ASML Raised Guidance Twice While Chip Stocks Sold Off

Early July was a bloodbath for semiconductor equities. The Philadelphia Semiconductor Index fell hard after a historic run, wiping out more than a trillion dollars in market value. Micron dropped as much as 13% in a session. Intel bled 21% over seven trading days. The thesis writing itself in real time was that AI capital spending had finally outrun the demand that justified it — that Meta's plan to resell surplus compute had rewritten the supply-demand equation, that SK Hynix delaying HBM4 expansion in favor of DDR5 was the tell, that the cycle had turned.

Then, on July 15, ASML reported.

The numbers

Net sales of €9.3 billion against a consensus near €8.8 billion. Net income of €2.2 billion. Both sales and gross margin came in above the company's own guidance.

And then the part that matters more than any quarter: ASML raised full-year 2026 guidance to €43-45 billion, up from €36-40 billion — for the second time this year. At the top end, that is a 19% increase over the prior view, delivered in the middle of a selloff premised on AI spending rolling over. Gross margin guidance went up too, to 54-56% from 51-53%. Shares rose about 6% pre-market.

Then ASML said it would expand capacity by roughly 30% annually across both EUV and DUV lines for each of the next two years — and disclosed that the guidance already incorporates demand from projects including Musk's planned Terafab in Texas.

Companies do not raise revenue guidance 19% and simultaneously commit to two consecutive years of 30% capacity expansion because they think the next eighteen months are uncertain. That is a company reading a booked, contracted, deposit-backed order file and telling you what's in it.

Why ASML is the tell

Every argument about AI capex eventually reduces to a question about physical constraint, and ASML sits at the bottom of it. Nvidia's chips are printed on TSMC's advanced nodes. TSMC's advanced nodes exist because of EUV lithography. EUV lithography is ASML, and only ASML — a genuine monopoly on the machines without which no leading-edge AI silicon gets made by anyone, anywhere, at any price.

That position makes ASML's order book the least noisy signal in the industry. Nvidia's revenue can be inflated by a customer stockpiling. Micron's memory pricing swings on inventory cycles that have nothing to do with AI. Meta's capex line is a strategic announcement as much as a spending plan. But an EUV order is a multi-year, nine-figure commitment to a machine that takes years to build and install, placed by a foundry that must already know who is buying the wafers. Nobody orders a scanner speculatively.

So when the market says AI spending is cracking and ASML says demand is 19% better than we told you three months ago, one of those is a price and one is a purchase order. They are not the same class of evidence.

What the selloff actually was

This is where the two stories reconcile, because the bears weren't hallucinating — they were looking at the wrong layer.

The early-July damage concentrated in memory and in Intel. Micron's drop followed SK Hynix's HBM4 deferral, which is a real datapoint about a real thing: the memory attach rate to AI compute, and whether HBM's margin premium holds. Intel's 21% is a company-specific story that has been running for years. Meta Compute is a genuine threat — to the pricing power of everyone renting out GPUs, which is a crowded, undifferentiated business with no moat at all.

None of that touches lithography. A GPU glut at Meta doesn't reduce the number of scanners TSMC needs; if anything, a world where surplus compute gets resold cheaply is a world where compute demand is more elastic, not less. The selloff was a repricing of the rentier layer of the AI stack — the people who bought hardware hoping to lease it — dressed up as a verdict on the whole thing. ASML's quarter is the reminder that the layer underneath is sold out.

The counterexamples were already there for anyone reading past the index. SK Hynix raised $26.5 billion in the largest foreign listing in U.S. history and immediately committed $8.6 billion to buy EUV equipment from ASML — a company allegedly moderating its AI exposure spending nearly nine billion dollars on the machines that make it. Micron raised its U.S. investment target to $250 billion through 2035. These are not the capital allocation decisions of firms that think the cycle just ended.

What actually turns on this

The honest caveat: ASML's visibility is long, which cuts both ways. An order file that extends years out is slow to reflect a downturn as well as an upturn. If AI demand genuinely breaks in 2027, ASML would be among the last to see it in the numbers, not the first. A raise today is evidence about what was contracted over the last several quarters — a strong claim about the past and present, a weaker one about the far side of the decade.

But that is precisely the timescale the bear case was making claims on. The July selloff wasn't a thesis about 2030. It was a thesis that the spending was breaking now. ASML answered that question with its order book, and the answer was that it isn't — that the buyers with the longest planning horizons and the most information about actual wafer demand just committed harder, and that the company with a monopoly on the binding constraint is spending two years' worth of capacity expansion to keep up.

A trillion dollars left the sector on a story. The one company positioned to know whether the story was true responded by building 30% more machines.