Newsletter · · Ashutosh Agarwal
China Flexes and a Trillion Dollars of Chip Value Vanishes - Foundry & Chip Equipment Weekly - Week of August 2, 2026
A synthesis of what investor and operator podcasts said about foundry and chip equipment for the week of July 26 to August 2, 2026, including a bear market in semis, China's state-backed lithography push and CXMT's Shanghai debut, order books running to 2028, and quantum's best week in months.
Foundry & Chip Equipment Weekly
Week of August 2, 2026: China Flexes and a Trillion Dollars of Chip Value Vanishes
Two weeks ago the puzzle was that the best businesses in the chip supply chain reported blockbuster numbers and their stocks went nowhere. Last week the market decided that raising your spending was something to be punished for. This week it stopped being a mood and became a rout.
Between the previous Friday and the middle of this week, the big chip names shed more than a trillion dollars of combined market value. The Philadelphia Semiconductor Index, the main basket of the 30 largest US-listed chip stocks, fell into a bear market, down more than 20% in a month. South Korea, where the memory boom had been most crowded, came apart: its main stock index dropped more than 40% in 40 days. A 25-year-old hedge-fund manager who had ridden the AI trade from $225 million to something like $20 billion got margin-called and had to sell his whole public book to Citadel.
And the spark for a lot of it came from one place: China. This was the week China's home-grown chip-equipment and memory ambitions stopped being a footnote and became the story: a state-backed lithography machine, a memory IPO that went vertical, and a cheap open-weight AI model, all landing at once on a market that was already nervous. Then, on Thursday, the whole thing bounced hard on a good Microsoft print and a great Lam Research one, as if to remind everyone that nothing about the demand had actually changed.
So this week is about a scare, who caused it, and whether it was real. And, in a welcome change of pace, it's also about quantum computing, which for the first time in months had a genuinely good week.
TL;DR
- The chip trade cracked. The semiconductor index fell into a bear market (down more than 20% on the month), roughly a trillion dollars of chip market value evaporated in under a week, South Korea's index fell more than 40% in 40 days, and a heavily leveraged AI-and-chips fund blew up and was bought out by Citadel. By Thursday the group had bounced about 7% off the lows.
- China lit the fuse, three ways at once. A state-backed Chinese company started making its own advanced chip-printing (lithography) machines, the kind ASML dominates; Chinese memory maker CXMT went public in Shanghai and surged roughly 450 to 500% on day one to a market value north of $450 billion; and a cheap Chinese open-weight AI model reopened the fear that AI itself is being commoditized. ASML fell around 6% in a day (one host cited 17% over the stretch); memory names got hit worldwide.
- But the operators and the earnings said demand is fine. A prominent tech investor argued SK Hynix has a 65% share of high-bandwidth memory at 80% gross margins, is booked out on long-term contracts, and that Chinese commodity-memory makers are "three, four, five years" from even entering that market. Samsung affirmed its backlog is full out to 2028. Lam Research beat and jumped 18%; Seagate beat with record free cash flow. Only KLA stumbled: it beat but guided softly and fell 8%.
- The fabless squeeze is turning political. There's a reported price fight between Apple and Micron over memory, and Apple has reportedly gone to the White House to ask that US companies be allowed to buy cheaper Chinese memory chips, a request that runs straight into national-security politics.
- Quantum finally had a good week. D-Wave's expanded partnership with AT&T, cutting a network task from over an hour to under 15 seconds, lifted the whole group, with IonQ and Rigetti up 3 to 4%. IBM's CEO bought a quantum lab from Boeing and GM and raised his forecast to a "trillion-dollar" market by the end of the 2030s. And a look inside the US quantum supply-chain build gave the clearest read yet on where the picks-and-shovels money goes: cryogenics, optics, lasers, vacuum systems, and precision components.
