Rasgon covers US semis & semicap at Bernstein (he rates NVDA/AMD/INTC/QCOM/AVGO/AMAT/LRCX/KLAC; memory, ASML and software are colleagues' coverage). Stance reflects how each name was framed in this conversation, not a formal rating. Research: QT Qualtrim · SA Seeking Alpha · STK Stock Analysis.
| Ticker | Name | Research | View | What he said | At |
|---|---|---|---|---|---|
| NVDA | Nvidia | QT · SA · STK · FA | Positive | Growth is accelerating (85% vs 65% two quarters ago, ~75% GM, $91B guide vs the famous $11B "big bang" of May 2023) yet the stock lags because investors play the constraint names — a divergence that "has to normalize" with Nvidia coming up. Plus a ~$20B CPU story this year (Grace, standalone CPU racks) as agentic AI explodes CPU demand; supply secured across the value chain. | 13:17 |
| AMD | Advanced Micro Devices | QT · SA · STK · FA | Positive | Just upgraded it (admittedly late): genuinely good server CPUs taking "oodles of share" from Intel (x86 server revenue share 0.1% in 2015 → low-mid 40s; up 70%+ this year, not the street's 50%), plus OpenAI/Meta multi-gigawatt GPU deals (warrants he dislikes but understands). His 2027 EPS ~$14+; if Lisa Su's doubled $120B CPU TAM holds, ~$20 EPS by 2028 — the 2030 target two years early. | 27:26 |
| AVGO | Broadcom | QT · SA · STK · FA | Positive | AI business (custom chips — e.g. Google's TPUs, a 15-year relationship — plus AI networking) guided to $100B next year, "way bigger than the entire company was a year or two ago," and "they'll probably do a lot better than 100." Stock lagged like Nvidia (accelerator fatigue + the software taint), but its software (VMware) is infrastructure AI runs on — a baby thrown out with the SaaS bathwater. | 41:52 |
| ASML | ASML Holding | QT · SA · STK · FA | Positive | Owns lithography — the patterning step that defines how small a feature you can print, "the most critical step" for advanced chips: ~90% market share, almost 100% in the most advanced (EUV) tooling. China can't buy ASML — a key reason its chips aren't competitive. | 45:46 |
| LRCX | Lam Research | QT · SA · STK · FA | Positive | Of his three semicap names, "Lam's probably done the best" — up on the order of 200% year-over-year (deposition/etch, levered to the memory boom). Semicap correlations are high: "you could own the basket and be okay." | 46:24 |
| AMAT | Applied Materials | QT · SA · STK · FA | Positive | One of the big five (70%+ of all wafer-fab equipment); does the "put stuff on / take stuff off the wafer" steps. "They're all doing good… if it's working, they will all work to greater or lesser degree." | 45:19 |
| KLAC | KLA Corporation | QT · SA · STK · FA | Positive | The process-control / inspection specialist — monitors wafers for defects while they run. The "quote-unquote worst" of his three semicap names lately, but still up ~100% year-over-year; the whole group works. | 46:09 |
| Anthropic | Anthropic (private) | — | Positive | His monetization proof-point: annualized revenue run-rate has "gone vertical" — ~$1B a year ago, $9B in December, $14B in January, $30B, then $44B a few weeks ago. Agentic coding (Claude) is the real use case that's "reached takeoff velocity." | 53:14 |
| INTC | Intel | QT · SA · STK · FA | Neutral | "You take lucky over good": CPU demand is so hot Intel got ~200bps of margin selling previously written-off, by-their-own-admission uncompetitive server parts. Balance sheet fixed (government/Nvidia stakes; bought out the dilutive Apollo Ireland-fab deal), 18A/14A yield statements "very carefully worded" (yields still clearly not good), packaging IP a real AI angle — and the unshortable overlay: "Trump wants the stock to go up." Likes Lip-Bu; a lot is priced in, but the narrative is going their way. | 32:22 |
| MU | Micron Technology | QT · SA · STK · FA | Neutral | 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. | 4:41 |
