 2018, when Micron's margins spiked, then fell, and the shares fell with them the following year. Today's margins are higher than in 2018.
The conclusion is deliberately modest rather than a short call: the market is pricing these profits as permanent, "maybe this time really is different. But that's not a bet that I want to make. And I don't think you should do either."
In short: Split: Baruch's largest position (sold out through 2026, HBM likely into '27, ~40%+; support 850–900; stays a low multiple — only a 20× rerating would top it), vs Weiss (took profit Monday off a 350 cost, "didn't want to be piggish," rebuys ~925), vs Harrington ("you trade it, I'm not sure it's investable — cash flows too ambiguous," a 100% run already prices '27–'28). All agree a 20× multiple is where it "gets egregious."
Micron makes memory chips that AI servers need in bulk. After a huge run the committee openly disagrees on it. Bill Baruch's largest position — he argues the company is "sold out" of capacity through 2026 (and its high-end AI memory likely into 2027), so earnings are locked in; he'd only call a top if the stock re-rates to 20× earnings (it usually trades far cheaper). Steve Weiss made a fortune on it but sold Monday to bank the gain ("didn't want to be piggish") and would only buy back around $925.
Jenny Harrington is the skeptic: a stock that has already doubled is "pricing in" the best case for 2027–2028, competition is intense, and a single software breakthrough could cut how much memory AI needs — so the future cash flows are "too ambiguous." Her verdict: "you trade it, it's not investable." Net: a name to rent, not marry — hence the Neutral.
In short: Lebenthal doubled his position at $60 (a value stock to him — strategic contracts + capital returns coming, ~7× on next year's estimates), following through on his final trade. Bryn bought it too, sells Sept-70 calls for $7 (cost basis ~53), notes it touched the 50-day. Should be ~15–20% of the entire S&P's 2026 earnings growth; +257% YTD, ~7× earnings. Jason: value and momentum at once.
Micron makes memory chips (DRAM and flash), which AI servers need in huge quantities. The committee is actively buying the pullback: Jim Lebenthal doubled his position at $60, calling it a "value stock" — it looks expensive on today's depressed earnings but only about 7× next year's expected earnings, and the company has locked in long-term supply contracts and will soon start returning cash to shareholders. One striking data point: Micron alone could be 15–20% of the entire S&P 500's earnings growth this year.
Bryn Talkington bought it too and, like with Nvidia, sells "call options" against it — she collects a $7 premium for agreeing to sell at $70 by September, which lowers her effective cost to about $53 and pays her to wait. The stock is up 257% this year yet still only ~7× earnings, which is why Jason Snipe says it's both a value and a momentum stock at the same time.
In short: The valuation comparator — SK Hynix comes to market priced "almost exactly like Micron," both near ~7x forward earnings and ~18x trailing EV/EBIT. At parity, buying SK Hynix "requires making the same peak-cycle bet on the same memory boom" — the only difference being SK Hynix leads HBM. Micron's blowout quarter (broken down the prior week) benefited from the same AI-memory shortage. (Recap, not a stance call.)
Micron is the big US memory-chip maker and the natural yardstick for pricing SK Hynix, because both sell essentially the same commodity (DRAM and NAND memory) and are riding the same AI-driven shortage. The article's key point is that SK Hynix is coming public priced at almost exactly the same level as Micron — roughly 7 times next year's expected earnings. That means buying SK Hynix at the IPO is making the identical bet you'd make on Micron (that today's boom-level memory prices and margins hold up), with the one difference that SK Hynix leads in the premium HBM niche. In other words, the "reward" for owning the clear leader isn't showing up in a higher price yet — the market is paying the same for both. Referenced as the valuation comparator, not a stance on Micron.
In short: The semi-momentum poster child he's rotating away from — "companies like Micron… have jacked up their prices 10 times causing incredible margins and earnings in the short term." Scarcity pricing that is historically cyclical; the driver of the ~20% semi weight and the momentum "TikTok trader" bid.
