Amazon AWS Raises GPU Instance Prices 20% on July 1

What happened

AWS will raise hourly reservation rates for its machine learning GPU instances by approximately 20% starting July 1, 2026, according to an update posted to the company's official documentation page. The new rates apply to EC2 Capacity Blocks across AWS's most powerful Nvidia-powered instance families: P6-B300 instances will cost $14.04 per hour, P6-B200 instances $12.355 per hour, and P5 instances in US regions $5.191 per hour. Other instance families including P5e, P5en, and P4de also saw rate adjustments, while all other EC2 pricing remains unchanged. This marks the second price increase for the same GPU capacity in 2026, following a 15% hike implemented in January. The cumulative effect of both increases brings the total pricing adjustment to approximately 38% from baseline rates. AWS attributed the changes to supply and demand dynamics in the high-bandwidth memory and GPU markets. EC2 Capacity Blocks for ML are reserved-capacity products that allow enterprises to secure scarce GPU instances on future dates for time-bound workloads such as large-scale model training, offering guaranteed availability at a premium over spot-market rates.

Why it matters

The pricing action signals a fundamental shift in cloud economics and directly impacts the cost structure for companies running AI workloads. AWS revenue climbed 28% year-over-year to $37.6 billion in the first quarter of 2026, the cloud unit's fastest growth rate in more than three years, demonstrating strong demand that gives AWS considerable pricing leverage. For AI startups and enterprises whose financial models assumed stable or declining cloud costs, the 38% cumulative increase in GPU pricing reshapes burn rates and margin calculations. A single P6-B300 instance running continuously for a month now costs roughly $10,250, making large-scale model training significantly more expensive. The price hikes reflect physical constraints in AI infrastructure: tight supply of high-bandwidth memory, which is packaged alongside advanced AI chips, limits GPU production and data center expansion. Amazon has committed roughly $200 billion in capital expenditure in 2026 to AI infrastructure and is set to receive 1 million Nvidia GPU chips by end-2027 under a cloud supply agreement, underscoring how supply-constrained the high-end GPU market remains. The increases also give AWS pricing power because customers have few alternatives when GPU capacity is scarce, allowing the company to pass infrastructure costs directly to users.

Bigger picture

The AWS price increases mark a departure from the decade-long trend of cloud price deflation that defined the industry through mid-2025. As the world's largest cloud provider, AWS pricing decisions often set industry benchmarks, and similar moves by Microsoft Azure or Google Cloud would confirm that the era of cheap cloud compute for AI workloads has ended. The pricing environment also creates openings for alternatives: decentralized compute platforms have been marketing GPU rates 60–90% lower than AWS, and Google Cloud has been actively promoting its TPU-based instances as cost-competitive options. The tight supply dynamics have propelled memory chip makers such as Micron and SK Hynix to record valuations, reflecting investor expectations that AI-driven demand will keep the market tight and prices elevated for years. The constraint extends beyond AWS: tight memory chip supply limits how many GPUs can be produced, which in turn limits how many data centers can be built industry-wide. This creates a broader bottleneck in AI infrastructure expansion, affecting the entire cloud computing and AI development ecosystem.

What to watch

Monitor whether Microsoft Azure and Google Cloud follow with similar GPU pricing increases, which would confirm industry-wide margin pressure and supply constraints. Track AWS customer migration patterns to alternative platforms, including decentralized compute networks and rival cloud providers offering TPU or lower-cost GPU options. Watch for updates on Nvidia GPU supply agreements and high-bandwidth memory production capacity, as these will determine whether pricing pressure persists or eases. Observe whether AI startups adjust their business models or fundraising strategies in response to higher infrastructure costs, and whether venture capital valuations reflect the new cost structure. Pay attention to AWS's capital expenditure execution and data center expansion pace, as delays or accelerations will signal shifts in the supply-demand balance for AI compute capacity.