Western Digital beat every cited operating expectation, and its stock still fell about 12%.
The company exceeded expectations for revenue, adjusted EPS, gross-margin guidance, and expense guidance. Yet its shares declined after gaining more than 200% year to date.
That reaction was not isolated. Barclays reported that 85% of S&P 500 companies beat Q2 EPS estimates, while reactions to both beats and misses were unusually negative.
AI demand is translating into revenue, operating leverage, and free cash flow. But when valuations already anticipate years of exceptional execution, the market is no longer asking whether the numbers are good. It is asking whether they justify what investors have already paid for future growth.
AI operating leverage carries an extraordinary capital bill
SpaceX reported Q2 revenue of $7.8B, up 92% Y/Y. Its AI business generated $2.6B, up 247% YoY, while AI EBITDA swung from a $600M loss in Q1 to a $1.1B profit.
Those figures suggest compute expansion is already producing meaningful operating leverage. It is no longer merely a promise about future demand.
The capital requirements are equally real. SpaceX reported $18.4B in CapEx, including $15.8B attributed to AI. It is targeting more than 2GW of compute by year-end and potentially approaching 10GW by the end of 2027.
Management also reported ambitions of reaching a $100B run-rate revenue by December 2026 and $1T in revenue by 2030. Those ambitions are enormous, but there was not enough detail to assess them independently.
The buildout may reward suppliers beyond GPUs
Musk said SpaceX will build exclusively on Nvidia and called Vera “the best architecture and the best AI computer.” No volumes, pricing, or contract terms were disclosed.
He separately identified memory as the current bottleneck, claiming supply is growing around 20% annually against demand of 200% or more. If that assessment holds, the economics of the buildout may spread beyond GPUs. $MU, $SKHY, and Samsung could benefit from a constraint that investors focused only on compute accelerators may underestimate.
“Cheap” valuations still demand years of execution
The bullish case for CoreWeave values $CRWV at 2x projected 2027 sales, compared with 5x for $NBIS. It argues that converting the backlog while deploying 8 GW of capacity could produce approximately $79 billion to $80 billion of annual revenue.
The $135 target assumes GPU hourly rates remain constant, capacity is deployed successfully, and interest expense falls from approximately 22% of revenue to about 7% in 2030. A low sales multiple offers limited comfort when the projected sales and financing improvements require years of successful execution.
AMD makes the valuation tension even clearer. Q2 revenue reached $11.5B, up 50% YoY, while Data Center revenue rose 107% to $6.72B.
Yet one valuation analysis found limited upside even after assuming 40% annual growth through 2030, $200 billion of revenue, a 40% net margin, and a 20x earnings multiple. AMD could exceed those assumptions, particularly if customers seek more supply and alternatives to Nvidia. But under that analysis, extraordinary performance produces only limited upside.
Strong fundamentals can win, but price determines how much investors keep
Shopify beat every cited Q2 benchmark and guided Q3 revenue growth to 32%, above the estimated 26%. It generated $654 million of free cash flow, and the stock closed roughly 17% higher.
That revaluation reflected genuine operating strength. The unresolved question is whether the business improvement justified the full change in price.
Western Digital shows how narrow the margin for error has become. AI demand is real. The harder investment question is how much of that reality investors have already purchased in advance.