AI Costs Rise: AWS Increases Prices for Reserved Compute Resources for Model Fine-Tuning

The demand for AI chips, RAM, and SSDs has sharply driven up the prices of components and computers in general. Services are also starting to cost more. In the case of AWS, the company has updated the prices as of July 1st for reserving Amazon EC2 Capacity Blocks for ML, necessary to run AI and machine learning workloads. This is a significant increase, with costs now having risen by 20%.

AWS Raises Prices for Certain AI Workloads

The service with Amazon EC2 Capacity Blocks for ML allows the reservation of the amount of acceleration capacity necessary to execute machine learning workloads. The price of EC2 Capacity Blocks includes a reservation fee and a fee for the operating system. As indicated on the company’s website, "Starting from July 1, 2026, the hourly rates for accelerators will be: P6-B300 at $14.04 (all available AWS regions except AWS GovCloud regions), P6-B200 at $12.355 (all available regions except AWS GovCloud regions), P5 at $5.191 (all available US regions), P5 at $4.72 (all non-US available regions), P5e at $5.97 (all available regions), P5en at $6.865 (all available US regions), P5en at $6.241 (all non-US available regions), and P4de at $2.214 (all available US regions). All other prices will remain unchanged."

This isn’t the first time; back in January 2026, AWS had already raised prices by about 15%.

Prices for other AI services, however, remain unchanged. It should be noted that Amazon EC2 Capacity Blocks for ML is not a necessary option for anyone running AI workloads. In fact, they are intended to ensure the availability of massive hardware resources for intensive and immediate workloads (like training and fine-tuning AI models) while avoiding queues and without needing to make long-term commitments, and are primarily used by those developing AI models for fine-tuning purposes.

Despite other services not experiencing price increases, it is evident that the success of AI is putting pressure on hardware manufacturers, both for chips and memory, who are struggling to keep pace with the enormous demand. This, in turn, is leading to ever-increasing costs for users.