What's new
The meltdown: what actually happened
Start with the scoreboard, because the numbers are the story. On All-In (July 31), the hosts laid it out plainly: the Philadelphia Semiconductor Index, 30 of the biggest US-listed chip names, Nvidia and TSMC and AMD and Micron among them, was "down over 20% over the last month. That's bear market territory." Between the prior Friday and Wednesday, "leading chip companies shed over a trillion dollars in market cap combined." The two big Korean memory makers, which aren't even in that index, fared worse: "Samsung down 38% over last month. SK Hynix down 14% since going public three weeks ago. The Kospi... is down over 40% in the last 40 days."
The most vivid casualty was a person. Leopold Aschenbrenner, a 25-year-old who left OpenAI to start a fund he named after his own AI manifesto, "Situational Awareness," had reportedly compounded $225 million in 2024 into roughly $20 billion, running it up as high as $45 billion, on leverage. This week he got margin-called and, per the reporting the hosts cited, sold his entire public portfolio, and Citadel bought the book. Chamath's lesson was the oldest one in the business: "You have to manage leverage incredibly carefully. Because when it runs ahead of you, the unwind is incredibly violent, and it's incredibly quick." Running around three-and-a-half times levered, he noted, "a 25% move is amplified 75%." Jason's coda: leverage "is the only way that smart people go broke... if you're not using leverage, your portfolio would just be down 30% this month, and then it would already be up 7% today."
That last point matters. As the hosts taped, the index had already bounced 7%. Which raises the only question worth asking about a sell-off this size: was it about the businesses, or about the money?
China is the reason, and the read-through cuts both ways
The trigger was China, and it came from three directions at once. The Rundown (July 28) walked through it cleanly. First, lithography: "According to the information, a state-backed Chinese company has begun producing their own immersion DUV lithography machines," the machines that print circuit patterns onto silicon wafers, a market "the Dutch company ASML dominates" and from whose most advanced tools China is banned. "ASML stock fell nearly 6% yesterday, and it's down another 5% this morning." Lam Research and SanDisk got hit too.
Host Zaid Admani was careful to keep the scare in proportion, and this is the crucial detail for anyone tempted to panic-sell the equipment names: "this Chinese company is expected to ship about five machines this year and about 20 machines next year. ASML, on the other hand, shipped 131 of these systems last year alone. Not to mention, these Chinese machines still aren't as good as ASML when it comes to performance and build quality." His sharper point was strategic, not near-term: "every time the U.S. government tightens their export controls, China gets even more motivated to make their supply chain self-sufficient... they're also forcing China to build the exact capabilities the U.S. doesn't want them to have."
Second, memory: Chinese DRAM maker CXMT went public in Shanghai on Monday, and "the stock surged over 450% on its first day of trading," giving it "a market cap of nearly half a trillion dollars, making it the second most valuable company in China behind Tencent." That's what sent SK Hynix and Samsung each down 13% in a day on The Rundown's telling, and it's what put South Korea's index down almost 11% in a single session. Third, the cheap Chinese open-weight model, Kimi K3 from Moonshot AI, reopened the debate about "how quickly China is closing the gap with the U.S. when it comes to AI."
On All-In, Friedberg took the China angle to its most uncomfortable conclusion for the whole complex, and this is the genuine bear case, not a valuation quibble. China, he argued, "may deflate the value of models by releasing open source AI models... the value may just sit with the compute infrastructure and the compute layer." If a big chunk of AI's projected value creation gets "deleted" or captured by China, "it really puts into question the 30-year timeline for the United States economy." He also confirmed the lithography scare, naming a Chinese company (phonetically "Aishengna") that "started mass producing lithography machines" and pegging ASML "down 17% on news that China is getting into that business."
Two honest flags on this thread. One: the near-term math (5 to 20 Chinese machines versus ASML's 131 last year) says this is a multi-year worry, not a next-quarter one. Two: China's domestic toolmakers, SMEE, Naura, AMEC, Piotech, went from an afterthought to the whole conversation, if mostly by implication. That's the single biggest change in the narrative in over a month.
The earnings quietly said the opposite of the tape
Here's the tension. While the market was pricing a China-driven demise, the companies reporting into the panic mostly beat.
