Amazon's AWS Hikes AI Cloud Prices 20% After January Increase

Amazon Web Services is raising prices on several key AI cloud offerings by 20%, the latest sign that memory chip shortages are driving up costs across the technology sector. The move follows a 15% price increase in January.

The price hikes apply to services that rely heavily on high-bandwidth memory chips, a component that has been in short supply as demand for AI training and inference workloads surges. AWS is the largest cloud provider by market share, so the increases will ripple through the startup ecosystem and enterprise budgets alike.

Amazon did not specify which services are affected. The increases target compute instances that use Nvidia's H100 and upcoming Blackwell GPUs, which require HBM3 memory modules. Those chips have been constrained for months, pushing up spot prices and forcing cloud providers to ration capacity.

The January increase was attributed to rising energy costs and infrastructure investment. The new round is explicitly tied to hardware supply, a shift in messaging that suggests the chip shortage is deeper than previously acknowledged.

For companies building on AWS, the math changes. A startup running a large language model training job that cost $100,000 a month in December now faces a bill of $138,000 after the two increases. That kind of jump can break a burn-rate model or force a pivot to cheaper alternatives.

Google Cloud and Microsoft Azure have not announced similar increases. AWS's pricing power reflects its dominant position. Smaller providers like CoreWeave and Lambda Labs, which specialize in GPU rentals, have already raised prices multiple times this year.

The chip shortage traces back to Samsung and SK Hynix, the two main HBM3 suppliers. Both have struggled to ramp production fast enough to meet AI demand. Nvidia's GPUs are the bottleneck for most AI workloads. The memory modules that feed them are the bottleneck within that bottleneck.

Amazon's move signals that the cloud pricing cycle has not peaked. If chip supply does not improve in the second half of the year, another round of increases is likely. For now, the cost of AI compute is going up, and the cloud giants are passing it through.