| SNDK | SanDisk | QT · SA · STK · FA | Neutral | His jaw-dropper for how hot memory is (colleague's coverage): SanDisk just guided to a single-quarter EPS (~$31–32) higher than the entire stock price when it went public ~18 months ago. | 3:36 |
| QCOM | Qualcomm | QT · SA · STK · FA | Neutral | ~70–75% smartphones (plus the pure-profit cellular-IP licensing), now chasing data center: a CPU and 200MW AI-rack deal with Saudi Arabia's Humain (nothing delivered yet) and a just-announced mystery hyperscale ASIC — no part, customer, size or date — that still sent the stock up 70%: "tells you how nuts things are." Analyst day end of June. | 38:14 |
| TSM | Taiwan Semiconductor (TSMC) | QT · SA · STK · FA | Neutral | Cited as the leading-edge logic supply Nvidia has locked down — leading-edge logic and CoWoS packaging are "always tight," but Jensen "saw this coming" and secured supply across the value chain. | 13:52 |
| TOELY | Tokyo Electron | SA · STK | Neutral | One of the semicap big five (with AMAT/Lam it does the deposition/etch steps); the group holds 70%+ of the wafer-fab-equipment market and "they're all doing good." | 43:50 |
| GOOGL | Alphabet (Google) | QT · SA · STK · FA | Neutral | Builds its own TPU custom AI chips with Broadcom (a 15-year relationship now exploding) and is spending $180B of capex this year — part of the "table stakes have exploded" hyperscaler arms race he calls existential but rational. | 41:30 |
| META | Meta Platforms | QT · SA · STK · FA | Neutral | "Poor little Meta" — spending $135B with no data-center rental business to monetize it, so investors fret each capex raise; but it has heavy internal AI uses, he thinks it is getting a return, and an AMD GPU customer (multi-gigawatt deal). The day a Meta cuts capex "it's all over… but by the time you see it, it's too late." | 50:14 |
| AMZN | Amazon | QT · SA · STK · FA | Neutral | The biggest spender in the arms race — $220B of capex this year (vs Google's $180B, Meta's $135B); cited as evidence the table stakes "have exploded" and capex keeps being revised up, not down. | 49:55 |
| ARM | Arm Holdings | QT · SA · STK · FA | Neutral | Passing mention: Nvidia's Grace CPU — 36 of them in every GB300 NVL72 rack — is built on the ARM architecture, putting ARM inside the agentic-CPU wave. | 16:04 |
| APO | Apollo Global Management | QT · SA · STK · FA | Neutral | Bought half of Intel's Ireland fab for $11B when Intel was desperate — "private equity guys don't work for free"; Intel just paid to unwind it ("a pretty good return over the three years"), removing the earnings dilution. | 32:57 |
| AAPL | Apple | QT · SA · STK · FA | Neutral | Rumored again as an Intel foundry customer — "it will be small," but people hope it leads to something else. | 34:47 |
| OpenAI | OpenAI (private) | — | Neutral | Eisman floats the bear theory (a "shell game," the "weak sister" whose IPO opens the kimono); Rasgon doesn't take the bait — his counter is monetization evidence and use cases. Also one of AMD's two multi-gigawatt GPU deals (with warrants). | 47:21 |
| Huawei | Huawei (private) | — | Negative | China's chips "are not competitive right now" — partly because the semicap sanctions have been successful (no ASML access). Huawei's weekend announcements aren't unknown tech, just pursued earlier than the rest of the world "because they have no choice." | 55:48 |
"View" is Rasgon's framing in this episode (Positive / Neutral / Negative), not a formal price rating — Bernstein ratings are mentioned only for AMD (upgraded) and Intel (not short). Research: QT Qualtrim · SA Seeking Alpha · STK Stock Analysis.
A jargon-free summary of the view on each name — what the business does and why he frames it that way. (Plain-language companion to the table above; renders on each ticker's consolidated page.)
