AWS Booked Solid Through 2027 as AI Demand Strains Cloud Capacity

Amazon's cloud infrastructure is running up against real capacity limits. CEO Andy Jassy recently disclosed that AWS has largely pledged its committed compute resources through 2027, with significant portions of 2028 already claimed—a sign of how aggressively organizations are locking in cloud capacity amid the AI arms race.

What's Being Reserved#

Speaking on a podcast with venture capitalist Ben Horowitz, Jassy explained that customers are securing access to computing power through On-Demand Capacity Reservations and Capacity Blocks for ML, mechanisms that let enterprises lock in instances across specific availability zones or time windows. AWS does maintain a reserve for general on-demand users and its Bedrock inference service, but the bulk of available capacity is now committed to long-term contracts. The company previously raised prices for Capacity Blocks for ML by roughly 15% across all regions, reflecting tightening supply.

The Debate Over "Sold Out"#

Linthicum, a cloud analyst, pushed back on the alarm, noting on LinkedIn that Jassy's remarks conflate two different scarcities. There's a real squeeze on multi-year, large-scale blocks—exactly what AI model providers need—but ordinary workloads can still find on-demand instances. The distinction matters: "Maybe not the exact massive, pre-reserved, hyperscale capacity block some large model provider wants for the next three years, but that's not exactly the same thing as 'AWS is sold out,'" he cautioned. Linthicum worried that headlines blur these categories and breed poor architectural decisions.

Real-World Tightening#

Yet on Reddit and in discussion forums, ordinary developers reported genuine friction. Standard EC2 instances have become harder to secure, with some reporting production Kubernetes clusters facing "concerning" resource constraints over the past month. Across the three major providers—AWS, Microsoft Azure, and Google Cloud—users described difficulty obtaining sufficient compute capacity, even beyond specialized GPU resources. This prompted concerns about an unintended consequence: faced with scarcity, teams reserve more capacity than they might use, reversing cloud's pay-as-you-go economics.

AWS already offers tooling to address this risk. The EC2 Capacity Manager, released last year, aggregates reserved-to-used ratios across accounts and regions to help teams spot inefficient reservations. Still, the underlying tension remains: when capacity feels scarce, organizations hoard.