Blackwell GPU rents spike
Hourly rental rates for Nvidia’s Blackwell GPUs have jumped to about $4.08 per hour, up roughly 48% from two months ago as demand from agentic AI workloads surges. Spot pricing and reports of outages and rationing suggest premium compute remains scarce even as firms invest in more capacity. (intellectia.ai) (alltoc.com)
Renting Nvidia’s newest Blackwell chips now costs about $4.08 an hour on the spot market, up 48% from $2.75 two months ago. (techmeme.com) The increase comes from data in the Ornn Compute Price Index, which tracks cloud rental prices for Nvidia graphics processing units, or GPUs, used to train and run artificial intelligence models. Other reports tied the jump to heavier demand from “agentic” tools that keep models working longer on multistep tasks. (techmeme.com) (gurufocus.com) A GPU is the specialized processor that does the matrix math behind modern artificial intelligence, and Blackwell is Nvidia’s newest data-center generation after Hopper. Nvidia introduced the Blackwell platform on March 18, 2024, saying it was built to run trillion-parameter models with lower operating cost and energy use than the prior generation. (nvidianews.nvidia.com) The price jump shows that buying more servers has not erased the bottleneck in premium compute. Reports circulating with the Ornn data said cloud providers were rationing access, extending commitments, or warning users about limited availability even as more Blackwell systems come online. (fxbus.com) (the-decoder.com) That squeeze is showing up in products people actually use. Anthropic’s public status page shows the Claude application at 98.4% uptime over the last 60 days and the Claude application programming interface, or API, at 98.73%, below the near-perfect reliability large business customers usually expect. (anthropic.statuspage.io) OpenAI has also said it will discontinue the Sora web and app experience on April 26, 2026, and end the Sora application programming interface on September 24, 2026. OpenAI’s help page does not cite compute shortages as the reason, but the shutdown landed in the middle of a broader scramble for top-tier graphics processors. (help.openai.com) Blackwell’s appeal is straightforward: faster chips can serve more users or bigger models from the same rack, so companies will pay more per hour if the math still works. Nvidia said at launch that Blackwell added new tensor cores, faster links between chips, and systems aimed at real-time generative artificial intelligence workloads. (nvidianews.nvidia.com) For now, the market signal is simple: the most coveted artificial intelligence chips are getting more expensive to rent, not cheaper. As long as spot prices keep rising above $4 an hour, cloud customers will keep treating Blackwell capacity as scarce inventory rather than commodity compute. (techmeme.com) (the-decoder.com)
