AWS raises GPU instance prices 20% from July 1 the second hike this year as AI compute demand peaks
Category: AI Cost
What it is
AWS has increased hourly rates on its EC2 Capacity Blocks for ML – covering Nvidia
Blackwell, H100, H200, and other high-demand GPU instances – by approximately
20%, effective July 1. This is the second price increase in 2026 after a 15% hike in
January. AWS revenues grew 28% YoY to $37.6B in Q1 2026, and the company has
committed ~$200B in AI infrastructure capex this year. No pricing justification was
provided beyond supply-demand mechanics.
Why it Matters for Enterprises
Reserved GPU compute is getting materially more expensive, and the trend may
spread to Azure and Google Cloud. Enterprises training or fine-tuning LLMs should
audit their reservation strategies and explore spot instances, open-weight models, or
alternative infrastructure before costs compound further.
