Amazon's AWS is raising prices on EC2 Capacity Blocks for ML, a service that lets companies reserve GPUs in advance to keep AI workloads from being interrupted, by roughly 20% starting in July, on top of a 15% increase in January. AWS said it revisits these reservation rates from time to time as supply and demand shift, and the service is popular with serious AI developers running big-budget projects like training or fine-tuning large models. Amazon stressed the hike applies to one purchasing option and that it remains committed to fixed pricing on alternatives customers can choose for AI workloads. Behind the increase is a broader squeeze, as tight high-bandwidth memory supply and strong GPU demand raise costs and give hyperscalers like AWS, Microsoft, Google, and Oracle more pricing power when capacity is scarce.
