← BackJan 6, 2026

AWS Raises EC2 Capacity Block Prices for ML by 15%: What It Means for Enterprise Customers

Amazon Web Services announced a 15% price hike for its EC2 Capacity Block instances used for machine‑learning workloads. The increase, effective June 2024, will affect teams that reserve GPU capacity for long‑term training jobs, prompting questions about future pricing patterns. Competitors may leverage the move in their sales pitches, while customers with Enterprise Discount Programs will need to renegotiate terms to offset the new base rates.

AWS quietly raised prices for its EC2 Capacity Block instances that are devoted to machine‑learning (ML) workloads, marking the first time the company has announced a straight increase to a line item in its public pricing catalog. Beginning on Saturday, June 22, 2024, the p5e.48xlarge and p5en.48xlarge instances—each equipped with 48 NVIDIA H200 accelerator cards—rose from $34.61 to $39.80 per hour and from $36.18 to $41.61 per hour, respectively, in most regions. In U.S. West (N. California) the p5e.48xlarge rate climbed from $43.26 to $49.75 per hour. The AWS pricing page indicated that “current prices are scheduled to be updated in January 2026,” but did not specify the direction, leaving customers to interpret the change without a clear preview. Capacity Blocks are AWS’s dedicated offering for customers who need guaranteed GPU capacity for fixed windows, typically ranging from one day to several weeks. They appeal to organizations that run large‑scale, time‑sensitive training jobs where spot instances risk interruption. Because the reservation model locks in a rate, the price adjustment directly affects budgets that are already measured in millions of dollars. The announcement comes roughly seven months after AWS announced “up to 45 % price reductions” for its on‑demand GPU instances and Savings Plans. Those reductions applied only to the on‑demand and Savings Plan pricing tiers, not to the Capacity Block offering that many customers rely on for consistent performance. According to an email from an AWS spokesperson, “EC2 Capacity Blocks for ML pricing vary based on supply and demand patterns, as described on the product detail page. This price adjustment reflects the supply/demand patterns we expect this quarter.” The company’s rationale underscores a broader resource‑constraint narrative: GPUs have become globally scarce as enterprises adopt generative‑AI and other compute‑intensive workloads. Historically, AWS has rarely implemented direct price increases; the company often changes pricing dimensions or reclassifies services, sometimes positioning the shift as a cost saving for many customers. This method has fostered a perception that cloud prices trend downward over time. The June 2024 hike marks a departure from that legacy, hinting at a potential shift in how AWS manages limited resources. For enterprises with Enterprise Discount Programs (EDPs) or other negotiated agreements, the impact can be substantial. EDPs typically guarantee a certain discount off the published public rate. When the base rate rises, the absolute cost for the same discount percentage increases, effectively eroding the financial benefit of the program unless AWS revises the discount tier. The price increase also provides competitors such as Microsoft Azure and Google Cloud Platform with a narrative lever. Both clouds have been actively courting ML customers, and the headline that “AWS just raised GPU prices by 15 %” can be a persuasive point in sales discussions. Whether those competitors can absorb the demand, especially given their own supply constraints, remains to be seen. Looking beyond GPUs, this development may foreshadow adjustments in other highly contested resource types. Supply bottlenecks for ARM‑based Graviton instances, data‑transfer fees, or memory‑intensive services could see similar pricing experiments. Historically, AWS has used aggressive pricing to spur adoption, but as supply chains tighten, price stability may give way to strategic hikes. In sum, AWS’s straight‑forward price increase for its EC2 Capacity Block ML instances signals a shift in the company’s pricing philosophy. Future changes may become easier to implement as customers and partners adjust to the new baseline. Cloud architects and procurement officers should revisit their cost models, renegotiate discount agreements where possible, and monitor other highly utilized services for similar adjustments. The next months will likely bring further scrutiny: will additional capacity‑related services see price changes, and will competitors mirror this approach? Observers who have followed AWS for two decades will regard the June 22 announcement as a watershed moment, one that challenges the long‑standing assumption that cloud pricing only decreases.