Micron is the poster child for the trade Carlson is deliberately avoiding, not joining. It makes memory chips, and right now demand so far exceeds supply that Micron has raised prices roughly tenfold ("scarcity pricing"), producing spectacular short-term margins and earnings. That's the engine behind semiconductors ballooning to ~20% of the whole S&P 500.
His caution is that this is historically cyclical: scarcity pricing eventually normalizes, and much of the current gain is price rather than more units sold. He doesn't own it, and he frames the crowd piling into names like Micron — "buying whatever is going up" — as the momentum that's leaving his quality compounders cheap.
5:05Of course these companies have an influx of demand. They are being priced with scarcity pricing. It's as if everybody wants these products all at the same time and there's not enough of them. So when supply is far below demand, the prices go up dramatically. And companies like Micron and many others have jacked up their prices 10 times causing incredible margins and earnings in the short term.
In short: A historic beat-and-raise that became a sell-the-news (rallied 16% June 25 then faded to ~$992 on over-extended retail): revenue +346%, DRAM ASPs +250%, 85% GM, all four units at record revenue, data-center >$25B, HBM4 >$1B, DC SSD >$5B. The structural news is 16 five-year SCAs with four hyperscalers covering 20% of DRAM / 33% of NAND vol with $22B deposits; $18.3B FCF, +30% dividend, 100% of excess cash to buybacks from Dec 2026 (CHIPS expiry); FY26 capex $27B, FY27 >$10B/qtr. Floor prices give margins well above prior-cycle peaks; supply tightness "structural, not cyclical."
Micron makes memory chips (DRAM and the stacked "HBM" used in AI servers). It just posted an enormous quarter — 85% profit margins, $18.3 billion of spare cash generated — and, more importantly, locked four giant customers into 16 five-year contracts that guarantee big volumes at prices with a floor above any past peak, plus $22 billion of cash paid up front. That turns a famously boom-bust commodity into something far steadier.
Oddly the stock still fell after the report ("sell the news") because too many small investors had piled in — it slid to about $992. But Singh sees the demand as structurally short of supply for years, a point underlined when rival Samsung immediately announced another 20% price hike. Micron is also raising its dividend 30% and will hand 100% of extra cash back via buybacks starting late 2026.
Full passage: premium transcript (PDF).
In short: The US leg of the three-player memory oligopoly. Same protected economics — management told him that even with all the patents, engineers and a burned-down fab's full team, they're "not sure we could rebuild it at the same cost… there's black magic." He owned it, broke his never-sell rule and regrets it; same hold-don't-chase stance.
Micron is the American member of the three-company memory-chip oligopoly, alongside Samsung and SK Hynix. Pabrai uses it to explain just how deep the moat is: Micron's own management told him that even if a factory ("fab") burned down and they still had every patent, engineer and builder, they aren't sure they could rebuild it at the same cost and output — some of the process is "black magic." That's how hard it is for anyone new to enter.
As with the Korean two, his stance is hold-don't-chase. He owned Micron, sold it against his own never-sell rule, and openly regrets it — a caution to keep the winner rather than a fresh buy signal.
15:13Micron the memory business is a very protected business it is almost impossible for a fourth new player to come into this business. When I was interacting with the senior management of Micron few years back, they told me that if one of our fabs burnt down, you know, got destroyed — and we have all the patents and we have all the engineers and we have everybody who built that fab — and we try to rebuild the fab again.
In short: Bought it last week. "A value stock" at roughly 7× forward earnings with estimates going higher — the Citi/Adam-Parker logic (if you won't buy waiting for a 5% pullback, you're saying it's a buy here). "Up 600%, so what" doesn't matter — the fundamentals changed dramatically, like those who refused to buy Nvidia in 2023. A cyclical stock, but "on where we are in the cycle, middle innings."
Micron makes memory chips (the DRAM and flash that go into phones, PCs and — now the big driver — AI data centers). Lebenthal bought it last week and calls it "a value stock," which sounds odd for something up 600%. His point: the price tells you nothing on its own — what matters is earnings. The stock trades at about 7× next year's expected profits (cheap), and those profit estimates are still being revised higher as AI demand drives memory prices up. So even after a huge run it's inexpensive relative to what it now earns.
