Amazon Web Services (AWS) will increase pricing for its Amazon EC2 Capacity Blocks for Machine Learning by approximately 20% beginning July 1, 2026, marking the second price increase this year for the reservation service that provides guaranteed access to GPU infrastructure for AI training and inference workloads. AWS says Capacity Block reservation prices are updated periodically based on supply and demand, with customers paying the prevailing rate when the reservation is purchased, even if the reserved capacity starts after a future price change.
The updated hourly reservation rates per accelerator include P6-B300 at $14.04 in all commercial AWS Regions except GovCloud, P6-B200 at $12.355, P5 at $5.191 in U.S. Regions and $4.72 outside the U.S., P5e at $5.97, P5en at $6.865in U.S. Regions and $6.241 elsewhere, and P4de at $2.214 in U.S. Regions. AWS said all other Capacity Block pricing remains unchanged. Reservation fees are charged upfront when the Capacity Block is booked, while operating system charges are billed separately based on the operating system used and the actual runtime of the instances.
Amazon introduced EC2 Capacity Blocks to provide customers with guaranteed access to scarce AI accelerator capacity without requiring long-term Reserved Instance commitments. The service supports scheduled reservations for NVIDIA Blackwell, Hopper and Ampere GPU platforms, along with AWS Trainium instances, enabling organizations to reserve GPU infrastructure for planned AI training and inference workloads. AWS previously increased Capacity Block pricing in January 2026, and the latest adjustment reflects continued demand for dedicated AI compute resources.
AWS EC2 Capacity Blocks for AI Pricing Update Effective July 1, 2026 | |
| Service | Amazon EC2 Capacity Blocks for Machine Learning |
| Purpose | Reserve guaranteed GPU accelerator capacity in advance for AI training and inference workloads. |
| Effective Date | July 1, 2026 |
| Price Increase | ≈20% vs. current Capacity Block pricing (second increase in 2026) |
| New Rates | P6-B300 $14.04/hr P6-B200 $12.355/hr P5 $5.191 (US) / $4.72 (Intl.) P5e $5.97 P5en $6.865 (US) / $6.241 (Intl.) P4de $2.214 (US) |
| Supported AI Hardware | NVIDIA Blackwell, Hopper, Ampere GPUs, plus AWS Trainium accelerators. |
| Pricing Model | Reservation fee paid upfront; operating system charges billed separately while instances run. |
| Pricing Basis | AWS updates Capacity Block prices periodically based on supply and demand. |
| Previous Increase | January 2026 (approximately 15% for selected Capacity Blocks). |
| AWS Source | https://aws.amazon.com/ec2/capacityblocks/pricing/ |
• Reservation fees are paid upfront; operating system charges are billed separately.
“Amazon EC2 Capacity Blocks for ML reservation prices are updated periodically based on supply and demand.”
Source: AWS – Amazon EC2 Capacity Blocks for ML Pricing
https://aws.amazon.com/ec2/capacityblocks/pricing/
🌐 Analysis: The latest increase underscores the sustained demand for reserved AI accelerator capacity as enterprises and hyperscalers continue deploying large-scale GPU clusters for foundation model training and inference. Rather than relying solely on on-demand availability, many organizations now reserve compute capacity weeks in advance to ensure access to Blackwell- and Hopper-based infrastructure during critical AI development cycles.
AWS’s pricing strategy also highlights the growing importance of reservation-based GPU services across the cloud industry. As AI infrastructure demand continues to outpace supply in many regions, guaranteed-capacity offerings have become an increasingly valuable premium service, giving customers predictable access while allowing cloud providers to align pricing with market demand.
🌐 We’re tracking the latest developments in AI infrastructure. Follow our ongoing coverage at:
https://convergedigest.com/category/ai-infrastructure/
🌐 We’re launching the “Data Center Networking for AI” series on NextGenInfra.io and inviting companies building real solutions—silicon, optics, fabrics, switches, software, orchestration—to share their views on video and in our expert report. To get involved, send a note to [email protected] or [email protected].
