Speaker · Updated
Stacy Rasgon: latest market views
Bernstein Research
Summary
Stacy Rasgon is Managing Director and Senior Analyst covering U.S. semiconductors and semiconductor capital equipment at Bernstein Research, a top-ranked chip analyst who joined the firm in 2008 after working at McKinsey & Company.
Chorus has 24 views from Stacy Rasgon across 2 episodes on CNBC Closing Bell and Monetary Matters since October 8, 2026. Their latest calls: bullish on NVIDIA, Advanced Micro Devices and Broadcom; bearish on Intel.
Where they stand
Their latest bullish or bearish call on each company.
- IntelBearish ·
- NVIDIABullish ·
- Advanced Micro DevicesBullish ·
- BroadcomBullish ·
What they talk about most
The latest views
Newest first. Every quote is checked against the episode's transcript.
IntelBearish
Intel's Q4 guide is extremely weak, driven by PC client collapse; the drop is warranted.
The quarter itself was fine, with revenue and margins above expectations, but the guide of $15.5B versus $17B consensus is extremely weak. He has long believed CPUs were over-shipping relative to PCs, and Q3 PC numbers were horrendous, likely the worst in at least twenty years, implying a sizable channel flush into Q4. To reconcile the guide, client revenue must fall at least 20% sequentially; even then data center would need to rise only about 10%, which he doubts is enough. Gross margin guidance of 38% sits about 400 basis points below the Street, and pricing upside does not appear to be enough to help. He says he is not surprised by the move.
“I suspect that's where the weakness is. The quarter actually was fine... The guide is extremely weak though.”at 34:53
CNBC Closing Bell: Closing Bell Overtime: Big Tech Faces Another Earnings Test 10/29/26 ·
IntelUncertain
Intel is very expensive on current earnings; buyers are pricing a speculative multi-year foundry turnaround.
Rasgon says Intel is a different company than when the government took its stake: the balance sheet is in better shape and the server market is helping. He still says valuations have risen and the stock is very expensive on earnings. He argues buyers are not paying for next year's earnings but for a structural change over about five years as Intel ramps its third-party foundry business. He thinks the market has much higher earnings power priced in for a few years out, but calls that speculative and up in the air.
“The valuations have gone up. It's very expensive on earnings. People are buying it now. We're not buying it for like the earnings over the next year. They're buying it for like a structural change over the next like five years as they ramp you know third party foundry business.”at 37:16
CNBC Closing Bell: Closing Bell Overtime: Big Tech Faces Another Earnings Test 10/29/26 ·
PC shipments are weak because of a memory shortage, not falling demand; weakness likely lasts into 1H next year.
Rasgon has been nervous about PCs and smartphones and flagged PCs as the biggest risk in his Intel preview. Third-party data show Q3 PC shipments down over 20% year over year and down double digits sequentially, where they usually rise double digits. He says this is the worst sequential quarter for PC shipments in about 20 years of data. He attributes it to PC makers being unable to build machines without memory, plus an inventory flush and PC CPUs having over-shipped for several quarters. He expects smartphones and broader consumer electronics to see the same pattern. Intel itself said the PC dynamics would last at least through the first half of next year, and Rasgon says the memory situation may not resolve for a while.
“And the reason PCs are so weak, it's not because PC demand is necessary, it's because there's no memory. And you can't build the PCs.”at 38:22
CNBC Closing Bell: Closing Bell Overtime: Big Tech Faces Another Earnings Test 10/29/26 ·
NVIDIABullish
Nvidia-type AI leaders are very cheap, if the AI bullish case plays out.
Asked for relative value after the group's pullback, Rasgon says he likes the Nvidias of the world because they are very cheap. He conditions this on the AI bullishness: if the AI bull case does not work out, these companies are the ones driving the hype and would be hurt. The support is thin and the view is brief.
“We'd like, you know, the NVIDIAs of the world. They are very cheap.”at 39:44
CNBC Closing Bell: Closing Bell Overtime: Big Tech Faces Another Earnings Test 10/29/26 ·
Semiconductor equipment (semi cap) is a good way to play the memory shortage and capacity buildout.
Rasgon says there is a memory shortage and chipmakers are adding capacity, but building it takes time. The tool vendors that supply that buildout are at the front of the trade. He cites KLA, which reported the previous night, as evidence the group is already seeing this. He says the firm likes semi cap as another way to play the memory shortage, alongside Nvidia-type AI names.