His mental model is the one analyst Adam Parker laid out: if you're refusing to buy only because you're hoping for a small (5%) dip first, you're really admitting it's a buy at today's price — so just buy it. He compares it directly to investors who wouldn't touch Nvidia in 2023 because it had "already gone up," and missed a massive further move. He concedes Micron is "cyclical" — memory booms and busts — but argues we're only in the "middle innings" of this upcycle, so there's more to come.
In short: "Double-click on time": the cycle is bullish and will be extended relative to the past — the "strategic customer agreements" are the validation that memory is more secular than cyclical, a much bigger story than the up-228% move suggests.
Micron makes memory chips (the DRAM and flash storage that data centers and AI servers gobble up). The stock is up 228% and the obvious worry is that memory is a boom-bust "cyclical" business — prices spike, everyone builds factories, then prices crash. Terranova's counter is that this cycle is different: it's bullish and will run longer than past cycles.
His evidence is "strategic customer agreements" — big AI buyers signing long-term, locked-in supply contracts (a take-or-pay style commitment) rather than buying chips on the open spot market quarter to quarter. That contracted demand is what makes him call memory "secular, not cyclical" — i.e. a durable multi-year growth story, not a quick price spike. So he'd hold through the noise rather than treating it as a trade to flip.
In short: The structural story of the call: more profit this quarter than Nvidia a year ago (Q3 $41.5B sales / $25.11 EPS, ~85% gross margin; ~$50B / $31 next-Q guide; +16% after hours), and Q4 operating income (~$41.5B) tops every Mag7 name. The 47-year commodity cycle is "structurally broken" by 16 five-year (2026-2030) take-or-pay SCAs with floor pricing above any prior peak margin and $22B of upfront deposits — covering >half of revenue. HBM4 sole-ish supplier for Vera Rubin; "no line of sight" to supply catching demand. (Bear case noted: Chinese funds say memory has "no moat.")
Micron makes memory chips. For 47 years that was a brutal boom-bust business: prices rise, Micron makes money; prices fall, it loses money. This quarter it earned more profit than Nvidia did a year ago (85% gross margins; ~$50 billion of revenue guided for next quarter).
The real news is 16 "take-or-pay" contracts — five-year deals (2026-2030) where customers must buy set volumes or pay penalties, with a guaranteed minimum price set so high that even a bad memory downturn would leave Micron's margins above any past peak. Customers even pre-paid $22 billion to lock in supply. That converts the most volatile business in chips into a steady, high-margin one. (The bear case: Chinese funds argue memory has "no moat" and rivals can build capacity over time.)
Full passage: premium transcript (PDF).
In short: The archetypal "shiny object" — "it's not hard to just go buy Micron because it's going up," but "you're not going to be able to make a successful investing career out of" chasing momentum. Reiterates his standing chip-bubble caution; retail "piling into semiconductor stocks… always the bag holders."
31:46And so that's what I'm taking advantage of and seeking out value wherever I can find it, or in different industries that, looking for what's bombed out or what may be turning around and have upside. That's what you have to be on the hunt for. It's not hard to just go buy Micron because it's going up. Are you going to be successful chasing shiny objects for a while? You might be, but you're not going to be able to make a successful investing career out of that.
In short: The subject — fiscal Q3 revenue +346% to $41.5B, gross margin 85%, op margin 80%, EPS $25.11, adj FCF $18.3B (vs $1.9B); Q4 guide ~$50B / ~86% GM / ~$31 EPS. Almost all price, not volume. Using 16 take-or-pay SCAs (~$100B min, ~$22B deposits) to set a margin floor above the prior ~60% cyclical peak. Cap ~$150B → ~$1.3T in a year. (Recap, not a stance call.)
Micron makes memory chips — the DRAM and flash storage inside phones, PCs, cars and AI data centers. Memory has always been a brutal boom-bust business: when demand is hot, all the makers flood the market with chips, prices crash, and profits evaporate. That "the boom always invites the bust" pattern is why investors have historically refused to pay much for memory stocks.