Nvidia makes the GPUs (graphics processors) that train and run AI — the picks and shovels of the entire boom. Rasgon's headline: growth is speeding up, not slowing down. Revenue grew 65% two quarters ago and 85% last quarter, at roughly 75% gross margins, and the company just told investors to expect $91 billion next quarter — nearly ten times the $11 billion guide that shocked Wall Street in May 2023.
So why has the stock lagged the rest of the chip sector? Three reasons: it already had a huge run; investors prefer to chase whatever part of the supply chain is scarcest (memory, equipment, optics — where earnings estimates have jumped tenfold); and at over $5 trillion — about 8% of the S&P 500 — many funds are simply not allowed to own more of it. Rasgon sees that gap as the opportunity: the bottleneck stocks can't be right while Nvidia is wrong, so "either the constraints come down or Nvidia comes up." Bonus: Nvidia now has a CPU business (~$20 billion this year, about the size of Intel's or AMD's) riding the agentic-AI wave, and it locked up its own supply chain years in advance.
AMD is the number-two maker of both server CPUs (the general-purpose brains in data centers) and AI GPUs. Rasgon just upgraded it — admitting he's late and that his record of recommending AMD is "horrific," while staying out of it has cost him more. Two engines drive the call. First, CPUs: "agentic" AI (AI that goes off and does tasks — booking trips, writing code) runs much of its work on ordinary CPU cores, so CPU demand is exploding. AMD has genuinely better server chips than Intel — its share of that market went from 0.1% a decade ago to the low-40s — and its server business now looks like it grows 70%+ this year, not the 50% Wall Street had modeled.
Second, GPUs: AMD signed huge multi-billion-dollar supply deals with OpenAI and Meta. It had to hand both customers warrants (rights to buy ~10% of the company each) to get the deals — effectively paying for a ticket on the rocket ship because its chips alone weren't winning enough share — which Rasgon dislikes but understands. The math: if CEO Lisa Su's (recently doubled) forecast for the CPU market holds, AMD could earn about $20 a share by 2028 — hitting its 2030 target two years early — "and at that point you can underwrite quite a bit of upside."
Broadcom is two companies in one: a sprawling chip business and a software business (VMware and others). The part that matters now is AI: Broadcom designs "custom" AI chips for the giants — most famously Google's TPUs, a partnership running 15 years — plus the networking chips that tie thousands of AI processors together. That AI business is forecast to do $100 billion of revenue next year, more than the entire company's sales a year or two ago, and Rasgon's guess is "they'll probably do a lot better than 100."
The stock has lagged for the same reasons Nvidia has (investors chasing scarcer parts of the supply chain), plus guilt by association with software stocks, which the market has been dumping on fears AI will replace them. Rasgon calls that a mistake here: Broadcom's software is plumbing (the "virtualization" layer that cloud computing — including AI — actually runs on), not the kind of application AI threatens. A baby thrown out with the bathwater.
ASML (Netherlands) makes lithography machines — the tools that project a chip's circuit pattern onto silicon. That patterning step decides how small a feature you can print, which is what makes a chip "advanced," so it's the most critical step in the whole process. ASML has about 90% of the lithography market and essentially 100% of the most advanced (EUV) machines: a true monopoly on the choke point of progress.
It's also a geopolitical weapon: China is barred from buying ASML's best machines, which Rasgon says is a big part of why Chinese chips "are not competitive right now" — China has the electric power to run AI data centers, but not the chips to fill them.
Lam Research makes the equipment that deposits material onto silicon wafers and etches it away — two of the four basic steps repeated dozens of times to build a chip. Lam is especially levered to memory chips, which are the hottest corner of the boom (memory prices "have gone crazy"), and it has been the best performer of Rasgon's three equipment names — up on the order of 200% year over year.
His broader point about the equipment makers: the "big five" (Applied Materials, Lam, KLA, ASML, Tokyo Electron) control 70%+ of the market and move together — "you could own the basket and be okay."
