AMD and SpaceX Both Beat Estimates Tuesday. Wall Street Sold Both Stocks Anyway.

At 4:05 p.m. Tuesday, AMD told Wall Street it had just booked $11.5 billion in quarterly revenue, a record, up 50 percent from a year earlier. Then the stock fell anyway. Forty minutes later, SpaceX, reporting its first earnings as a public company, said revenue jumped 92 percent to $7.8 billion, crushing estimates. Its stock fell too, and harder.

AMD posted adjusted earnings of $1.66 a share on revenue of $11.54 billion, beating consensus of roughly $1.62 on $11.3 billion, according to its SEC filing and Wall Street estimates cited by Quartz. Data center revenue reached $6.7 billion, more than doubling year over year, and that unit alone made up 58 percent of the company's total sales. By any normal reading, that is not a quarter that gets punished. AMD's report landed the same week as Intel's: revenue of $16.1 billion, up 25 percent, with its own data-center and AI business climbing 59 percent to $6.3 billion, CNBC and Quartz reported. Two chipmakers, same week, same story: AI demand is real and it is showing up in the numbers. Our read: the market has quietly changed what it grades a beat against, and Tuesday's stock drops were the first clean look at the new rubric.

SpaceX reported second-quarter revenue of $7.81 billion against a Wall Street estimate of $6.93 billion, and narrowed its net loss to $541 million from $1 billion a year earlier, CNBC reported. It closed the quarter, its first as a listed company, with $100 billion in cash and marketable securities and a $47.5 billion backlog, the company said in its earnings release. That is a debut most CEOs would frame and hang on a wall. The stock dropped anyway. Shares fell as much as 8 percent in after-hours trading on concerns over capital spending, which came in at $18.4 billion for the quarter, well above the $10.1 billion spent the prior three months and roughly $5 billion higher than analysts had modeled, Quartz reported. AMD's stock told the same joke: shares had climbed as much as 7 percent during the regular session, then slumped more than 8 percent after the print, even with a raised outlook, Quartz and CNBC reported.

Neither company missed. Both spent more than expected to keep growing, and that is what got punished. Meanwhile the broader market did the opposite. The Nasdaq-100 gained more than 3 percent as chip stocks rallied alongside strong earnings from Amazon and Palantir, with Palantir up 30.7 percent on the day, and a full 68 percent of U.S. stocks were advancing by midday, TheStreet reported. So the index closed at a fresh record the same afternoon its two most closely watched AI-infrastructure bellwethers got sold off after the bell. That split is the story: breadth at the top of the market, discipline at the single-name level, and the line dividing them runs through the capital-expenditure column, not the revenue line.

It is a small shift in mechanics with a large consequence. A beat used to be the finish line. Now it looks more like the opening bid, and the market's counteroffer is a simple ratio: what did you spend against what you made. AMD and SpaceX both cleared the first bar and tripped on the second.

None of this means AI spending is slowing. AMD's own numbers argue the opposite, and it is entering what its executives call a multi-year buildout, with Intel's data-center business climbing right alongside it. What changed Tuesday is narrower, and in its way more useful: investors now have two audited data points, not projections, showing that capital expenditure has become a line of scrutiny separate from revenue growth. That is a shift in market structure, not a mood swing. It means the next AI-linked earnings report will be read less for whether it beats and more for whether its spending-to-beat ratio looks like AMD's, wide but explainable, or like something the market decides it cannot explain.

Two companies posted the best quarters of their public lives on the same afternoon. Both got sold anyway, for the same reason: they spent to earn it.

Watch that ratio, not the headline number, on whichever AI-infrastructure name reports next.

How this edition was made →