Amazon Web Services has nearly exhausted its available cloud capacity for 2027, with a substantial portion of 2028 already pre-sold, according to Amazon CEO Andy Jassy, underscoring the extraordinary demand pressures facing the world's largest cloud provider as the artificial intelligence boom accelerates.
Speaking on the a16z podcast, Jassy said AWS has effectively sold out its 2027 capacity, with only a limited reserve held back for on-demand services and the company's Bedrock inference platform and EC2 compute offerings. He indicated that a significant share of 2028 capacity has also been committed, reflecting a demand curve that shows no signs of flattening. "Even spending $220 billion this year, we're leaving a meaningful amount of demand on the table," Jassy said, adding that "a good chunk of '28 is sold out."
The disclosure highlights the scale of capital expenditure underway at hyperscale cloud providers as they race to build out infrastructure capable of supporting generative AI workloads alongside traditional cloud services. Jassy said Amazon is "probably putting more capacity [live] than anybody else" and remains committed to maintaining its leadership position in both infrastructure and AI. He noted that the company could deploy significantly more capital than the $220 billion planned for this year without fully satisfying existing demand, describing the current market environment as one of "incredible demand" across both AI and non-AI cloud segments.
However, Jassy acknowledged that supply chain constraints are tempering the pace of expansion, with memory components among the most acute bottlenecks. "I wish I could tell you that I thought that the supply constraints would be just a 2027 issue," he said. "I think we'll see them in different components for several years." The remarks suggest that the industry-wide scramble for critical hardware — from high-bandwidth memory to advanced networking equipment — will remain a structural challenge for cloud operators well beyond the near term, forcing providers to make difficult tradeoffs about which workloads and verticals to prioritize as they allocate scarce capacity.