This quarter was staggering — revenue up 346% and an 85% gross margin (meaning 85 cents of every sales dollar is gross profit) — and almost none of it came from shipping more chips. It came from prices: DRAM jumped 60%+ in a single quarter because AI created a memory shortage. The real story the article tells is Micron's attempt to break the cycle using "Strategic Customer Agreements" — multi-year "take-or-pay" contracts where big customers commit to buy a set volume at a set price (and put down ~$22 billion in deposits), whether they end up needing it or not. That locks in both how much Micron sells and at what price for years, setting a profit "floor" that's actually higher than memory's previous best-ever peak. If it works, a notoriously cyclical business starts to look more like one with a dependable baseline. The piece is a recap of that momentum, not a recommendation — and it flags the open risks: prices may be starting to soften, and a handful of customers carry most of the contracts.
In short: Simpson: the week's validation — "blew it out of the water" (top line, bottom line, guides), proving the hyperscaler AI spend is still there. Talkington: the memory names are now a bigger and bigger QQQ weight, the reason the Qs are up while the MAG 7 lag. (Weiss had shaved his Micron back before the print.)
In short: +16% on a blowout Q3 (~$41B rev, +346% YoY, ~81% GM, growing faster than Nvidia at its peak) — but the bear tell: >90% of the gain is price, not volume. "They're not selling more product… big companies are willing to pay substantially more." He doesn't own it; it's the source of the consumer-price inflation.
Micron makes the memory chips that AI hardware needs, and it just posted staggering numbers — about $41 billion of revenue in a quarter, up 346% from a year ago, with 81% gross margins, growing even faster than Nvidia did at its peak. Carlson, who doesn't own it, makes a key observation that doubles as a warning: more than 90% of that growth came from raising prices, not selling more chips. They have the same product line as before; big customers are simply being forced to pay far more.
That matters because it's where his "AI boom is cracking" thesis starts. Those price increases are now landing on companies like Apple and Microsoft, which are passing them to consumers. So while Micron's quarter looks like a triumph, Carlson frames it as the momentum darling whose boom is built on pricing power that is becoming the rest of the economy's inflation problem.
1:54They're not selling more product, they're not really inventing anything new. In fact, if you look at the exact math, over 90% of the gain in Micron's revenue and earnings and so forth is because price increases alone. Micron just has a list of products that they've had before, and big companies are willing to pay substantially more for those products.
In short: The frothy momentum he won't chase — semiconductors "going up 15% a day," moving "110 [points] in price every day." It brings "a level of frustration" watching it outrun his book, but he expects these bid-up cyclicals to "turn… it's not going to last forever," so he stays patient.
In short: The week's blowout and the purest scarcity play. AI data centers created a shortage of every kind of semiconductor; prices soared. EPS of $2.51 = a 1,215% YoY increase; revenue 41.5B = a 345% YoY increase; stock up 267% into the print yet the 2027 PE is "only 8.7 times" because earnings outran the stock. Management: chip supply stays constrained past 2027.
Micron makes memory chips (the DRAM and flash storage inside servers, phones and PCs). AI data centers gobble up so many chips that there's now a shortage of essentially every kind of semiconductor, and shortages let sellers raise prices hard. The result was a stunning quarter: earnings per share of $2.51, up 1,215% from a year ago, on revenue up 345%.
The remarkable part is valuation. Even though the stock had already risen 267% before the report, it trades at just 8.7× expected 2027 earnings — because profits grew even faster than the price. And management says the chip shortage lasts "past 2027," so the tailwind isn't over. This is Eisman's cleanest example of "play scarcity": own the supplier whose product is in short supply, not the buyer burning cash to get it.
7:15Because of the growth in AI data centers, there is a shortage of every kind of semiconductor. Prices have soared. Micron beat on both revenue and earnings. But that statement does not even begin to capture what is happening here. The company reported EPS for the quarter of $2.51, which was a, get this, 1,215% year-over-year increase.
