Newsletter · 2026-07-19 · Ashutosh Agarwal

TSMC and ASML Blow Out, and Shares Fall Anyway - Foundry & Chip Equipment Weekly - Week of July 19, 2026

Foundry and chip-equipment podcast intelligence for the week of July 13 to 19, 2026. TSMC and ASML both reported blowout quarters with order books running to 2029 and 2030, yet both stocks fell, and the debate turned on valuation rather than fundamentals.

Foundry & Chip Equipment Weekly

Week of July 19, 2026: TSMC and ASML Blow Out, and Shares Fall Anyway

Foundry & Chip Equipment Weekly, July 19, 2026.

Here's the puzzle that defined the week. The two most important companies in the entire chip supply chain, the one that builds the machines, and the one that builds the chips, both reported earnings that were, by almost any measure, spectacular. Records. Raised forecasts. Order books stretching years into the future. And both stocks went down.

That's not a contradiction. It's a signal. After a run this big, "great" isn't good enough anymore, the price already assumed great. This week the market started asking whether the last leg up was the crowd, not the fundamentals. The businesses look better than ever. The stocks are having a harder time.

Let's get into it.

TL;DR

TSMC posted a monster quarter, revenue around $40 billion (up 34% from a year ago) and profit up 77% to roughly $22 billion, its fifth straight record. It raised full-year growth guidance above 40% and lifted this year's capital-spending budget to as much as $64 billion, with the CFO saying the next three years of spending will be "significantly higher" than the last three. It also added $100 billion to its Arizona plan, bringing the US total to $265 billion. The stock fell about 4%.
ASML, the only company that makes the extreme-ultraviolet (EUV) machines needed for the most advanced chips, raised its full-year sales forecast for the second time this year to €43-45 billion, lifted margins, and said every EUV order it needs for 2027 is already booked with 2028 pre-orders in hand. The stock barely moved.
The common thread from both management teams: demand now runs into 2029-2030, and they're finally expanding capacity aggressively, but carefully.
The selloff in chips was real but was framed on the podcasts as a "parabolic unwind", hot money leaving crowded trades, not a change in the underlying story. A quirk in South Korea's leveraged-ETF rules got some blame for draining buying power.
Intel's former CEO gave a candid post-mortem on how a great American company lost its way, and a genuinely chilling stat about Taiwan's fragility.
On the frontier: a fault-tolerant quantum computer is now being promised on Amazon's cloud in 2028, and the math behind it has quietly gotten far less crazy.

What's new

The earnings that mattered: two blowouts, two shrugs

Start with TSMC, the Taiwanese giant that manufactures the majority of the world's most advanced chips for Nvidia, Apple, AMD, and Broadcom. On The Rundown (July 16), host Zaid Admani of Public.com walked through "another blowout quarter": revenue of roughly $40 billion, up 34% year over year, and net income jumping 77% to almost $22 billion, the fifth straight quarter of record profits. Management raised full-year revenue growth guidance to slightly above 40% and lifted the capital-expenditure budget to as much as $64 billion.

That capex number is the tell. As Admani put it, TSMC "has historically been very conservative when it comes to expanding capacity." Their spending sat around $30 billion a year from 2021 to 2024, through the first years of the AI boom, before rising to $40 billion in 2025. Now it's headed to $60 billion-plus, and the CFO said spending over the next three years will be "significantly higher than the past three years." In plain terms: the company that is famously cautious about building too much is now building a lot. You don't do that unless you're convinced the demand is durable.

TSMC also announced another $100 billion for Arizona, bringing its total US commitment to $265 billion, which it calls the largest foreign direct investment in US history, enough for four more chip plants. There's a catch, and it's the one number that wasn't pretty: TSMC warned that its overseas factories (which cost more to run than Taiwanese ones) could shave several percentage points off gross margins over time.

So why did the stock drop about 4% on a blowout? Two reasons came through clearly. First, expectations were sky-high, the stock was already up more than 70% over the prior year, and, as Admani said, "the market is kind of numb to it." Second, and more subtly, on Bloomberg Tech (July 16), Tammy Chu, Berenberg's head of tech equity research, made the counterintuitive point that raising spending over multiple years can spook investors in the near term even as it thrills them long term, because bigger capex raises the bar for everyone to justify current valuations. Her bottom line, though, was constructive: TSMC "called today that the industry is new and demand will go all the way into 2029, 2030," and the pullback is "just short-term profit taking instead of anything fundamentally went wrong."