“And frankly, all of these shortages, there's no memory. I mean, they are adding capacity now to make more of it, but it does take time to add that capacity... We like semi cap as another way to play this.”at 39:44
CNBC Closing Bell: Closing Bell Overtime: Big Tech Faces Another Earnings Test 10/29/26 ·
Semiconductor fundamentals are still improving and the sector is not yet near a cycle peak.
Sector earnings are up more than 100% while the SOX index is up roughly 80% year to date, so multiples have compressed. Forward earnings estimates have kept rising even since the June peak, so the stock pullback is driven by sentiment and fear of a peak rather than falling fundamentals. Nvidia and Broadcom say they can grow 70% to 100%+ next year, memory makers expect tighter supply in 2027 and 2028 than in 2026, and hyperscalers say nobody has enough compute. Rasgon accepts that a peak will eventually come, but he does not see one in 2026 or 2027, and he suspects there is still room to run from this base.
“fundamentals seem better or not worse. I'm not really worried about a cycle peak yet. ... It's not this year. It's not that next year.”at 3:16
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
AI infrastructure & data centers
The evidence points to AI infrastructure earning a return: neoclouds' payback is about 1 to 1.5 years.
The main bear case is that hyperscalers are spending heavily without a return. Rasgon says the incremental data points are landing on the side of a return. Neoclouds renting out GPU capacity at high prices imply payback on the infrastructure of about one to one and a half years. Some foundation-model companies that rent out their compute have seen revenue grow very quickly. Meta's Muse agent is an early example of a use case a normal consumer could use. He admits it is still early and that the question of whether customers will pay is the key test for every AI bear case.
“the incremental data points around is there a return or not on these investments, I think is landing pretty much decisively on the side of yes, there is a return.”at 8:27
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
AI chip demand is real: capacity being built is used immediately, with no inventory stockpiling.
Rasgon argues that AI demand is genuine, not phantom. He contrasts it with the COVID period, when customers stockpiled chips and inventories swelled across the supply chain. He says nobody is stockpiling AI hardware, and everything that is built and deployed is used straight away. Order visibility is strong and improving in every part of the industry he covers, and the roadshow with 11 semiconductor companies was uniformly positive. Rasgon acknowledges that the companies themselves only see the orders in front of them.
“Nobody's really stockpiling AI hardware. Everything that's actually getting built and deployed is being used immediately.”at 12:24
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
AI infrastructure & data centers
Data center backlash is a real midterm political risk, but it will not stop the buildout.
Local backlash against data centers exists in some places and is bipartisan, and it is becoming a political talking point ahead of the midterms. Rasgon says this is a concern. He expects progress will not be stopped because there is enough money and power behind the buildout. The US has a lot of land, and if US sites are blocked, data centers will simply be built elsewhere, and space-based data centers could sidestep local opposition altogether.
“it does make a really good political talking point... We are going into the midterms. So, that is a concern, although my guess is progress will not be stopped.”at 12:59
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
AI infrastructure & data centers
Power and build constraints are timing issues on individual projects, not a broad brake on data center spending.
Adding tens or hundreds of gigawatts a year is more than the electrical grid can support, so companies rely on behind-the-meter, on-site generation, and some capacity ships before it can go online because power is not ready. Rasgon says this is not broad-based. He treats it as timing noise on specific builds that does not slow overall spending. He points to Nvidia's growth guidance, which is capped by the land, power and shell it expects to be built, and says Nvidia would raise numbers if more capacity came online.
“But I don't think it's a broad based kind of thing. It's more timing issues on some of the builds. And I don't think it's slowing anything down.”at 15:00
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
NVIDIABullish
Nvidia trades cheaply on likely earnings if demand holds; the stock is only expensive if a peak arrives.
Nvidia has said it can grow revenue at least 70% next year and has demand for 100%. At a share price of about $240, 70% growth implies about $15 to $16 of earnings, and 85% growth implies $17 to $18, so the stock trades at roughly 11 to 12 times if revenue doubles. Rasgon argues that if the AI cycle is not peaking, the stocks are absurdly cheap. The market is pricing in peak fears instead, and he says the buy side expects more than the company's own guidance. Nvidia's guidance is set at the level it believes land, power and shell capacity will support, so it has line of sight to that growth.
“to pick on Nvidia, if they grow revenue 70%, they'll probably do $15 or $16 in earnings next year.”at 16:43
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
BroadcomBullish
Broadcom's shares look cheap if its doubling-revenue path holds; the stock is only expensive at a peak.
