AWS capacity sold out through 2027? What Andy Jassy said, and what it means for on-demand users
Victoria Shinder2 min readAmazon CEO Andy Jassy said in a podcast with Andreessen Horowitz's Ben Horowitz and Erik Torenberg, reported by Reuters on 23 October, that AWS's large committed cloud capacity is largely sold out through 2027 and that a good chunk of 2028 is already allocated. In the same remarks he said AWS always leaves a certain amount for on-demand use, for EC2 and for its Bedrock inference service. InfoQ collected the reactions on 1 November.
The distinction matters, and it is the one the cloud analyst David Linthicum drew: Jassy was describing the big pre-reserved blocks sold through On-Demand Capacity Reservations and Capacity Blocks for ML, where a customer books instances in a named Availability Zone, in the second case for a fixed future window. Linthicum's point was that he can still open the console today and allocate GPU capacity, just not a three-year hyperscale block, and that headlines saying AWS is sold out lead to bad assumptions.
The engineers' reports quoted by InfoQ point the other way for everyday work. One developer on Reddit described trouble getting enough on-demand capacity for ordinary EC2 instances over the past month, mainly for CI but sometimes for production Kubernetes clusters; others reported similar constraints at other providers and beyond GPUs. A commenter warned that the natural response, reserving capacity just in case, is an anti-cloud pattern.
What it means
For most teams the statement is not that capacity is gone, but that the cheap assumption behind on-demand, that a reasonable instance is always there when asked for, is getting weaker at the margin. The useful move is to measure it: CI fleets that can fall back across instance types and regions absorb a shortage, and reservations bought only out of fear show up as unused capacity, which AWS's own EC2 Capacity Manager reports as a ratio.