Amazon reported better-than-expected Q2 2026 earnings on July 30, 2026, with net sales rising 20% and AWS revenue climbing 37% year over year to $42 billion. The stock jumped nearly 10% in after-hours trading. The results arrived alongside Amazon's decision to raise its 2026 capital expenditure forecast from $200 billion to $220 billion, and investors did not flinch. The reason is straightforward: Amazon's spending on AI infrastructure — the data centers, servers, and networking equipment that run AI applications — is producing measurable cloud revenue growth.

The contrast with Meta was immediate. Meta's stock fell 8% after its own Q2 2026 earnings, with investors focused on a cash flow squeeze and continued spending without a clear AI revenue source. Amazon, Microsoft, and Google all saw their stock prices rise after reporting strong cloud revenue in the same quarter. The market is drawing a line between companies monetizing AI through cloud infrastructure and those spending on AI without a corresponding revenue stream.

Amazon's capital intensity is enormous by any historical measure. The company spent $173 billion on property and equipment for the fiscal year ended June 30, 2026, up from $107.65 billion the prior year. The raised $220 billion capex forecast for calendar 2026 extends that trajectory. Amazon ended Q2 2026 with $7.6 billion less cash than it held twelve months earlier, marking its first period of negative free cash flow in 2026. That figure did not deter buyers. AWS revenue growth at 37% year over year provides the revenue-side answer to the spending-side question.

CEO Andy Jassy, speaking on the Q2 2026 earnings call, said that AWS and Amazon Bedrock — Amazon's service that lets customers access and build on multiple AI models through a single platform — can have a successful business without its own frontier model. The strategic implication is that Amazon is positioning itself as the compute and inference layer for third-party AI models rather than competing primarily at the model layer. Inference, in this context, is the process of running a trained AI model to produce outputs — the step where the model is actually used, as opposed to the training step that creates it.

Anthropic's AI compute spending with Amazon, described by Jassy as "literally the same money" flowing between the companies, illustrates the circular economics at work: Amazon invests in Anthropic, Anthropic pays Amazon for compute, and both sides book the transactions.

Amazon's Q2 2026 net income also included non-operating pre-tax other income of $53.4 billion, primarily from its investments in Anthropic, per the company's earnings release. That paper gain substantially flatters reported profitability even as operating cash flow turns negative on a capex-adjusted basis. Investors appear to be looking through the non-operating income and focusing on AWS revenue acceleration, but the composition of net income is worth tracking in future quarters if Anthropic's valuation stabilizes or declines.

Amazon's custom silicon strategy adds another layer. The Trainium chip and the Arm-based Graviton processor do not appear in capex numbers but can improve cloud margins by reducing dependency on third-party accelerator pricing — that is, what Amazon pays companies like Nvidia for their chips. If AWS can shift inference workloads onto internally designed silicon at lower cost per token (the basic unit of text an AI model processes) than Nvidia-based alternatives, the unit economics of AI cloud services improve materially. The capex figures capture data centers, power infrastructure, and rack-level hardware; the chip design investment is an R&D line item that does not show up in the same spending bucket but directly affects gross margin downstream.

Microsoft and Google, the other two hyperscalers — the handful of companies operating cloud infrastructure at a global, massive scale — reporting strong cloud revenue in Q2 2026, received similar investor approval. The pattern across all three is consistent: cloud revenue growth that can be plausibly attributed to AI workloads earns capex forgiveness. Meta's miss on that test cost it 8% in a single session.

The broader context here is that the market is pricing AI infrastructure as a land grab with a short payback window for cloud providers and an uncertain one for everyone else. Amazon's negative free cash flow, $220 billion capex guide, and the $53.4 billion in Anthropic-driven non-operating income all sit inside the same quarter, and the stock went up 10%. That is a market making a directional bet: that AWS revenue growth will compound faster than the infrastructure depreciation and capital costs that underpin it. The bet is not unreasonable given the 37% AWS growth rate, but it carries embedded assumptions about AI compute demand sustaining or accelerating through 2027 and beyond. If enterprise AI inference demand plateaus before the installed data-center base is fully utilized, the hyperscaler economics shift quickly. Amazon's custom silicon advantage narrows that risk by improving per-unit margins, but it does not eliminate it.

For now, the market's verdict is clear and the revenue is real. Whether the spend-to-revenue ratio holds at current levels is the question that will define 2027 earnings season.