Broadcom has suggested that it can reach over $30 in earnings by 2028. Rasgon says that if revenue doubles next year and doubles again, that earnings figure is achievable. At a share price of about $363, that would put the stock at roughly 11 to 12 times earnings. He frames this as the same valuation question as Nvidia: if the cycle is not peaking the stock is cheap, and if it is, the market is right to be cautious.
“Broadcom suggested by '28, they will do over $30 in earnings, which if you think that they can double the revenues next year and double them again, yeah, you'll be over $30 in earnings.”at 17:12
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
Market sentiment, positioning & volatility
The AI chip rotation into bottleneck stocks, and AI megacap selling to fund it, must normalize.
This year investors rotated from accelerator leaders such as Nvidia and Broadcom into bottleneck stocks in memory, semi-cap, networking, optics, power semis and CPUs. Those bottleneck stocks ripped while the AI leaders lagged, and Rasgon says one of the two groups must be wrong, so the gap has to close one way or the other. He says AI megacaps are seen as safe and their shares are sold as a stable source of funds for buying higher-torque names. Nvidia's size, about $5 trillion or 8% to 9% of the S&P 500, also means some clients have position limits that prevent them buying more, which is a technical factor unrelated to fundamentals. Nvidia has started to behave better recently and Broadcom less so, and market-share concerns are weighing on Broadcom.
“it felt to me like it has to normalize at some point one way or the other.”at 21:41
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
Memory supply stays very tight into 2027-2028 as HBM absorbs wafer capacity and new fabs take years.
HBM is made by stacking DRAM dies, and yield losses from stacking mean it takes three to four times as many wafers per gigabyte as standard DRAM. That cuts effective capacity by two-thirds to three-quarters, which is new this cycle. On top of strong AI demand, this keeps supply very tight. NAND is also tight because AI workloads store large model weights and generated content, and new factories take years to come online. Rasgon expects material new capacity only in 2027 and 2028, and says a rack of Nvidia's GPUs is about 80% memory by silicon wafer area. Memory stocks trade at single-digit P/Es on fears of peak pricing, but he sees the tightness continuing.
“To make a gigabyte of high bandwidth memory DRAM takes three or four times as many wafers. ... It'll be '27, '28 before you start to see material amounts of capacity start to come on.”at 25:08
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
Semiconductor equipment spending will exceed $250B by 2028, and more if cleanroom capacity is not a constraint.
Rasgon works backwards from data center gigawatt forecasts to racks, then to silicon content per rack, then to the capacity fabs must add, and finally to equipment spending. That gives WFE above $150B in 2026, above $200B in 2027 and above $250B in 2028. He thinks the end-of-decade figure should be $300B or more. The binding constraint is cleanroom shortage: fabs need shells and cleanrooms before they install tools, so spending would be higher still, above $300B next year, if cleanrooms and semi-cap capacity were unlimited. He is quite bullish on the semi-cap group (Applied Materials, Lam Research, KLA), which he says are highly correlated and can be owned as a basket.
“We will probably wind up spending $150,000,000,000 plus on WFE, on equipment this year in 2026. We are over $200,000,000,000 next year and I think over $250,000,000,000 in 2028.”at 30:14
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
NAND faces more competitors than DRAM, so its margins are weaker.
Rasgon says NAND has more players than DRAM. DRAM has four: the big three plus CXMT in China. NAND has six to eight, including Solidigm, Kioxia and YMTC. He says the competitive dynamics in NAND are sometimes worse and that this shows up in the margin profiles of some NAND makers. He also notes that DRAM costs more per bit, while NAND is cheaper per bit because of density.
“So the competitive dynamics in NAND sometimes are a little worse. And we've seen that in terms of some of the margin profiles in some of the names.”at 37:00
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
Chip equipment spending will keep rising into the decade, constrained by clean-room capacity.
Rasgon's published wafer fab equipment (WFE) forecast is about $154B this year, about $200B next year and a little over $250B in 2028. His covered companies, including Lam Research and Applied Materials, are modeled consistently with that. He thinks WFE will need to rise further into the end of the decade and possibly earlier, since demand looks strong. The binding constraint is clean-room space, because a tool can't ship until there is a place to put it. Micron's capex is rising, with facility spend growing faster than equipment spend next fiscal year because clean rooms must be built first.
“I suspect we will need higher WFE into the end of the decade. I suspect demand is high enough to do it earlier. I just don't know if the clean rooms are available to enable it.”at 38:24
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
NVIDIABullish
Nvidia's moat holds; competition is the wrong question while the market stays this large.
