Opinion · July 2026
Same Earnings Season, Opposite Verdicts: Amazon, Apple, and the AI-Capex Market
Amazon and Apple both beat earnings estimates on Friday. Amazon rose 15.32%. Apple fell 7.35%. Read side by side, the two reactions are the clearest data point yet for a question I’ve been circling since the July 17 piece on AI-capex sustainability and the July 28 piece on CXMT: the market isn’t grading whether AI infrastructure spending is happening. It’s grading, company by company, whether that spending shows up as revenue or only as cost.
One beat rewarded, one beat punished
Amazon’s case for the market was concrete: AWS revenue grew 37% to $42.2 billion, its fastest pace in 18 quarters, and its AI and custom-chips businesses each now clear $25 billion in annual run-rate revenue with triple-digit growth. Investors shrugged off $7.6 billion of negative trailing free cash flow and a raised $220 billion capex plan because operating cash flow still grew 33% to $161.4 billion — the spending has a visible revenue line attached to it.
Apple’s problem wasn’t uncertain demand — iPhone sales grew 22% to $54.3 billion, ahead of expectations. It's that AI infrastructure spending, which Amazon can point to as a source of revenue, shows up on Apple’s income statement purely as a cost: rising memory and component prices squeezing guided gross margin from over 50% to 47–48%, because Apple is a buyer of the same scarce NAND and DRAM that hyperscaler data centers are bidding up. Same underlying shortage, opposite balance-sheet exposure.
Where this connects to my own research
This is the same free-cash-flow-versus-capex tension from my AI Capex Reality Check, seen from a different angle. For hyperscalers, AI capex is an investment with a visible revenue return: AWS, Azure, Google Cloud. For a buyer of memory like Apple, it's a straight cost input with no offsetting revenue line. It's also why Micron and Sandisk both fell Friday even as Apple’s and Amazon’s own earnings calls confirmed durable memory demand — that move traced more to Treasury yields hitting 19-year highs and pressuring growth stocks broadly than to any new doubt about the shortage itself, per Motley Fool’s reporting. The memory-scarcity trade underlying my Sandisk and Micron thesis is a different, better-supported bet than the general “is AI capex sustainable” question, but Friday was a reminder it isn’t immune to a broad, rate-driven risk-off day.
None of this closes the debate; it sharpens it. The next few weeks of earnings will keep testing the same split — does this company’s AI spending show up as revenue, or only as cost — and I’d rather track that distinction company by company than treat “the AI trade” as one undifferentiated bet.