Amazon raised its 2026 capital expenditure forecast to $220 billion from $200 billion on Thursday, and investors sent the stock up nearly 10% in after-hours trading. AWS revenue grew 37% year over year to $42 billion for the quarter, and total net sales rose 20%. The combination — heavy spending justified by heavy demand — is the setup investors are willing to pay for right now.
The infrastructure bill is now visible in the accounts. Amazon spent $173 billion on property and equipment in the fiscal year ended June 30, up from $107.65 billion the year before. That line item covers GPUs, natural gas turbines, and the plots of land they sit on. Amazon ended the quarter with $7.6 billion less cash than it held 12 months earlier, its first period of negative free cash flow this year.
Under most conditions, ballooning capex and shrinking cash would draw punishment from Wall Street. AWS's growth rate is what changed the calculus. A 37% jump on a base of $42 billion in a single quarter is enough to signal that data-center demand is scaling alongside supply, and the multi-year lag between breaking ground on a facility and selling its capacity makes forward demand signals especially valuable.
Key facts
- 01Amazon raised its 2026 capex forecast to $220B from $200B, after spending $173B on property and equipment in the fiscal year ended June 30.
- 02AWS revenue grew 37% year over year to $42B in Q2, and Amazon shares rose nearly 10% in after-hours trading.
- 03Amazon's cash reserves fell $7.6B over 12 months, marking its first period of negative free cash flow this year.
- 04Meta's stock dropped 8% after its own earnings, as investors punished heavy AI spending without a matching cloud-revenue engine.
- 05Property-and-equipment spending jumped from $107.65B the prior year to $173B — a category covering GPUs, natural gas turbines, and land.
Andy Jassy used the Q2 earnings call to argue that AWS's AI business will follow the same margin curve the retail cloud did a decade ago, and to make a strategic point about frontier models — that Amazon does not need to own one to win in AI infrastructure. Amazon Bedrock hosts models from third parties, most prominently Anthropic, whose Claude Opus 5 is served through the platform.
“We see the AI business following very much the same margin trajectory we saw in the core business before”— Andy Jassy, Amazon CEO
Beyond the data-center buildout, Amazon is pushing on its own silicon. The Trainium TPU and the Arm-based Graviton processor do not show up in headline capex numbers, but both are aimed at improving the margin structure of AWS's AI workloads over time. Custom chips let Amazon capture more of the economics that would otherwise flow to Nvidia.
Jassy's framing on models was blunt on the call.
“AWS and Amazon Bedrock can have a wildly successful business without its own frontier model, and the reason is that there's not going to be a single model to rule them all.”— Andy Jassy, Amazon CEO
The pattern repeated across the hyperscalers this earnings season. Microsoft and Google also posted strong cloud numbers and saw their shares rise. Meta went the other way — its stock fell 8% after quarterly earnings this week, as investors zeroed in on cash-flow strain and heavy AI spending without a matching cloud-revenue engine to offset it. The market is drawing a clear line between companies that host AI workloads and companies that only consume compute.
The catch is that hosting revenue and AI-lab spending are the same dollars viewed from opposite sides of the ledger. Anthropic's compute bill flows directly into AWS revenue. If the underlying economics of frontier labs and AI startups do not hold, the cloud revenue that investors are currently rewarding will not be stable either. AWS looks insulated from that risk because it is a step removed, not because it is exempt.
That is the version of David Cahn's $3 trillion question that matters for public-market investors. Either end-user demand for AI justifies the buildout at Amazon, Microsoft, Google, and Meta, or it does not. Cloud hosts sit further down the chain than the labs, but the chain is short.
For now, Amazon has the cleanest story in AI infrastructure: 37% growth on a $42 billion quarterly base, a plausible margin trajectory from Jassy, custom silicon coming online, and enough model-partner revenue through Bedrock to avoid betting the company on any single frontier lab. The market is treating cloud hosts as the safest position in the AI stack, and Q2 gave it a reason. The risk investors are quietly carrying is that safety is relative — AWS's growth depends on customers who are themselves being asked, quarter by quarter, whether the AI economy can pay for itself.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