Rasgon says the competition narrative moves semi stocks but is the wrong fundamental question. What matters is whether the market is big enough, and it is. He expects Nvidia to do close to $700B next year, 70%+ growth, and says it could beat that. He argues Nvidia can absorb share loss because it is large, while rivals like AMD and Qualcomm gain meaningfully from small share shifts. He cites Nvidia's moat: CUDA software, a broad mix of GPUs, LPUs, CPUs and networking, and a balance sheet used to take equity stakes and backstop ecosystem infrastructure.
“If it's big, I think everybody's fine. And if it's not, I think everybody's screwed.”at 41:08
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
BroadcomBullish
Broadcom is a favored AI name; custom-chip competition doesn't threaten the overall AI market.
Rasgon lists Broadcom among the favored traditional AI names. He cites Broadcom's guidance for next year's AI revenue, which he says could be exceeded. Broadcom is among the custom-chip (ASIC) makers he sees as competitors to Nvidia, but he argues that competition matters less than whether the total market is big. The AI market is roughly a trillion dollars and still growing, so custom chips don't threaten the story.
“Broadcom said they'll do [$115B] in AI revenue next year, again, could do more.”at 41:29
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
IntelNeutral
Intel is better than in years, but its turnaround will be a slog and foundry revenue is years away.
Rasgon has been negative on Intel for most of his career but says he no longer hates it, which he calls new. He credits luck: strong server demand lets Intel sell server products that it admits are not competitive, and customers bought products Intel had written down to zero. He says the narrative is turning its way on tight supply and its foundry business, but he doesn't expect material foundry revenue for quite a few years. Packaging revenue will help, and its IP is decent. He praises CEO Lip Bu Tan for resetting the cost structure and for industry relationships, while expecting a slog.
“I don't hate it right now, which is new for me.”at 42:44
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
Traditional cycle warnings like high channel inventory are less reliable after years of demand-driven strength.
Rasgon's cycle tracker flags high channel inventory as the biggest bear signal, and inventories have been high. But they have stayed high for four or five years. He argues the industry hasn't had a typical cycle in almost ten years. In 2018 his estimates fell about 40% while the stocks ended higher, because markets anticipate and the trade war backdrop mattered. COVID, the tariff whiplash of Trump 2.0 and now an AI demand cycle of unusual strength have all broken the historical pattern. He doesn't know how representative the old indicators are anymore, and calls high inventories the new normal.
“we haven't really had a typical cycle quote unquote for almost ten years.”at 46:12
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
Chip build-out follows demand and multiples are far below 2000 levels, so no bubble.
Rasgon compares today with 2000. Back then there was dark fiber built for demand that never came, and build-out ran ahead of demand. Today he sees the industry building out to meet demand, with nobody having enough capacity. Valuations are the bigger difference: Cisco traded around 200x forward earnings in 2000, while Nvidia is around 12x. If you are worried about a bubble, he says, it has to be on the multiple side, and he doesn't think multiples are that high.
“We're actually building out with demand versus building out ahead of demand, and then the valuations are much different.”at 49:21
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
Semiconductor fundamentals are improving, so he remains fairly constructive on the sector.
Rasgon acknowledges nervousness after the semi rally and the June sell-off. He says the sector is still up a lot year to date. He believes fundamentals are getting better, not worse, and he remains fairly constructive. He likes AI names, semi cap equipment and AMD, and expects a bumpy path, not a monotonic one.
“I still think fundamentals overall are getting better, not worse.”at 49:49
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
Advanced Micro DevicesBullish
AMD is a favored name; it could grow AI revenue if it ramps, and share shifts matter more to it.
Rasgon says he has liked AMD on the CPU side and that it is where he has spent most of his time. He says AMD could do a large amount of AI revenue next year if it ramps its products. Because AMD is small relative to the total market, he says it doesn't need much share to matter meaningfully, though he doubts market-share shifts are moving much.
“AMD is kind of where I've been spending most of my time at least.”at 50:17
Monetary Matters: Stacy Rasgon: “Demand Is Off The Charts” in Semiconductors… And Set To Double Again Soon ·
Shows
- CNBC Closing Bell · 1 episode
- Monetary Matters · 1 episode
Ask your AI what Stacy Rasgon thinks
Chorus gives Claude, ChatGPT and other assistants every view Stacy Rasgon has shared on the shows it covers, next to everyone else's. Ask where they stand, what changed, and who disagrees.
Chorus reports what speakers said. It isn't investment advice.