Blackwell GPU Rent Spike

Renting Nvidia’s newest Blackwell graphics processors by the hour has gotten markedly more expensive, with one market index putting the rate at $4.08. (techmeme.com) Techmeme, citing a Wall Street Journal report published April 13, said Ornn’s compute price index showed Blackwell hourly rent up 48% from $2.75 two months earlier. Ornn describes itself as a U.S.-regulated marketplace for graphics-processor futures and swaps tied to data-center capacity. (techmeme.com) (ornn.com) The price jump points to a tighter market for premium inference hardware, the chips rented to run artificial-intelligence models after they are trained. The Wall Street Journal summary linked by Techmeme said demand from “agentic” artificial intelligence workloads was a driver of the increase. (techmeme.com) Inference is the serving side of artificial intelligence: a user asks a model for an answer, and rented chips generate it in real time. Nvidia said in February 2025 that Blackwell systems were being positioned for “reasoning models and agents,” which require more memory, faster links between chips, and more compute to answer each prompt. (blogs.nvidia.com) Blackwell is Nvidia’s latest top-tier graphics-processing architecture for large artificial-intelligence jobs, and cloud providers have only gradually brought it online. Nvidia said CoreWeave made Blackwell generally available in the cloud on February 4, 2025, while Amazon Web Services later said its Blackwell-based P6 and P6e instances had reached general availability. (blogs.nvidia.com) (aws.amazon.com) That matters for pricing because a young supply base can move sharply when demand concentrates in a few sought-after chip types. Ornn’s business pitch is built around that volatility: the company says customers use its products to hedge swings in artificial-intelligence compute costs and data-center capacity risk. (ornn.com) Published cloud price sheets show Blackwell rates already sit in a high band, though they vary by provider and contract terms. A March 21, 2026 pricing roundup listed single Nvidia B200 rentals at $5.98 an hour on RunPod and $6.08 on Lambda, with an eight-chip CoreWeave cluster priced at $68.80 an hour, or $8.60 per chip. (deploybase.ai) Other market trackers show a wide spread even within the same chip family. GetDeploying said on April 12 that it was tracking Blackwell B200 prices across more than 23 providers, and DataStorage said some listings were as low as about $2.25 an hour, underscoring how spot, reserved, and negotiated rates can diverge from an index reading. (getdeploying.com) (datastorage.com) Cloud operators have also been signaling tight conditions. A market report summarizing the Ornn data quoted Vultr Chief Executive J.J. Kardwell saying the compute shortage was “extremely severe,” as providers balance limited supply against a rush of customers building and serving larger artificial-intelligence systems. (fxbus.com) The immediate question is whether more Blackwell capacity reaches the market fast enough to cool rents, or whether agent-style artificial-intelligence products keep bidding them higher. For now, the hourly price of top-end inference hardware is moving in the wrong direction for anyone trying to scale cheaply. (techmeme.com)