AWS to Implement 20% Price Hike on GPU Instances for Machine Learning Workloads

Amazon Web Services (AWS) has announced a substantial price increase for its EC2 Capacity Block reservations, specifically targeting machine learning GPU instances. Starting July 1, 2026, customers utilizing these reserved capacities will see an approximate 20% hike in hourly rates. This decision, as stated by AWS, is a direct response to the prevailing supply and demand dynamics within the high-end GPU market. The price adjustments will impact AWS's most potent Nvidia-powered instance families, including the P6-B300, P6-B200, and various P5 series instances across different regions. For example, the P6-B300 will be billed at $14.04 per hour, and the P6-B200 at $12.355. While these specific GPU instance prices are rising, AWS has indicated that other EC2 prices will remain unchanged. This is not the first price adjustment for these services; AWS had previously raised prices by about 15% in January 2026. The consistent increases underscore the intense enterprise appetite for GPU compute, driven by the escalating demands of AI training and inference workloads. Customers who rely on Capacity Blocks for ML typically reserve these instances for time-bound, large-scale projects, often willing to pay a premium for guaranteed availability amidst scarcity. The broader context for this price surge includes Amazon's significant investment in AI infrastructure, with roughly $200 billion committed in 2026. Reports also suggest that Amazon is slated to receive one million Nvidia GPU chips by the end of 2027, highlighting the ongoing supply constraints in the high-performance GPU market. The move by AWS will likely be closely watched by competitors like Microsoft Azure and Google Cloud, as their responses could influence the competitive landscape for AI workloads.