Then ASML, the Dutch company that makes the lithography machines chipmakers can't live without, and the only maker of the most advanced EUV machines (which run up to €400 million each and can only be built a few dozen a year). On The Rundown (July 15), Admani called it "a blowout quarter": Q2 revenue of €9.3 billion (up 21% from a year ago), profit of €2.9 billion, and, the part that stood out, guidance raised to €43-45 billion for the year, up from a previous €36-40 billion, with gross-margin guidance lifted to as high as 56%. The killer detail: ASML says it has received all the EUV orders it needs for 2027, and some customers are already lining up for 2028.

Why does that matter more than almost any other earnings report? Admani nailed it: "Companies like Nvidia or AMD can tell you that demand is strong today, but ASML is taking orders for machines that chipmakers will need two years from now. So when ASML says that customers are accelerating their plans, that tells you that these companies are still betting the AI demand will be there in 2027 and 2028."

The Canadian Investor (July 16) added texture that's easy to miss. ASML sold 86 lithography systems in the quarter (up 28% from Q1), but EUV units were flat at 16, and for the first time, its system sales split roughly 50/50 between memory and logic, versus prior years when logic (the processors from the likes of Nvidia) was often double memory. That shift is the AI memory boom showing up in the machine orders: memory makers, historically reluctant to over-build, are finally placing orders because they now have take-or-pay contracts locked in with the hyperscalers years out. By region, South Korea took 43-45% of shipments over the last two quarters, a direct read on how much of the world's new capacity is memory. And ASML plans to add 30% to its low-NA EUV capacity for 2027, and is investigating another 30% for 2028. As the hosts wryly noted, ASML said the strong demand gives it "increased flexibility for future pricing adjustments", "a nice way of saying they'll be increasing prices," which is what happens "when you have a monopoly."

The stock? Flat. And that flatness became its own story.

Why the market yawned: "priced to perfection"

On The Joseph Carlson Show (July 15), Carlson, who owns ASML, captured the mood better than anyone. "ASML just released one of the best earnings reports that I've ever seen," he said. "They didn't just beat and raise guidance. They raised guidance substantially on a multi-year basis." And the stock was up half a percent. "All of that, everything ASML's doing, couldn't get this stock to budge."

His explanation is the phrase of the week: price to perfection. ASML is, "practically speaking, a perfect company", an indisputable moat, a monopoly product, super-profitable, fast-growing. But it trades at a 44x forward price-to-earnings multiple (38x on next year), and it's up 50% this year and more than 130% over the past year. "What are investors expecting here? The stock to just continue rocketing up 50% every couple of months without end?" Carlson has been trimming, and he flagged a broader rotation: over the past month, while ASML, TSMC, and Micron gave back gains, money flowed into DoorDash, Texas Roadhouse, Meta, Moody's, and Mastercard. His read: "This could be the first indication that investors are starting to look outside of semiconductors."

Over on Squawk on the Street (July 16), Jim Cramer made the same point in his own idiom, calling it "a great unwind" of "parabolic moves." He was emphatic that this is technical, not fundamental: "I don't find that there's any change in the fundamentals. But the charts themselves, you get a lot of hot money, too much borrowed money. This is what happens. They unwind." He also pointed to a specific, under-discussed mechanism: South Korea changing margin requirements and halting listings of leveraged products tied to SK Hynix, which "means that there's just not as much buying power as there was." And on the supply story he was blunt: paraphrasing Arm CEO Rene Haas, "it's not a demand problem, it's a supply problem", "there's not enough right now Applied Material, not enough Lam."

Two heavyweights, two blowout quarters, and the market's verdict was a shrug. That gap, between how good the businesses are and how the stocks behaved, is the single most important thing to sit with this week.

Intel's former CEO on what went wrong (and a stat that should scare you)

The most substantive operator voice of the week wasn't about this quarter at all. On All-In (July 15), Intel's former CEO Pat Gelsinger, who joined Intel at 18 and was mentored by Andy Grove, Gordon Moore, and Bob Noyce, delivered a remarkably candid post-mortem on how one of the greatest American companies got "absolutely demolished."

