TL;DR
Over a weekend in January 2026, Amazon Web Services increased prices for EC2 Capacity Blocks for machine learning by roughly 15 percent. The change affects large GPU instances and may force conversations between AWS and enterprise customers with volume-based contracts.
What happened
AWS implemented a price increase for EC2 Capacity Blocks for ML, with headline hikes of roughly 15 percent across several large GPU instance types. Examples cited include the p5e.48xlarge rising from $34.61 to $39.80 per hour in most regions and the p5en.48xlarge moving from $36.18 to $41.61; in US West (Northern California) p5e rates increased from $43.26 to $49.75. The company had a notice on its pricing page saying prices were scheduled to update in January 2026 but did not specify whether rates would go up or down. An Amazon spokesperson said pricing for these Capacity Blocks varies with supply and demand patterns and that this adjustment reflects expected supply/demand for the quarter. Capacity Blocks are a reserved, time-bound way to guarantee GPU capacity for ML training by paying up front at a locked rate, a product commonly used by organizations running large training jobs.
Why it matters
- Enterprise costs could rise even for customers with negotiated discounts, since those discounts are often tied to public list prices.
- The move breaks a long-standing industry expectation that cloud prices trend downward, potentially making future increases easier to justify.
- Competitors courting ML workloads may use the change as a sales argument, affecting enterprise procurement conversations.
- The adjustment highlights pressure on GPU supply and could foreshadow similar changes for other constrained resources.
Key facts
- AWS raised EC2 Capacity Block prices for ML by about 15% over the weekend.
- p5e.48xlarge increased from $34.61 to $39.80 per hour in most regions.
- p5en.48xlarge moved from $36.18 to $41.61 per hour in most regions.
- In US West (Northern California), p5e went from $43.26 to $49.75 per hour.
- AWS's pricing page had stated prices were scheduled to be updated in January 2026 but did not indicate direction.
- An Amazon spokesperson attributed the change to supply and demand patterns for EC2 Capacity Blocks for ML.
- Capacity Blocks let customers reserve guaranteed GPU capacity for fixed time windows and pay up front at a locked rate.
- Seven months earlier AWS had touted up-to-45% reductions for some GPU pricing categories, but those applied to On-Demand and Savings Plans, not Capacity Blocks.
- Historically, AWS has more often altered pricing models or dimensions than issued straight list-price increases; straight increases have tended to be tied to regulatory actions in the past.
What to watch next
- Whether AWS applies similar list-price increases to other constrained resources such as memory-optimized instances or data transfer — not confirmed in the source.
- How Microsoft Azure and Google Cloud Platform respond in pricing or sales messaging to the GPU price change — not confirmed in the source.
- Conversations between AWS account teams and customers with Enterprise Discount Programs over the next weeks as billing impacts are assessed — likely but not confirmed in the source.
Quick glossary
- EC2 Capacity Blocks for ML: A reservation product that lets customers lock in guaranteed GPU instance capacity for a defined time window by paying up front.
- p5e.48xlarge / p5en.48xlarge: Large EC2 instance families that bundle multiple GPUs for heavy machine learning training workloads.
- On-Demand: Cloud billing model where compute resources are paid for by the hour or second with no long-term commitment.
- Savings Plans: A pricing program that provides lower compute rates in exchange for a commitment to a certain level of spend over a period.
- Enterprise Discount Program (EDP): A negotiated commercial agreement that typically provides a customer-specific discount off public cloud list prices.
Reader FAQ
Did AWS raise GPU prices?
Yes. The company increased EC2 Capacity Block rates for certain GPU instances by about 15%.
Which instance types were cited?
The article cites p5e.48xlarge and p5en.48xlarge, with region-specific rates including a larger jump in US West (N. California).
Why did AWS make the change?
An Amazon spokesperson said the price adjustment reflects expected supply and demand patterns for the quarter.
Will negotiated enterprise discounts protect customers?
Not necessarily; many programs discount from public list prices, so higher public prices can raise absolute costs even if percentage discounts hold.
Is this the start of broader cloud price increases?
not confirmed in the source
PAAS + IAAS AWS raises GPU prices 15% on a Saturday, hopes you weren't paying attention An anomaly or the beginning of a new trend? My bet's on the latter…
Sources
- AWS raises GPU prices 15% on a Saturday, hopes you weren't paying attention
- AWS raises GPU prices 15% on a Saturday
- Announcing up to 45% price reduction for Amazon EC2 …
- Amazon EC2 Capacity Blocks for ML Pricing
Related posts
- Why Refactoring Should Be Done In-Flow, Not As Backlog Stories
- Refactoring — Why It Shouldn’t Be Put on the Backlog as a Story
- How Twitch Managed Standards Across a Million Lines of TypeScript