Fluence Report Reveals Hyperscalers Charge Nearly Twice the GPU-Cloud Price

A recent report titled The AI Compute Race by Fluence has uncovered that hyperscalers are charging approximately 1.98 times the median GPU-cloud price for GPU compute resources. This finding is particularly relevant for companies outside of major buyers like frontier labs and hyperscalers.

The report indicates that the pricing varies significantly across different models. For instance, the H100 model is priced at up to 2.8 times the median, while the B200 model is at 1.4 times. The H200 model's hourly rates range from $4.08 to $10.60, depending on the provider.

Commitment lengths also play a crucial role in pricing. A one-year reservation can reduce the median price by 21%, while a three-year commitment can lead to a 48% reduction. Spot capacity is available at 53% of the on-demand rates for workloads that can tolerate interruptions.

Despite expectations of falling GPU prices, the report reveals that 17 out of 22 tracked GPU models have increased in cost over the past year. The median rental price for the H100 has risen by 13% to 30%, depending on the timeframe analyzed.

Fluence attributes the rising prices to a shift in demand towards open models, with the share of Chinese models among companies in the US, Europe, and the Global South increasing significantly. The report also notes that the largest hyperscalers are set to guide capital spending of $720–740 billion in 2026.

In addition, the GPU-hour is evolving into a financial asset, with exchanges like CME and ICE announcing GPU futures. Lenders are also beginning to offer loans against GPUs, indicating a growing financialization of this resource.

To help buyers manage costs, the report suggests several strategies: comparing chip prices across different providers, committing to longer terms when workloads are stable, matching GPU generations to specific models, and exploring options beyond hyperscalers, as much of the newest capacity is already contracted.