Amazon's Cloud Growth Gives Investors Patience for a $220 Billion AI Buildout

Amazon's second-quarter results showed why public investors are treating cloud platforms differently from other companies spending heavily on artificial intelligence. Net sales increased 20%, AWS revenue rose 37% to $42 billion and the company's shares gained nearly 10% after hours.

At the same time, Amazon raised its 2026 capital-spending forecast from $200 billion to $220 billion. Property and equipment spending reached $173 billion for the year ended June 30, covering data centers, accelerators, power generation and land. Cash declined from a year earlier as the investment program expanded.

The difference is visible revenue. AWS customers are already paying for compute, storage and AI services, giving investors evidence that demand is rising alongside construction. Data centers take years to plan and build, so current cloud growth helps support spending that will not produce its full capacity immediately.

Amazon is also developing its own Trainium and Graviton chips, which could improve margins by reducing dependence on more expensive third-party processors. CEO Andy Jassy said the AI business can follow the margin path of AWS even without Amazon owning the single dominant frontier model.

The circular flow of AI spending deserves attention. Model companies buy cloud capacity, cloud providers invest in model companies and both sides report growth linked to the relationship. The arrangements can be strategically valid, but investors should separate contracted infrastructure commitments from broad customer demand when judging whether the market can sustain current construction plans.

Cloud providers are benefiting because every model developer needs computing capacity, even when the identity of the eventual application winner remains uncertain. That position resembles selling infrastructure during a rush, but it comes with enormous capital commitments and energy requirements. Investors should compare revenue growth with depreciation, contracted demand and the risk that customers build their own hardware or become insolvent. The host can win across several models, yet an overbuilt data center is not automatically valuable. Utilization and durable customer economics matter more than the announcement price of a new campus.

The market's patience is not unlimited. Amazon has entered negative free-cash-flow territory and must keep utilization high as new capacity arrives. Still, its results suggest investors are more comfortable funding AI infrastructure when it is attached to a proven cloud revenue engine. The same spending looks far riskier when the path from compute to paying customer remains unclear.