Opinion · August 2026
The “Beat and Get Sold Anyway” Pattern Just Claimed Its Biggest Trophy Yet
On July 24, I wrote about Intel, Micron, and CoreWeave all reporting genuinely good news and getting sold anyway in the same session. Wednesday supplied the cleanest, biggest example of that pattern yet: SpaceX guided to a $100 billion run rate by December and $1 trillion of revenue by 2030, and the stock fell 13.61%. AMD posted record revenue with data-center sales doubling to $6.7 billion, and fell 7.04%. Neither company missed anything.
Two different mechanisms wearing the same disguise
It’s worth separating what actually happened, because the two selloffs aren’t really the same story. SpaceX’s drop is mostly mechanical: a post-IPO lockup expiring Thursday makes roughly 911.5 million insider shares tradable for the first time, which is a scheduled supply event, not a re-rating of the underlying business. Layer onto that Musk naming Nvidia the exclusive AI-chip supplier for SpaceX’s own compute buildout — a real competitive data point, and one that cost AMD a customer relationship the market had apparently been assuming would persist. AMD’s drop is the more informative one: flat 56% gross-margin guidance against 41% revenue growth, with capital spending nearly tripling to $808 million, is the market doing real work — asking whether AMD’s data-center growth is arriving at acceptable returns, not just whether it’s arriving.
Where this connects to my own thesis
Flat margins next to tripling capex is precisely the kind of number my Physical Limits of Compute whitepaper and the follow-up AI Capex Reality Check study flagged as the thing to watch for: not whether AI infrastructure revenue keeps growing, but whether the capital required to generate it keeps growing faster. AMD’s quarter is one data point, not a verdict, but it belongs in the same file as Alphabet’s and Meta’s capex-versus-free-cash-flow numbers from that study. The pattern I flagged in July hasn’t gone away; it just found two bigger names to run through.