Hyperscalers List GPUs at Near... Note
NewsBTC

Hyperscalers List GPUs at Nearly 2× the GPU-Cloud Price, Fluence Research Finds

Fluence's AI Compute Race report reveals GPU compute pricing disparities for companies outside of frontier labs and hyperscalers. Hyperscalers list GPUs at nearly double the median price found elsewhere, with significant variation based on the specific GPU model. Longer commitment periods drastically reduce GPU-hour costs, with three-year terms offering a median price cut of 48%. Spot capacity provides a cheaper option for interruptible workloads, costing 53% of on-demand rates. Despite expectations, 17 out of 22 tracked GPU models have increased in price over the past year. This rise is driven by the rapid adoption of open-weight AI models. These models allow any company with GPUs to serve advanced AI on their own hardware, increasing demand. The largest buyers have secured significant future GPU supply through massive capital spending and large contracts with neocloud providers. Consequently, the GPU-hour is emerging as a financial asset, with futures markets and lending options now available. The report advises companies to pay less for compute by comparing providers, committing to steady workloads, matching GPUs to models, and looking beyond hyperscalers for capacity.
CdXz5zHNQW_OKjltxhzzC.jpeg