Amazon Web Services (AWS) is raising prices for its graphic processing unit (GPU) reservation service for the second time this year, with service costs set to jump by around 20%.
The EC2 Capacity Blocks for ML service lets users reserve GPU compute capacity across Nvidia and AWS custom silicon up to eight weeks in advance. Enterprises can also share Capacity Blocks across multiple AWS accounts.
After prices for the service rose around 15% at the turn of the year, Capacity Block reservation costs are set to rise again with the hyperscaler confirming price changes, effective July 1.
For example, an H200-based p5e.48xlarge instance tcost around $34.61 per hour before January. After July 1, the instance will cost around $39 for most global regions, a figure that jumps to $49.74 for users in the U.S. West.
The hyperscaler said prices are “updated periodically based on supply and demand.”
Pricing in the compute space is steadily rising amid component bottlenecks like the ongoing memory shortage. And yet the insatiable demand for compute to power AI workloads shows no signs of slowing, with Gartner forecasting worldwide IT spending to reach $6.31 trillion in 2026, up 13.5% from 2025 figures.
Demand might be high, but there is a dearth of reliable supplies of GPUs. Hyperscalers like AWS have locked down the bulk of production for Nvidia and other chipmakers, leading to multibillion-dollar forward orders – and leaving enterprise buyers to face longer lead times or force themselves into bed with hyperscalers just to access compute.
AWS said in March it plans to deploy more than one million Nvidia GPUs as it looks to continue building out its AI infrastructure, with CEO Matt Garman claiming a month prior that it was still running six-year-old Nvidia A100 servers owing to “so much more demand than supply.”
Comments