Amazon Web Services has implemented a roughly 15% price increase on some of its most powerful GPU-based EC2 Capacity Block instances, a move that will ripple through enterprise machine learning budgets and may even cause concerns about the future direction of cloud pricing.
The changes affect the p5e.48xlarge and p5en.48xlarge instances, both of which are built around eight NVIDIA H200 GPUs and are typically used for large-scale, business-critical machine learning training. Across most regions, hourly pricing for the p5e.48xlarge has risen from about $34.61 to $39.80, while the p5en.48xlarge has increased from roughly $36.18 to $41.61. Customers in the US West region have seen sharper jumps, with p5e rates climbing from $43.26 to $49.75 per hour.
The increases were first identified by industry watchers and later confirmed by AWS pricing updates. While AWS had previously noted that prices for these Capacity Blocks would be updated in January 2026, it did not specify the direction of the change. The result was a price shift that landed without a formal announcement.
Reserved Capacity
Capacity Blocks differ from standard on-demand or spot instances. They allow customers to reserve specific GPU capacity in advance, typically from one day up to several weeks, at a locked-in rate paid upfront. This model is designed for organizations that cannot risk losing GPU access mid-training due to capacity shortages, making it especially popular with enterprises running time-sensitive or high-value AI workloads.
In June 2025, AWS announced significant price reductions, up to 45%, across several GPU instance types, including H100- and H200-based offerings, for on-demand usage and Savings Plans. At the time, AWS emphasized that regular price cuts were a way to pass efficiencies from scale back to customers. Capacity Blocks were excluded from those reductions, but the contrast between last year’s cuts and this year’s increase has not gone unnoticed.
For customers running large fleets of reserved GPU instances, a 15% hike can translate into a substantial increase in monthly cloud spend. Enterprises operating under negotiated Enterprise Discount Programs may also feel the impact, since discounts are typically applied to public list prices. When those prices rise, the absolute cost increases even if the discount percentages remain unchanged.
The GPU Shortage, and Customer Expectations
The broader backdrop to this price boost is an ongoing global shortage of advanced GPUs, driven by surging demand for AI and machine learning workloads. All major cloud providers are competing for limited supply, and guaranteed access has become increasingly valuable. From that perspective, higher prices for reserved capacity reflect the realities of constrained hardware availability.
Still, the move has large significance, in addition to the out-of-pocket expense. AWS has spent years conditioning customers to expect cloud prices to trend downward over time. A visible increase on a flagship AI service challenges that assumption and may prompt enterprises to re-evaluate long-term cost models, diversify providers, or reassess how and when they reserve premium GPU capacity.
For now, AWS has indicated that the next formal pricing review for Capacity Blocks is scheduled for April 2026. Until then, customers planning major machine learning projects will need to factor higher reservation costs into their budgets, and consider whether guaranteed capacity is worth the new premium in an increasingly competitive and resource-constrained AI landscape.