In short: The centerpiece. Terranova: a "paradigm shift" — not a cyclical commodity but an "irreplaceable resource," 16 strategic agreements securing pricing to 2029; JOET owns it from $223; buy 25% now, scale lower. Lebenthal: just initiated — cyclical but mid-innings, ~7× forward once estimates rise, big buybacks coming. Brown: it's all on price not volume — at extreme pricing customers find workarounds (DeepSeek echo); now a $1.3T co (9th biggest) — "that's eventually how the cycle turns."
Micron makes memory chips — DRAM (the fast working memory in phones, PCs and AI servers) and NAND (flash storage). Memory has always been a brutal boom-bust "commodity": when prices spike, rivals flood the market with supply and prices crash. The committee's debate is whether AI has changed that. Terranova argues it's now an "irreplaceable resource" — AI chips literally can't run without enough memory, and Micron has signed long-term supply contracts (16 of them, out to 2029) that lock in high prices regardless of where the spot market goes. The proof prices have surged: average selling prices are up 60% for DRAM and 80% for NAND — so much that Apple has to raise product prices to absorb the cost.
Lebenthal agrees it's a great business right now but calls it cyclical, just early in the cycle — and cheap, because once next year's much-higher earnings estimates catch up the stock is only ~7× forward earnings, with big stock buybacks coming (returning cash to shareholders, which he loves). Brown is the skeptic: all of Micron's gains come from charging more, not selling more units. History says when a supplier pushes prices to extremes, customers eventually find ways to use less of the product — which is how the cycle finally turns. Practical advice from Terranova for anyone buying an already-extended stock: don't buy it all at once; buy a quarter now and leave resting orders lower, so you participate without overpaying.
In short: The momentum darling — up 277% YTD / 875% on the year as crypto/MicroStrategy money rotates in; fundamentals "going parabolic" too, so "part of this is very warranted." A name the crowd is chasing, not one he owns.
8:55They love semiconductors. They love AI. We can also look at Micron Technologies. This is one of the ones that all the attention's going to. Not only has Micron caught the attention of true AI enthusiasts, it's caught the attention of just momentum investors. Lots of people are fleeing crypto, selling out of MicroStrategy, and buying Micron.
In short: The most important earnings print next week; the memory/semiconductor segment is "completely at a blow-off top."
Full passage: premium transcript (PDF).
In short: His emblem of the "most dangerously overowned" asset class — chip stocks. "Somebody sent me a chart of Micron… I've known this about the company since the internet bubble. It was a commodity chip maker. It went up 10 times… These things are overvalued." A commodity-cyclical priced like a secular winner — the internet-bubble FOMO repeating.
Polomny isn't shorting Micron — he's using it as the flag for what he thinks is the most dangerously over-owned corner of the market: chip stocks. His point is that Micron is fundamentally a "commodity" chip maker — it makes memory chips whose price swings up and down with supply and demand, like a mining company — yet it's been bid up 10× and priced as if it's a permanent secular winner. He lived through the 1998–2000 internet bubble and says this is the same FOMO. Beating the index by owning a hot chip stock, he warns, is luck, not skill — the test is whether you can repeat it for 5–10 years.
17:47So the whole global south and east was entering its s-curve. So that was before the supply constraints that we've artificially added here and before this whole AI thing. We didn't need the AI thing for this to work but it's just accelerating, pulling it forward. — Excellent point again John. I got a very interesting question from one of my followers who asks: what asset class is most dangerously overowned today? What do you think? — That's all these chip stocks.
In short: "The Microns of the world" — the crowded semis that took over the NASDAQ-100 and are decimating big-tech free cash flow; semis "are a commodity… going to absolutely crash and burn."
6:44But there's a lot of tech stocks that are really carrying the boat. And the parabolic nature of it is a situation where you take all the cash flow of these big tech stocks and you're decimating that free cash flow and you're sending it to all these other companies like the Microns of the world, the Western Digitals.