His diagnosis: Intel stopped being run by technologists. "One of the things that went off the rail was when it started to be run by business people... the bean counters, the finance people." The proof is in the capital allocation. In the five or six years before he returned, he said, "Intel gave $100 billion to shareholders" in dividends and buybacks, while it "hadn't built a new factory in a decade" and passed on buying EUV machines. "How could you not buy EUV machines?" he asked. "There's just all of these things that you would only do as a technologist, because the economics behind them by themselves were not good."

He also gave the clearest framing of TSMC's rise: when he came back, "TSMC was producing 5x the wafers of Intel... and now it's more like 7x." TSMC's insight decades ago, "I don't care whose chip it is, I'll be your manufacturing partner", was dismissed by Intel as "a trivial piece of the business," and it became the model for the entire industry.

Then the part that should stop any investor cold. On Taiwan concentration risk, Gelsinger cited a recent Wall Street Journal report: the island of Taiwan has less than three weeks of energy reserves. "That should just put a chill in everybody's spine," he said, "because the blockade, after three weeks the island browns out. When you turn off a fab, it doesn't come back on for 90 days. The economic impact of a brownout of Taiwan is greater than the Great Depression." No shot needs to be fired, just three weeks without energy imports. He noted China "has blockaded the Taiwan Straits seven times over the last four years." On the brighter side: the CHIPS Act is working, slowly, US leading-edge chip production has gone from about 12% (2021) to roughly 18% now. "It's not 50. We have a long way to go."

His one source of comfort on the AI-bubble question was elegant: energy is the natural ceiling. "Nobody's going to build and buy GPUs and build data centers if they don't have energy," so "you have an upper bound on how aggressive and how hyped and bubbled we get."

The debate

The bull case dominated this week, and it's grounded in hard order books. ASML's 2027 EUV slots are sold out and 2028 pre-orders are in; TSMC is raising three-year capex after years of caution; both managements independently pointed to demand through 2029-2030. Berenberg's Tammy Chu argued the capacity expansion "is not coming to an end at any time," and stressed that ASML is careful, it "doesn't want to drag empty capacity through a downturn," so it only builds after talking to customers "on a very frequent basis." When the picks-and-shovels suppliers are this confident about orders two and three years out, that's the strongest evidence you can get that the AI buildout isn't a one-year sugar high.

The bear case, this week, was about price and cycles, not fundamentals. The clearest skeptic was portfolio manager Lance Roberts on Thoughtful Money (July 11). His warning is worth quoting because it's the mirror image of the bull case: semiconductors are now about 18% of the entire market's value, the group is "building a pretty defined head and shoulders" chart pattern, and "we're pricing in 2029 earnings right now for a lot of these companies. So the risk is any disappointment." His deeper point cuts at the supply story directly: much of the recent boom is price, not innovation. "They haven't really changed anything about their product except for the price... when supply starts increasing, there's nothing to protect those higher prices." High prices, he argued, always breed competition, and he pointed to China as the live example, forced by chip restrictions to "do more with less" by writing better software rather than throwing hardware at the problem.

Note the shape of this debate honestly: nobody on the podcasts this week made a fundamental bear case against TSMC or ASML specifically, no one argued their orders are fake or their moats are cracking. The bearishness was about valuation (Carlson, Roberts), about the mechanics of a crowded-trade unwind (Cramer), and about the cyclicality that always eventually visits this industry. That's a meaningful distinction. It means the market's anxiety right now is "how much is already in the price?", not "is the story real?"

One more strand worth folding in, from Monetary Matters (July 14), where investor Ben Pouladian reframed the whole shortage: "If there's not a GPU shortage, then... there's a shortage of powered land and tradesmen to build data centers." The bottleneck, in his telling, has moved downstream to electricity and construction. That's bullish for the equipment makers over years ("the backlog is measured in years and not quarters") but he flagged the same valuation tension as everyone else: the semi-cap names (ASML, KLA, Lam, Applied Materials, Tokyo Electron) "trade at by far the highest valuations in the semiconductor" space while Nvidia trades at a discount to its own history, "so there must be some sort of mean reversion at some point."

Read-throughs

Advanced packaging and memory (HBM/DRAM). The single most important structural shift in ASML's quarter was the memory/logic mix hitting 50/50 for the first time. That's the HBM boom pulling machine orders into memory fabs, concentrated in South Korea (43-45% of ASML shipments). Memory makers placing multi-year machine orders, enabled by take-or-pay contracts with the hyperscalers, is exactly the kind of forward commitment that historically the memory industry avoids. Watch whether that discipline holds.