Applied Materials is the biggest and broadest of the chip-equipment makers — like Lam, it handles the "put material on the wafer / take material off" steps, across the widest range of tools. Chips are built like a layer cake: 30–50 layers of microscopic wiring built up by repeating deposit-pattern-etch-inspect cycles, and Applied sells machines for much of that cake.
Rasgon's stance on the whole equipment group is simple: when AI forces everyone — logic, memory, foundries — to add capacity, all the toolmakers win "to greater or lesser degree." They're all up big; he covers AMAT, Lam and KLA and is positive on the basket.
KLA dominates "process control" — the inspection and measurement machines that watch wafers as they're built, hunting for microscopic defects before a chip is ruined. The smaller the features get, the more inspection matters, so KLA is a toll-taker on advancing complexity.
It's been the "worst" of Rasgon's three equipment stocks lately — which still means up roughly 100% year over year. The group rises together; KLA just rose slightly less.
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.
Intel's stock has ripped, and Rasgon's verdict is "you take lucky over good." CPU demand is so strong that Intel earned extra margin selling server chips it had already written off as unsellable garbage — by Intel's own admission its products aren't competitive, but right now customers will take anything. Meanwhile the balance sheet got fixed (the US government and Nvidia took stakes), it bought its way out of an expensive private-equity deal with Apollo, and its statements about new manufacturing processes (18A, 14A) sound better — though Rasgon notes they're "very carefully worded" and the margin guidance says yields are still poor.
Why won't he short it, despite making his career being negative on Intel? "Trump wants the stock to go up" — the President took a stake and tweets stock charts; in hindsight the right move was buying that day. He likes new CEO Lip-Bu Tan (underpromise, overdeliver — the opposite of Gelsinger's execution). Lots of wood left to chop, a lot priced in, but "the narrative is going their way."
Qualcomm makes the chips inside smartphones (processors, modems, Wi-Fi) — about three-quarters of its chip business — plus a lucrative licensing arm that collects a royalty on essentially every 3G/4G/5G phone sold, whether or not it contains Qualcomm chips. The problem: phones are a mature market, so the story is diversification into data centers.
So far that's mostly promises: a CPU and a 200-megawatt AI-rack deal with Humain (Saudi Arabia's AI consortium) with nothing delivered yet, and a just-announced custom AI chip ("ASIC") win where nobody knows the product, the customer, the size, or the timing — and the stock still jumped 70% on the announcement. Rasgon's read: that move "tells you how nuts things are," not that the business has been proven. Analyst day at the end of June should fill in details.
Anthropic (the private company behind the Claude AI models) is Rasgon's favorite evidence that AI spending is producing real revenue, not just hype. Its annualized revenue run-rate has "gone vertical": roughly $1 billion a year ago, $9 billion in December, $14 billion in January, $30 billion, then $44 billion a few weeks before this episode.
The driver is "agentic" coding — AI that writes software, spinning up dozens of sub-agents that each handle a piece of the work. Rasgon calls it the first AI use case to reach "takeoff velocity": real demand, and a real willingness to pay. It's also why he dismisses the "no return on AI" doomsday — companies are demonstrably paying for tokens, in some cases more than they pay for the staff they replaced.
Huawei is China's flagship chip effort, and Rasgon's point is geopolitical: China's AI chips "are not competitive right now," largely because Western export controls work — China cannot buy ASML's advanced lithography machines or the best US chip-making equipment. Huawei's much-hyped announcements involve techniques that aren't unknown; it's just forced to attempt them years early "because they have no choice."
The flip side of his memorable line — "the US has chips but no power, China has power but no chips" — is that China's constraint is solvable only by cracking chipmaking, while America's constraint (electricity) is at least a known engineering problem. If China ever gets the chips, it can power them instantly with coal plants.
Summary & timestamps derived from the public YouTube video (transcript in transcript.txt) for personal study. Not investment advice. © The Real Eisman Playbook / Steve Eisman for source material.