In short: Whipsawed with the memory complex: −13% Friday, +10% Monday; Vera Rubin memory uncertainty and the Taiwan-chip-threat headline weighed on semis (Korea −9% off the highs).
Full passage: premium transcript (PDF).
In short: Memory costs up ~10× in two years — AI hardware inflation from COVID-style overordering; memory is hyper-cyclical and prices typically fall a year before demand does. Memory names sell off first in his cascade sequence; its CEO among the table-pounders.
40:13So, this inflated demand that is created by companies just taking all their cash flow and investing in each other or roundtpping I think is concerning. What it does is it allows this AI development to accelerate at a very breakneck speed. But that speed also creates supply bottlenecks and it results in these guys overpaying, right? So you look at something like Micron where the cost the cost for memory has gone up like 10x over the last couple years, right? just go to your Best Buy and try to buy RAM even
In short: Referenced — memory demand has surged so hard it made Micron a ~$1T stock (net income nearing Google's); memory orders are "exploding past logic," boosting ASML.
19:46That's why we have companies like Meta saying they're having to buy far more memory. Now, the demand for memory has surged so much that it's made companies like Micron Technologies become a trillion dollar stock. Micron this year is going to make almost as much money in net income as Google.
In short: Up ~220% YTD and gone parabolic — but "that's not what I do"; he can't value it and won't comment, beyond noting that if he can't value it most holders probably can't either.
Micron makes memory chips and its stock has roughly tripled this year. Rule won't rate it — "that's not what I do." He says he has no reliable way to value a stock like this, and pointedly adds that if he can't value it, most of the people who own it probably can't either. A flag about froth, not a recommendation either way.
0:20Well, there's a lot to discuss, so I want to start by discussing the broader indices. The S&P is still up 8% on the year. The Nasdaq's up 15%, but we've seen weakness in the last few days, much of it driven by chip stocks which have gone parabolic — names like Micron still up 220% on the year, Intel up 150% — and we're only six months in. Are we at or near a top?
In short: The memory poster child (up well over 100% YTD; colleague's coverage): pricing "has gone crazy" and earnings revisions are up an order of magnitude or more, yet memory names trade at single-digit PEs — classic cyclical-peak skepticism. Part of the constraint trade he thinks must eventually normalize against Nvidia.
Micron makes memory chips — the components that store data rather than compute it. AI servers gobble memory, pricing has "gone crazy," and earnings forecasts for memory companies have risen tenfold or more; Micron's stock has more than doubled this year. Yet memory names trade at single-digit price-to-earnings ratios — the market's way of saying "memory is a boom-bust business and these are peak earnings."
Rasgon doesn't cover memory himself (a colleague does), and his framing cuts both ways: memory is the poster child for the bottleneck trade that has crushed Nvidia's relative performance — and that divergence, he argues, eventually "has to normalize" one way or the other.
4:41Like we haven't gotten anywhere near nuts yet, right? — In terms of valuation. — In terms of valuation, not not at all. And and and so and — so give me what's what's like the the PE of Micron right now. — Oh. Well, so the memory names are probably trading at single-digit PE. Micros and that's because you know look memory is is known to be a very cyclical industry and typically in cyclical industries when earnings — wait what's the difference between a memory chip and a CPU chip — different um types of processing
In short: Friday's −13.3% (DRAM ETF −15%, SOX −10.3%) was an overvaluation reset (SOX P/E 17→28×), not a broken memory cycle — triggered by a SemiAnalysis note halving Vera Rubin NVL72 DRAM (~55→28TB) + soft Broadcom guide. Shortages still forecast; Micron is the only US-listed HBM4 supplier for Vera Rubin.
Micron makes memory chips — the components that store data for AI processors. The most advanced kind, "HBM4," is in short supply, and Micron is the only US-listed company that can supply it for Nvidia's Vera Rubin systems.
The stock dropped 13% on Friday after a research note suggested the new AI systems would need only half as much memory as expected. The house reads that plunge as the stock simply being priced too richly and getting reset — not as the AI-memory boom being over. Shortages are still expected, so they stay invested.
Full passage: premium transcript (PDF).