On Closing Bell (July 28), CNBC's Christina Partsinevelos ran the numbers as they crossed. KLA, the "semi-cap equipment" company that does process control, "considered an AI CapEx play," "posted an EPS of $1.05, five cent beat with revenues of $3.66 billion. So a small revenue beat." The problem was the guide: next-quarter EPS "at $1.16 at the midpoint, that's only two cents higher than what the street anticipated," and the $4 billion revenue guide was exactly what analysts had said "needs to be higher... for the stock to really move." Result: "Shares down about 8%... already almost 40% off the highs." In a market this jumpy, a beat with an in-line guide is a sell.
But right alongside it, Seagate told a different story: "$4.1 billion revenue, that was higher than the street wanted on EPS of $7.30. So much higher than the 5.80... a record free cash flow of $3.1 billion. And... they see momentum continuing into 2027." The stock climbed after hours, "something we haven't seen from a lot of the chip names as of late."
Then Thursday flipped the mood entirely. On Stock Market Today with IBD (July 30), the hosts described "one of the stronger first day of a rally attempts for the NASDAQ that I've seen in a little while," fueled by good reports "from Microsoft and also from Lam Research." The Nasdaq 100 rose 3.4% and reclaimed 25,000; Lam Research finished up 18%. Their sober caveat, worth keeping: it was "probably a day of short covering," and "not much repair work was done today in terms of the technical damage," a bounce, not yet a bottom confirmed.
The most useful frame on why the equipment names get bought on days like this came from Best Stocks Now (July 30). Bill Gunderson's "two sides of the street" metaphor is exactly the WFE thesis in plain English: on one side sit the spenders, Meta and Microsoft, "borrowing money to keep up with the AI race"; on the other sit "the chip stocks... the biggest recipient of those funds," who "need equipment. A lot of it." Lam Research, he noted, is up 157% over 12 months even after falling 33% in the month, and, crucially, he relayed that "Samsung affirming... it's clear out to 2028 before we'll be able to clear the backlog and to catch up. And in the meantime, we're going to need more Lam Research equipment." A full order book to 2028 is not what a demand cliff looks like.
The memory debate, and why it isn't cyclical anymore
The single best piece of operator-adjacent analysis this week came on Squawk Pod (July 29), from Steve Jang, founder of Kindred Ventures, an early investor in Coinbase, Uber and Perplexity, so a venture capitalist rather than a chip operator, but one who made the clearest bull case for memory heard in a while.
His starting point: the sell-off is about mismanaged expectations, not broken fundamentals. SK Hynix, he said, "have fundamentally very strong numbers as of the print late last night. They are seeing 80% gross margins on memory chips, which is never before seen." The key is that not all memory is equal: SK Hynix leads in high-bandwidth memory (HBM), the fast, stacked memory that sits next to AI chips, with "65% market share," while commodity DRAM from Samsung and Micron is "one step below that in tier." And demand, on his read, is only starting: HBM will "10X from here just over the next three, four years," pulled not just by model training and inference but by "long context memory that is required for agents" and, longer out, by physical AI: "Waymo, Tesla FSD... they all need HBMs onboard on the car."
Then the moat, which is the direct answer to the CXMT scare: "They make commodity DRAM. They are unable to make the stack-to-die structure and architecture of an HBM... it'll be many years before there's a fab for memory chips that will be able to compete effectively... we're talking about three, four, five years to even be able to enter into the market." His bigger claim is the one that would re-rate the whole group if it holds: "fundamentally, memory chips are out of the cyclical state that they were over the last 15, 20 years... it's sitting side-by-side with GPUs and logic processors." If memory really has stopped being a boom-bust commodity, every dollar the market just took out of Samsung, SK Hynix and Micron on China fears was priced on the old playbook.
The fabless squeeze goes to Washington
One thread turned unexpectedly political. On Market Mondays (July 28), the hosts flagged a reported fight between Apple and Micron over memory pricing. The role reversal is the fun part: Apple, "the bully for 10 years" that dictated terms to suppliers, is now on the receiving end of price hikes big enough that "they have to raise our prices on our next age of devices." And the response, per the report they cited, was striking: Apple officials "have gone to the White House to say, look, bring down the price," and asked "the president's team to allow American companies to buy Chinese memory chips."
