Amazon Web Services has effectively run out of cloud computing capacity through 2026 and is largely sold out through 2027, with a substantial chunk of 2028 already committed to customers, according to Amazon CEO Andy Jassy. Speaking on a video podcast hosted by Andreessen Horowitz, Jassy said the company is struggling to keep pace with the relentless demand pouring in from AI laboratories and enterprise customers. He described a situation where even Amazon's planned capital expenditure of around $220 billion for 2026 is insufficient to satisfy the backlog.

The disclosure is one of the clearest signals yet that the AI infrastructure buildout has transformed cloud capacity from a flexible, on-demand resource into a scarce, pre-allocated commodity. For enterprises hoping to book large blocks of compute next year, the shelves are largely bare. Jassy acknowledged that a small pool is kept available for on-demand use through services like EC2 and the Bedrock AI model platform, but the big reserved chunks are spoken for well into the future.

The revelation lands at a pivotal moment for the broader tech sector. Markets have been jittery about AI spending after several large platforms reported that costs are surging ahead of revenues. Yet Jassy's comments suggest that from the infrastructure supply side, there is no slowdown in sight. He also flagged that AI inference remains roughly ten times too expensive for widespread business adoption, positioning Amazon's in-house Trainium chips — which he said offer 30-40% better price-performance than competitors — as a key tool for driving costs down.