Fluence says hyperscalers charge 2x median GPU cloud prices

Fluence released The AI Compute Race, a pricing study of GPU cloud compute for buyers outside frontier labs and hyperscalers. Key findings: provider type and commitment length drive price dispersion. Hyperscalers list H100, H200 and B200 at 1.98 times the GPU-cloud median, ranging from 1.4 times on B200 to 2.8 times on H100. The same H200 lists from $4.08 to $10.60 per hour. A one-year reservation cuts the median price by 21 percent and a three-year term by 48 percent versus on-demand. Spot capacity lists at 53 percent of on-demand for interruptible jobs.

Market trends and price dynamics

• Price momentum: 17 of 22 tracked GPU models rose in the past year according to GetDeploying. Median H100 rental rates increased 13 to 30 percent depending on the measurement window.
• Demand shift to open models: OpenRouter token share for Chinese models among companies in the US, Europe and the Global South rose from 6–13 percent to 57–67 percent in under eight months, cited by CNBC. Open weights enable serving near-frontier models on owned hardware, which shifts demand toward securing GPUs.
• Supply concentration: The four largest hyperscalers guided $720–740 billion of 2026 capital spending. The four largest disclosed neocloud deals by AI labs and Meta total about $233 billion, three listed as “up to,” with significant capacity yet to be delivered.
• Financialization of compute: CME, ICE and Nodal Exchange announced GPU futures with different settlement indices. Lenders are financing GPUs: Lambda’s investment-grade loan priced at 6.78 percent and Sharon AI’s unrated loan at 9.95 percent.

Operational implications

• Open-weight models are approaching closed-model performance and gaining adoption. The bottleneck is securing GPU supply.
• Price optimization methods: compare the same chip across provider types, commit when workloads are steady, match GPU generation to the model, and look beyond hyperscalers because significant neocloud capacity is pre-contracted to labs years in advance.

Scope and collaborators

The report covers GPU pricing across H100, H200 and B200 and was produced with research partners NEAR AI, io.net, Boundless, Phala, Gensyn, Fortytwo, Theta, Compute Labs and GAIB.

About Fluence

Fluence Cloud is a GPU cloud provider sourcing capacity from independent data centers. Founded in 2017 by Dmitry Kurinskiy, Evgeny Ponomarev and Tom Trowbridge, it has raised $14 million from 1KX, Multicoin, Distributed Global, Blockchange, Tiger Global and Protocol Labs.