That request lands on a live wire, and the hosts knew it: letting Apple buy cheaper Chinese memory would come "at the cost of an American company like Micron," which one host bluntly called "financial treason." Their bottom line on Micron was unbowed, a $1,600 target held, on the logic that "the demand hasn't gone anywhere and they're sold out to 2026 and potentially sold out to 2027." But the episode is a useful preview of who ultimately pays for the memory shortage: the device makers, and eventually their customers.
Quantum's best week in months
Quantum delivered real, investable news this week, and, for once, from operators.
The catalyst was D-Wave. On Squawk on the Street (July 27), CEO Alan Baratz, in a first-on-CNBC interview, having just rung the Nasdaq opening bell after changing the company's listing, described an expanded partnership with AT&T that is one of the first real-world quantum use cases with a household name attached. It began with a network-outage-detection problem: AT&T "were able to perform that computation in less than 15 seconds leveraging our annealing quantum computers when otherwise it was taking them over an hour... classically." Now it's expanding to "technician routing to network planning," and, most interesting, combining D-Wave's quantum systems "with their AI agentic solutions." Baratz described a three-part revenue model, "professional services and quantum compute as a service in addition to system sales," and rattled off a customer list that is starting to look like a business: "BASF... NTT Docomo... UHG Optum," and AT&T. On the sell-off in quantum names, he was measured: "I think it has a lot more to do with just the general market dynamics... than anything else."
The read-through was immediate. On The Rundown (July 27), Admani noted the AT&T news "is lifting the entire quantum sector, including [IonQ] and Rigetti, which are also up around 3% to 4%," and made the point that matters: "this is the first time the quantum computing sector has gotten some buzz in the last few months." A single blue-chip customer actually using the technology did more for the group than any roadmap slide.
The heavyweight operator voice was IBM's Arvind Krishna, on two shows. On Everyone Talks To Liz Claman (July 25), he confirmed IBM had acquired HRL Laboratories, the quantum lab jointly owned by Boeing and General Motors, for its "strength in quantum materials, in quantum sensing, and in another semiconductor-based quantum qubit technology called spin electronics." Buying more approaches, he explained, "only then increases the chances that we are one of the winners." And he raised his forecast on the record: "quantum, by the end of the next decade, is likely going to be a trillion-dollar impact on the industry." He also noted the Trump administration has invested "about a billion dollars" in IBM's quantum efforts. He repeated the upgraded number to the WSJ Tech News Briefing (July 28): "I have actually upped my estimate. I think quantum will be a trillion dollar value adder by the end of the 30s." Context worth keeping: IBM's stock had fallen about 25% on a lowered outlook earlier in July, so Krishna is talking a bruised stock's book, but the HRL deal and the raised forecast are concrete.
The debate
The bull case, this week, came from the companies and the order books. Samsung says its backlog runs full to 2028. Lam Research and Seagate beat and, on the good day, were rewarded. A serious investor laid out why memory's 80% margins and 65% HBM share may be durable, not cyclical, and why the feared Chinese competitor is years from the table. And the clearest AI-capex bull on the podcasts, All-In's David Sacks, argued the whole sell-off "is driven by momentum," not fundamentals: the hyperscalers "have invested pretty much all of their free cash flow and then some in this boom," and "eventually there will be a return on that investment... this is sort of temporary market volatility amplified by leverage."
The bear case, this week, was voiced loudly and, for the first time in a while, it was fundamental, not just valuation. It has three legs, all Chinese. Homegrown lithography threatens the ASML monopoly (even if only 5 to 20 machines, versus 131, this year). CXMT's blockbuster IPO signals a state-backed memory challenger with half a trillion dollars of capital markets behind it. And cheap open-weight Chinese models threaten to commoditize the AI layer itself: Friedberg's warning that value could shift away from models toward "compute infrastructure" and energy, and that a piece of America's projected AI windfall could simply be "deleted." Layer on the plumbing: treasuries near 5%, a market pricing a real chance of a rate hike, and a leverage unwind that mechanically forced selling regardless of any thesis.
