The AI Statistics Race report the prices determined by the GPU outside of the biggest sellers: for three years the average price has been reduced by 48%, while 17 of the 22 GPU models have increased in price over the past year.
DALLAS, Texas – November 13, 2026 – Airplane has been published The AI Statistics Racea report that calculates the costs of cloud GPUs for each company that is not a local lab or a hyperscaler. Created with research partners YOU ARE CLOSE, io.net, Endless, Phala, Re-examination, Forty-two, That’s it, Computing Lab a GAIB. His main search: where the company sells the GPU-hour and determines the price. Across the H100, H200 and B200 models, the hyperscalers listed GPUs at 1.98 times the GPU-cloud average, ranging from 1.4 times on the B200 to 2.8 times on the H100. That H200 lists from $4.08 to $10.60 an hour.
Due diligence is important in supplier selection. Based on each provider’s bid rate, a one-year reservation cuts the average cost by 21% and a three-year period by 48%. Capacity lines are up to 53% above demand, for work that can be interrupted.
Consumers hoping for a drop in GPU prices should re-examine their budgets. Seventeen of the 22 GPU models were tracked by Download Image more than a year ago, and the average H100 costs 13% to 30% more, depending on the window measured.
The report also explains why prices have risen so quickly:
- The application was moved to open models. The Chinese model’s share of OpenRouter brands among companies in the US, Europe and the Southeast increased from 6-13% to 57-67% in less than eight months, according to CNBC. The open workloads that a company with a GPU offers are a close-up model on the hardware it manages.
- Major buyers shut down new items. The four major hyperscalers lead $720–740 billion of capital expenditure 2026, and the four projects that are expanded by AI and Meta labs and neoclouds around $233 billion, three of them “including” funds and mostly for the capacity that has not yet been released.
- GPU-hours are becoming a financial asset. CME, ICE and Nodal Exchange have announced the GPU future, and it will set it on a different price index, as lenders are already lending money to GPUs: Lambda’s loan is set at 6.78% and Sharon AI, no rating disclosed, at 9.95%.
“Open-heavy models are almost behind the best closed ones, and companies are turning to them. Every company with GPUs can do a close-enough model on the devices it manages. GPUs are the difficult part,” said Evgeny Ponomarev, Founder of Fluence.
The report closes with four ways to pay less for the same calculation: compare the same chip across different suppliers, work while maintaining performance, match the GPU generation to the model, and look beyond hyperscalers, because most of the new neocloud capacity is contracted to laboratories in the coming years.
The full report is available at fluence.ai/compute-race-2026.
About Fluence
Fluence Cloud is a GPU cloud provider with capacity from private data centers (run.ai). Founded in 2017 by Dmitry Kurinskiy, Evgeny Ponomarev and Tom Trowbridge, the company has raised $14 million from investors including 1KX, Multicoin, Distributed Global, Blockchange, Tiger Global and Protocol Labs.