In short: Went from a ~$60B market cap 13 months ago to ~$1 trillion — "unsustainable"; semis are notoriously cyclical and this looks like an ending move, not a beginning move.
Micron makes memory chips. Dowd points out it ballooned from about a $60 billion company to roughly $1 trillion in just 13 months — a move he calls "unsustainable."
Chip makers are famously cyclical (booms followed by busts as shortages turn into gluts), so to him this looks like the end of a surge, not the start of one.
5:46They're not they're not beginning moves. When you have Micron go from 60 billion market cap 13 months ago to one trillion market cap. It's unsustainable. Semiconductors are notoriously cycl cyclical. So that's where we are. We think the uh stock markets are painting a picture that is totally false given what is going on in the real economy and the most recent move in the indexes since the lows in April has been all AI driven.
In short: "Highly risky" — just crossed $1T (up ~10×, +18% on the day) on a one-time memory-demand spike; the most cyclical seller. "I'm not buying Micron." Will it be good when supply catches up? "I'm not so sure."
Micron makes memory chips, and it's the centerpiece of Carlson's warning about cyclical "scarcity sellers." Memory suddenly became scarce in the AI boom, so Micron's stock is up about 10× and just crossed a $1 trillion market value. The trap, he says, is that whenever a deeply cyclical stock soars, investors convince themselves it's no longer cyclical and that the fat profits will last.
He doesn't buy that. Memory is enjoying a one-time demand surge; when new supply comes online and contracts roll over, prices and profits can fall hard. So he calls it "highly risky" and is explicit: "I'm not buying Micron" (or SanDisk). It's a great business today, but he won't pay up for peak-cycle earnings.
8:39It is structurally changing to be a more durable seller. Then we also have Broadcom, which again still has some degree of cyclicality like most businesses, but I would say that Broadcom is becoming less cyclical over time. Now, when we get to the companies that I'm more concerned about, when supply eventually catches up with demand, I believe that these ones are far more risky are the most cyclical sellers of the bunch.
In short: "Hanging on to the memory stocks for dear life." Every prior cycle said sell, but this may be the first true capacity cycle since the mid-90s — the one cycle you absolutely do not sell; takes the over on every bullish memory number.
Micron makes memory chips (DRAM and the "high-bandwidth memory" that AI servers crave). Memory is famously cyclical: prices boom on a shortage, then crash when everyone builds capacity and a glut arrives. After 25 years of cycles, Baker says the playbook normally screams "sell" right about now — margins are at record highs.
But he's "hanging on for dear life" because of one exception. He thinks this is a true capacity cycle — like the mid-1990s — where demand structurally outruns the industry's ability to add supply, not just a normal inventory swing. In a capacity cycle the up-leg lasts far longer, so selling early is the mistake. He takes the "over" on every bullish memory forecast.
8:20I would say based on every memory cycle we have had for the last 25 years this is the time to be selling memory 100%. I was actually the Micron analyst in the year 2000, like I remember going to their analyst day in Sun Valley. I'm a veteran of many memory cycles and based on history this is the time to sell.
In short: The memory face of the circularity moving "beyond NVDA": its rally was so extreme it briefly replaced Berkshire Hathaway as the largest constituent in the Russell 1000 Value index. The Compound Bros call the buying "fully rational" (EPS "quadrupled," "throw out everything you know," "almost like IPOs"). Paulo's frame: hyperscaler capex is "inextricably linked" to chipmaker FCF, so memory's earnings are part of the same flow. He regrets not owning it ("I had this") but flags it as circularity/mania evidence, not a stated buy.
Micron makes memory chips — the DRAM and high-bandwidth memory that AI servers need in huge quantities. Its profit forecast suddenly went up fourfold, the stock went parabolic, and it briefly became the single biggest company in a major "value" stock index — displacing Warren Buffett's Berkshire. Cheerleaders say that's "fully rational" because the earnings really did quadruple. Paulo's worry isn't that the earnings are fake exactly, but that they're the flip side of hyperscalers spending money they don't have on chips and data centers — so memory's boom is bolted to the same wobbly capex machine. He isn't calling it a buy or a short here; he's using it as Exhibit A that the mania has spread from Nvidia into the whole chip complex, and admits he wishes he'd owned the move.