Note the shape of it honestly. The bull rebuttal to the China scare came from investors, earnings prints, and one VC's memory analysis, persuasive, but not from an incumbent foundry or tool-company executive; ASML appeared this week mainly as a falling share price.
Read-throughs
Wafer-fab equipment is where the China fear hits first and the demand story holds longest. The names that fell hardest on the lithography headline, ASML, Lam Research, KLA, are the same ones whose order books (Samsung to 2028; Seagate's momentum into 2027) argue the demand is intact. KLA's 8% drop was about a guide that was merely in-line, not about lost business; Lam's 18% pop the next day, on its own beat, is the counter-tell. The China lithography threat is real for the 2028-and-beyond debate and largely noise for the next few quarters, given 5 to 20 Chinese machines against ASML's 131.
High-bandwidth memory and advanced packaging remain the tightest link in the chain. The bull argument rests entirely on HBM being hard: stack-to-die architecture that commodity DRAM makers, including China's CXMT, reportedly can't replicate for years. If that moat is as deep as claimed, the packaging and test tool-chain that HBM depends on, the metrology and advanced-packaging names this newsletter tracks, sit on the durable side of the trade, even as the stocks trade on the scare.
Fabless customers are the pressure valve. The Apple and Micron fight, and Apple's reported appeal to Washington to buy cheaper Chinese memory, is the clearest sign yet that the memory shortage is flowing downstream into device prices, and into politics. Watch whether any US allowance for Chinese memory purchases materializes; it would be a genuine negative for Micron and a signal that the shortage has become a policy problem, not just a pricing one.
China-domestic toolmakers are now central to the equipment debate. SMEE, Naura, AMEC and Piotech are the implicit subject of the whole argument. Nothing this week changes the near-term picture, the machines are few and reportedly inferior, but the market has clearly decided to start pricing the long-tail risk that China builds its own WFE stack. This is the thread to watch most closely from here.
Quantum hardware is quietly becoming a supply-chain story, which is where the durable money is. The D-Wave and AT&T deal and IBM's HRL acquisition are the headlines, but the more investable idea came from The New Quantum Era (July 27), where the Chicago Quantum Exchange's Kate Timmerman described the push to build a domestic quantum supply chain, anchored by a newly funded regional tech hub with $55 million in federal money. Her point is the frontier read-through in a nutshell: the qubit companies grab the headlines, but the value pools in the components, "nano positioners, optics, photonics, vacuum systems," lasers, cooling systems, connectors, much of it today "single sourced" from outside the US, and much of it made by small and mid-sized manufacturers that could pivot in. She cited an economic-impact study putting the number near "$80 billion" with "70% of jobs created not requiring a master's degree." Whoever wins the qubit race, the cryogenics, control electronics, optics and precision-machining suppliers get paid, the same picks-and-shovels logic that makes the chip-equipment names the quiet winners of the AI build-out.
What changed from prior weeks
The mood completely inverted. Two weeks ago the crown jewels reported blowouts and shares went nowhere; last week Alphabet and Tesla were punished for raising capex; this week that anxiety detonated into a full bear market for chips, a trillion dollars of lost value, a South Korean crash, and a marquee fund blowing up. Then it bounced. The whipsaw itself is the lesson: this is a leverage-and-momentum unwind sitting on top of a demand story that, by the companies' own numbers, hasn't changed.
The bigger narrative shift is China moving from the margins to the center. Via a state-backed lithography maker and CXMT's IPO, China was the proximate cause of a global sell-off. The bear case on the whole complex is now a China-supply-chain case, not just a valuation case, and that is a meaningfully different and more serious debate.
Quantum also changed, for the better. After a run of roadmap-only weeks, it produced a real commercial deal (D-Wave and AT&T), a real acquisition (IBM and HRL), and a concrete supply-chain build, the first genuinely actionable quantum week in a while.