Amazon Web Services will raise its EC2 Capacity Block reservation prices for machine-learning GPU instances by about 20% effective July 1, 2026, according to an update posted to the company's official documentation. The adjustment targets reservation-based access to high-end Nvidia-powered accelerators and is tied by AWS to supply-and-demand dynamics.
The new per-accelerator hourly charges apply across AWS's most powerful Nvidia instance lines. The updated rates are:
- P6-B300: $14.04 per hour
- P6-B200: $12.355 per hour
- P5 (U.S. regions): $5.191 per hour
- P5 (non-U.S.): $4.72 per hour
- P5e: $5.97 per hour
- P5en (U.S.): $6.865 per hour
- P5en (non-U.S.): $6.241 per hour
- P4de (U.S.): $2.214 per hour
AWS said on its pricing page:
"Amazon EC2 Capacity Blocks for ML reservation prices are updated periodically based on supply and demand."
The company noted that other EC2 prices remain unchanged in the same documentation.
This price increase arrives as enterprises sustain elevated demand for GPU compute. AWS reported 28% year-over-year revenue growth to $37.6 billion in the first quarter of 2026, the unit's fastest expansion in over three years, a performance that the cloud business says gives it pricing leverage as customers rely on AWS for AI training and inference workloads. Amazon has planned about $200 billion in capital expenditure for 2026 directed at AI infrastructure, and reporting in March 2026 indicated Amazon is slated to receive 1 million Nvidia GPU chips by the end of 2027 under a cloud supply agreement - underscoring the tightness in the high-end GPU market.
EC2 Capacity Blocks for ML are a reservation product that lets customers secure scarce GPU instances on a specified future date for time-limited workloads, typically large-scale model training. Because Capacity Blocks guarantee availability, enterprises have been prepared to pay a premium above spot-market rates; the newly posted rates mark a material increase in that premium.
For additional pricing context, Spheron Network published an analysis on June 20 showing P6-B200 on-demand rates for an eight-GPU node were already running at about $14.24 per hour for the full node prior to the reservation-price adjustment.
For Nvidia, the pricing update presents a mixed signal. The constrained availability of P5 and P6 instances - built on Nvidia's Blackwell (B200, B300) and Hopper (H100) GPU architectures - confirms strong end-market demand for Nvidia silicon. At the same time, higher reservation costs could encourage some AWS customers to consider alternatives, including Nvidia-powered offerings on competing clouds or Google Cloud's TPU-based instances, which the article indicates Alphabet has been marketing as a cost-competitive option.
Whether Microsoft Azure or Google Cloud will mirror AWS with similar reservation price rises remains to be seen. Azure is identified as AWS's closest rival in enterprise cloud infrastructure; a one-sided upward repricing by AWS could prompt either competitive matching or create an opportunity for rivals to attract price-sensitive AI workloads.
It is also unclear from AWS's documentation whether Capacity Block reservations made before July 1 will remain at prior rates or be billed under the new schedule beginning July 1. That ambiguity leaves enterprise buyers with a near-term decision: either lock in remaining capacity at pre-July rates before the increase takes effect or accept the higher costs as part of the evolving AI infrastructure pricing landscape that AWS has helped shape.
Summary
AWS will increase EC2 Capacity Block reservation prices for several Nvidia GPU instance families by roughly 20% on July 1, 2026. The move affects P6 and P5 lines and follows strong AWS revenue growth and major capital spending plans for AI infrastructure. The change raises costs for guaranteed reserved GPU capacity and leaves open whether earlier reservations will keep prior pricing.
Key points
- AWS will implement a roughly 20% uptick in EC2 Capacity Block reservation prices for high-end Nvidia instances effective July 1, 2026, with specific per-accelerator hourly rates published by the company.
- The increase comes amid robust demand for GPU compute and a period of accelerated AWS revenue growth, alongside a substantial capital expenditure program focused on AI infrastructure.
- Sectors most directly affected include cloud infrastructure providers, enterprise AI users that rely on reserved GPU capacity, and suppliers of high-end GPUs.
Risks and uncertainties
- It is unclear whether Capacity Block reservations made before July 1 will be billed at the old or new rates, creating short-term billing uncertainty for enterprise customers - this impacts finance and procurement functions in AI-heavy firms.
- Higher reservation prices could prompt customers to evaluate competing cloud or specialized hardware alternatives, creating demand-side uncertainty for AWS and potentially redistributing workloads among cloud rivals.
- The constrained supply of top-tier GPUs creates exposure for AI infrastructure costs; continued supply tightness could sustain elevated prices and affect budgeting for large-scale model training programs.