In short: One of the memory names in the up-cycle — "what's the last time we talked about a NAND cycle? Must be a decade ago." The memory guys run ~80% margins and are signing long-term agreements "right and left," de-risking the cycle.
Micron makes memory chips — both DRAM (the fast working memory in computers and AI servers) and NAND flash (storage). Shaulov groups it with SK Hynix and SanDisk as the memory names riding a cycle he says we "haven't talked about in a decade."
What makes it attractive now: these companies are earning roughly 80% gross margins, and they're locking customers into long-term supply agreements (LTAs). Those contracts give them visibility and make the business less boom-and-bust than memory has historically been — which could earn the stocks a higher valuation.
11:43You've got memory companies like Hynix, Micron, SanDisk. What's the last time we talked about a NAND cycle? Must be a decade ago. And it's a powerful cycle. The profitability of these businesses is absolutely enormous. And if you look at forward CapEx indicators — historically how profitable the customers are leads to forward CapEx.
In short: "The Microns of the world" — the memory names driving the extreme semiconductor dislocation.
Micron makes memory chips (DRAM). He calls these "the Microns of the world" — the memory names driving the extreme, overstretched move in semiconductors.
They're a big part of why the chip sector looks so dangerously inflated to him, and a prime candidate to fall hard as that bubble deflates.
16:31Also, don't forget the Let me give you some stats. The SOXX ETF, right, which is the semiconductors, it's over 54% above its 200-day moving average because of these a lot of the Microns of the world. And then if you look at the EWY, which is the South Korea ETF, which has a lot of these semiconductor memory players in there, it's 156% above its 200-week moving average.
In short: Referenced — its stock fell after Google's "turbo quant" model needed less memory; a reminder that AI hardware/commodity demand can be disrupted.
32:24— That's already happening. On that note, RAM chip makers have been suffering. You recall Micron stock went down because Google announced a new turbo quant model that requires less memory. And who knows what that's going to do to copper or other raw materials. Copper demand is supposed to increase between 2 and 2.5% compounded for 10 years without AI. You had data centers on top of that.
In short: The DRAM shortage means "Micron is robbing the Mag 7" — memory costs balloon and crush data-center/Mag-7 margins (Micron the beneficiary).
Micron makes memory chips (DRAM) — the working memory inside computers and AI servers. There's a shortage, so memory prices have soared, and Micron is making a fortune.
His phrase: "Micron is robbing the Mag 7." The AI giants must buy huge quantities of this now-expensive memory, which inflates their costs and squeezes their profits. So Micron is a beneficiary of the very dynamic that's hurting the big tech names — interesting as a symptom, not a buy/sell call here.
7:09Like Micron, how much money Micron's making because they're robbing Mag 7. So, that is creating this downdraft in the Mag 7 equities, but the power side with the natural gas equities where they're going to actually take some of those data centers that are in the wrong locations, what we call NIMBY, not in my backyard, and they're going to move them toward other parts of say Canada and Texas where the gas is trapped and cheap.
In short: #4, +226.8% — "Micron sold out its production through 2025 and beyond… The stock surge is a result of revenue and margin expansion. In 2025, demand for high-bandwidth memory exploded. Pricing improved, utilization rose, and Micron's earnings power surprised even optimistic investors." Reported without comment here; six months later the same expanded margins become the archive's first argued short-side case (July 2026) — "memory is a commodity business. These companies don't have any pricing power."
In short: "Everyone's in Micron, Nvidia" — the over-concentrated chip trade he says to sell in favor of the power bottleneck.
Micron makes memory chips. He names it with Nvidia as the crowded chip trade to sell. His advice: stop piling into the chipmakers and instead own the "power bottleneck" — the gas, coal, nuclear and uranium that actually run the data centers.
Full passage: premium transcript (PDF).
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.