June 30, 2026, (Inside AI) — Amazon Web Services will raise prices for reserved AI GPU capacity by 20% on July 1, marking the second hike this year as demand for artificial intelligence compute continues to outstrip supply.
The increase targets Capacity Blocks for ML, a service that lets businesses book NVIDIA GPUs in advance for large training and fine-tuning jobs. It follows a 15% jump in January, making reserved AI compute 35% more expensive in just six months.
AWS confirmed the change after Business Insider first reported it. The company said pricing is reviewed periodically based on supply and demand. Other GPU buying options, including on-demand and spot instances, remain at current prices. Amazon’s own Trainium AI chips are also unaffected.
The root cause is a shortage of high-bandwidth memory (HBM), a critical component paired with advanced AI processors. Memory output constraints are now capping GPU production, slowing the global build-out of AI data centers.
Peter Berezin, chief economist at BCA Research, explained the bottleneck:
“As there is a limit to how much memory can be produced, then there is a limit to how many GPUs can be produced, which means that there’s a limit to how many data centers can be built.”
The memory squeeze is not limited to cloud GPUs. This week, Apple raised prices across its Macs and iPads, citing memory costs, while Xbox consoles also saw hikes. What began as a software scaling challenge has hardened into a hardware bottleneck.
The higher prices will hit startups and large firms alike. Many depend on AWS to train and run their AI models. As the world’s biggest cloud provider, AWS supports thousands of AI services, so even a narrow price rise can ripple across the industry.
Analysts say tight supply hands AWS more pricing power. Microsoft, Google, and Oracle gain the same leverage as alternatives shrink. For AI builders, reliable GPU capacity keeps getting costlier, with no quick fix in sight as memory fabrication capacity remains limited.
