AWS Quietly Hikes GPU Compute Prices — What It Means for You

Here we go again. Another cloud price increase, slipped out quietly, wrapped in technical nuance, and timed just carefully enough that most […]

Chris Fennell (Chief Operating Officer)

6th January 2026

News

1 min read

Here we go again. Another cloud price increase, slipped out quietly, wrapped in technical nuance, and timed just carefully enough that most customers won’t notice until the bill lands.

This time it’s AWS, nudging up prices on high-end GPU capacity while relying on complexity, fine print, and the sheer noise of the AI gold rush to keep the backlash muted. If this feels familiar, that’s because it is — and it raises uncomfortable questions about where “pay-as-you-go” cloud economics are really heading.

What have AWS done?

Over the first weekend of January 2026, AWS quietly “adjusted” pricing on its EC2 Capacity Blocks for machine learning, pushing costs up by around 15 percent for high-end GPU workloads. A p5e.48xlarge instance — packing eight NVIDIA H200s — now costs closer to $40 an hour instead of $34, with some regions being rewarded with even steeper increases. The p5en.48xlarge wasn’t spared either, because consistency matters when raising prices.

AWS’s latest “update” targets Capacity Blocks — their premium option for reserving guaranteed GPU power for machine learning. Unlike spot or on-demand instances, these blocks can’t be interrupted, making them perfect for companies with big AI budgets and a fondness for predictability.

And, of course, AWS had hinted at changes on its website… just not the tiny detail that prices would actually go up. Some customers will only notice after their next invoice lands in their lap.

But of course they will get some notifications from AWS… with some dry technical overview of affected SKUs.