Amazon reported better-than-expected Q2 earnings on Thursday, with net sales up 20% and cloud revenue as a standout driver. Shares jumped nearly 10% in after-hours trading. Despite investor worries about capex, Amazon is not slowing data-center spending. The company spent $173 billion on property and equipment (GPUs, turbines, land) for the fiscal year ended June 30, up from $107.65 billion the prior year. It raised its 2026 capex forecast from $200 billion to $220 billion, and ended the quarter with $7.6 billion less cash than 12 months ago, its first period of negative free cash flow this year. AWS revenue grew 37% year-over-year to $42 billion for the quarter, helping justify the spending given the long time lag between breaking ground and selling capacity.

Beyond data centers, Amazon is betting on custom silicon like Trainium and Graviton to improve margins. CEO Andy Jassy noted the AI business follows a similar margin trajectory to AWS‘s core business and that AWS can succeed without owning a frontier model because no single model will rule all. This pattern echoes Microsoft and Google, whose shares rose after strong cloud revenue. In contrast, Meta’s stock fell 8% after earnings, as investors focused on heavy capex without a clear revenue source.

Investors treat cloud hosts as the most reliable part of the AI stack, while remaining skeptical about AI labs and startups. However, Amazon‘s hosting revenue is someone else’s AI bill—in Anthropic’s case, literally the same money. If AI spending isn’t sustainable for labs and their clients, cloud revenue won’t be stable either. The article concludes that, as David Cahn’s $3 trillion question suggests, either enough demand exists to justify the buildout or it doesn’t, and cloud hosts aren’t insulated from that demand problem.