AWS Raises GPU Capacity Block Prices 20% for AI Workloads

Key Moments

Details of the Upcoming Price Changes

Amazon’s (NASDAQ:AMZN) Amazon Web Services unit will implement a roughly 20% price increase on EC2 Capacity Block reservations for machine-learning GPU instances, effective July 1, 2026. The adjustment was disclosed in AWS’s official documentation, which attributed the move to prevailing supply-demand conditions.

The new pricing will apply specifically to some of AWS’s most advanced Nvidia-based GPU offerings. According to the updated documentation, the hourly rate per accelerator will be set as follows:

AWS indicated in the documentation that all other EC2 prices will remain unchanged.

On its pricing page, the company stated: “Amazon EC2 Capacity Blocks for ML reservation prices are updated periodically based on supply and demand.”

Context: Surging Demand and AWS Growth

The increase comes amid sustained strength in enterprise demand for GPU computing capacity. AWS revenue rose 28% year-over-year to $37.6 billion in the first quarter of 2026, representing the cloud division’s fastest growth rate in more than three years. That growth trajectory has provided AWS with meaningful pricing power as customers expand AI training and inference workloads on its platform.

Amazon has earmarked roughly $200 billion in capital expenditure in 2026 for AI-related infrastructure. Reuters reported in March 2026 that Amazon is expected to receive 1 million Nvidia GPU chips by the end of 2027 through a cloud supply agreement, highlighting persistent tightness in the high-end GPU supply chain.

How Capacity Blocks for ML Fit Into AWS Strategy

Capacity Blocks for ML are a reserved-capacity mechanism designed to help enterprises secure limited GPU instances for scheduled, time-bound jobs, commonly large-scale model training efforts. Because this product guarantees access at future dates, customers have been willing to pay a premium relative to spot-market prices.

The new pricing schedule raises that premium meaningfully. For reference, ahead of this adjustment, on-demand pricing for a P6-B200 eight-GPU node was already around $14.24 per hour for the entire node, based on pricing analysis from Spheron Network published on June 20.

Implications for Nvidia and Competing Clouds

For Nvidia, the higher AWS reservation prices reflect a complex signal. The constrained availability of P5 and P6 instances, which are based on Nvidia’s Blackwell (B200, B300) and Hopper (H100) GPU architectures, underscores strong downstream demand for its chips. At the same time, steeper reservation costs on AWS may motivate some customers to weigh alternatives, including Nvidia-based services on other cloud providers or Google Cloud’s TPU-focused offerings, which Alphabet has been positioning as cost-conscious options.

Market participants will be watching closely to see whether Microsoft Azure or Google Cloud adopt similar moves on GPU reservation pricing. Azure is AWS’s closest competitor in enterprise cloud infrastructure, and an AWS-only increase could either drive broader repricing across the sector or give Azure and Google Cloud an opportunity to court cost-sensitive AI workloads.

Another unresolved issue is how existing Capacity Block reservations will be treated. It is not yet clear from AWS’s documentation whether reservations booked before July 1 will continue at the original rates or transition to the new pricing structure from that date.

Customer Decisions Ahead of the July 1 Change

With the new rates set to take effect in less than a week, enterprise customers face immediate trade-offs. Organizations must decide whether to secure any remaining Capacity Block inventory at current prices before July 1 or adjust budgets and project economics to accommodate the higher reservation costs as an ongoing feature of AI infrastructure built on AWS.