Semi-cap equipment (AMAT, LRCX, KLAC, TEL). Cramer's "not enough Applied Material, not enough Lam" and Pouladian's note that Lam earns roughly 30% of revenue from recurring service on its installed base both point to a durable, annuity-like cushion under the toolmakers even if the order cycle cools. The offsetting risk everyone named: these are the highest-multiple names in chips right now. The install base is the floor; the multiple is the risk.

Fabless customers and the wafer-price squeeze. TSMC's warning that overseas fabs dilute margins by several points is really a preview of a coming negotiation: someone pays for pricier US wafers. On the a16z podcast (July 15), the discussion of custom silicon underlined how locked-in these customers are, Google's TPUs and Amazon's Trainium chips are shipping "in the millions," yet the guest argued a challenger has to be "5x better" because Nvidia wins on supply chain, HBM access, process node, and time to market. Tellingly, "AMD got to 2 nanometer before Nvidia" with higher-density HBM and 3D stacking "and yet they still lose." The upshot for the foundry: even the giants trying to escape Nvidia still funnel their most advanced silicon to TSMC.

China-domestic toolmakers. No podcast this week discussed SMEE, Naura, AMEC, or Piotech. The only China angle voiced was Lance Roberts' broader point about China leaning on software efficiency because it can't get the best hardware, a long-run share-loss risk to Western tools, but nothing new this week.

Quantum hardware (the long-dated frontier). The most concrete operator soundbite came from Andy Ory, CEO of QuEra, on Next in Tech (July 14). Ory's company uses "neutral atoms" moved around by lasers, rather than the superconducting approach used by Google's Willow chip. His pitch on why that matters for the supply chain: superconducting qubits "have to be in these baths, these 15-millikelvin refrigerants," are size-limited and expensive, and can only talk to adjacent qubits, while neutral atoms use lasers for "anywhere-to-anywhere connectivity" and need no exotic helium-3 cooling. The key milestone: QuEra plans "Libra," which Ory called "the first machine to go on the Amazon cloud that's a fault-tolerant quantum computer," in 2028.

The reason to take that seriously, and this is the genuinely new information, is that the underlying math has collapsed. The ratio of physical qubits to a usable error-corrected "logical" qubit has fallen from as high as 1,000-to-1 down toward something like 2-to-1 in demonstrations (and averaging out around 10-25:1), while the number of qubits an algorithm needs has dropped from "2 million... then became 200,000, now looks like 20,000." Three forces, bigger machines, better error-correction efficiency, leaner algorithms, are all pushing the same direction. Ory's one supply-chain worry isn't cooling; it's "lasers and specialized laser power."

The realist counterweight came from Catalyst with Shayle Kann (July 16), where the guest cautioned we're "three to four years away" from quantum systems delivering performance gains a scientist would actually choose over a classical computer, and that the algorithms, the "roads" for the quantum "car", are the harder, more underfunded problem than the hardware. For a hedge-fund book, the read is the same as it's been: the hardware timelines are compressing fast enough to watch closely, but this is still a long-dated theme, and the near-term "picks and shovels" (lasers, cryogenics, control electronics) are the tangible exposures.

What changed from prior weeks

For three straight weeks this section has led with the same lament: no fresh operator voices from the foundry/tool ecosystem, pundit-heavy weeks, with TSMC and ASML sitting quiet ahead of earnings. This week the operators finally spoke, because they reported. We now have hard numbers where we had previews: TSMC's three-year capex ramp is real, ASML's 2027 book is sold out, and the memory/logic mix has tipped to 50/50. The memory story that dominated late June and early July (Samsung's coronation, SK Hynix's US listing) is now visibly feeding the equipment order book rather than living only in the stock charts.

Still quiet, for continuity: Samsung Foundry (its 2nm GAA yields and the Taylor, Texas fab went unmentioned again, Samsung surfaced only as a memory maker and a possible future US-listing rumor); Intel Foundry's actual 18A/14A execution (this week's Intel content was Gelsinger's retrospective and the political story of the government's stake, not the roadmap); and the China-domestic toolmakers and metrology names (Advantest, Onto, Camtek), which drew no coverage at all.