AWS Capacity Largely Sold Out Through 2027, With On Demand Headroom Reserved

Amazon CEO Andy Jassy revealed that Amazon Web Services (AWS) has largely sold out its capacity through 2027, with a significant portion of 2028 already reserved. The remaining dedicated capacity will be allocated for on-demand use, the company's inference service Bedrock, and EC2. Jassy made these remarks on a video podcast with Ben Horowitz and Erik Torenberg of Andreessen Horowitz, according to Reuters on October 23.

AWS provides capacity assurance through On-Demand Capacity Reservations and Capacity Blocks for Machine Learning, allowing customers to reserve instances in a specific Availability Zone for a fixed time frame. AWS raised the rates of Capacity Blocks for Machine Learning by approximately 15% uniformly across all regions in January, covering P5en, P5e, P5, and P4d GPU instances as well as Trn2 and Trn1 instances.

Jassy stated that AWS had already allocated nearly all the capacity for 2026, and the company could invest even beyond the projected $220 billion in capital expenditures for the year without it being a poor use of capital. However, cloud analyst David Linthicum argued that the reporting overstated the company's position, distinguishing between pre-reserved blocks and the capacity an enterprise could allocate daily.

Linthicum highlighted the existing confusion around reserved instances, on-demand services, and committed capacity, leading to misguided assumptions and strategies. The discussion on Reddit focused more on current availability than the 2027 commitments, with many users reporting capacity issues with regular EC2 instances. Some technical operations managers reported trouble obtaining sufficient compute from major providers, not limited to GPUs.

The $220 billion figure refers to the total capital expenditure, not solely AI spending. Amazon's third-quarter results, scheduled for October 29, will show how much of the committed capacity is contracted revenue.