AWS reports cloud capacity largely sold out for 2027 commitments

Article Summary

Amazon CEO Andy Jassy announced that AWS cloud capacity is largely committed through 2027 and partially allocated for 2028, though the company continues to reserve headroom for On Demand, EC2, and Bedrock services. While Jassy noted that AWS could exceed its planned $220 billion capital expenditure for this year, analysts and developers report ongoing difficulties in securing compute resources for both specialized AI workloads and standard EC2 instances. AWS previously increased Capacity Blocks for ML rates by 15% across all regions in January to manage high demand for GPU-intensive instances.

AI Analysis

## Core Assessment

AWS says its large capacity commitments are largely booked through 2027, while some On-Demand, EC2 and Bedrock capacity remains reserved for flexible use. That is not proof that all ordinary capacity has disappeared. It also does not mean every customer can obtain the instance they need: InfoQ quotes a developer who had trouble obtaining regular on-demand EC2 capacity and reported constraints affecting production Kubernetes clusters on some days. Those accounts are useful evidence of individual problems, not a measurement of availability across AWS.

## Blunt Take

“Sold out” can make the whole cloud sound unavailable; “there is still headroom” can make a real provisioning failure sound imaginary. Neither slogan tells an engineer whether a particular instance type will start in a particular region today.

## Why It Matters

Long-term capacity blocks and on-demand instances answer different planning needs. An analyst quoted in the report cautions against conflating booked hyperscale blocks with ordinary availability, while other practitioners describe shortages and the temptation to reserve capacity in advance. The January price rise for some ML capacity blocks adds cost pressure, but does not by itself establish why every instance shortage occurred.

## Impact and Advice

Cloud teams can track failed capacity requests by region and instance type, then assess reservations or alternatives for workloads that cannot tolerate delays. Reservations can improve certainty while leaving unused capacity to manage; the report points to EC2 Capacity Manager as a way to monitor reservation use across accounts and regions.

Explore this story further