Blackwell GPU rent spikes

Renting Nvidia’s newest Blackwell chips now costs about $4.08 an hour, according to pricing trackers that follow cloud GPU markets. (techmeme.com) The figure came from the Ornn Compute Price Index and was cited Monday by The Wall Street Journal via Techmeme and other aggregators. Ornn’s index said the same Blackwell rental rate was about $2.75 two months earlier, a jump of roughly 48%. (techmeme.com) (news.bensbites.com) A graphics processing unit, or GPU, is the chip that does the heavy math for training and running artificial intelligence models in the cloud. Nvidia introduced the Blackwell platform in March 2024 as its next data-center generation after Hopper, with the B200 aimed at large language model training and inference. (nvidianews.nvidia.com) Nvidia said Blackwell was built to run trillion-parameter models with lower cost and energy use than the prior generation. The B200 configuration Nvidia unveiled carries 192 gigabytes of high-bandwidth memory, the fast on-package memory that keeps large models fed with data. (nvidianews.nvidia.com) The reported trigger for the latest price move was demand for “agentic” artificial intelligence workloads, a term companies use for software agents that can carry out multistep tasks with less human prompting. Microsoft Research described agentic artificial intelligence as systems that act more like collaborators, while Cloudflare defines an artificial intelligence agent as software that can pursue goals autonomously. (microsoft.com) (cloudflare.com) That matters in the rental market because inference demand can rise quickly when companies move from demos to products that run continuously for users. The Wall Street Journal summary circulating Monday said some artificial intelligence companies were already rationing offerings and products as compute tightened. (news.bensbites.com) The Blackwell move also fits a broader pattern in leased AI compute. SemiAnalysis said last week that one-year Nvidia H100 rental pricing had climbed from about $1.70 per GPU hour in October 2025 to roughly $2.35 by March 2026, while on-demand rental capacity was sold out across GPU types. (semianalysis.com) (sechub.in) Ornn has been trying to position its index as a benchmark for this market, saying on April 2 that the Ornn Compute Price Index had been added to the Bloomberg Terminal. That gives investors and infrastructure buyers a standardized reference point for spot and contract GPU pricing, even though private cloud deals can still vary by term length, volume, and region. (morningstar.com) The immediate question is whether more Blackwell supply reaches cloud providers fast enough to cool the market. For now, the clearest signal from Monday’s pricing data is that frontier artificial intelligence compute is still getting more expensive, not less. (techmeme.com)