Blackwell GPU rents spike
Spot rental prices for Nvidia’s new Blackwell GPUs have risen sharply in recent months, with one industry index showing the hourly cost up about 48% from two months ago. The Ornn Compute Price Index put an hour on a Blackwell chip at $4.08, up from $2.75, a move market write‑ups blame on rising demand for agentic AI workloads. The price jump is being cited across tech press as a sign compute is getting materially more expensive for training and inference. (the-decoder.com)
Renting Nvidia’s newest Blackwell graphics processors on the spot market now costs about $4.08 an hour, up from $2.75 two months ago. (techmeme.com) The price move comes from the Ornn Compute Price Index, which tracks cloud rental markets for Nvidia chips. Techmeme’s roundup of the Wall Street Journal report pegged the Blackwell jump at 48% as of April 13, 2026. (techmeme.com) A graphics processing unit, or GPU, is the chip that does the heavy lifting for artificial intelligence systems, both when companies train models and when users run them. Nvidia’s Blackwell line is its latest data-center generation, sold in systems such as the B200 and GB200 for large training and inference jobs. (nvidia.com) Nvidia says a single GB200 NVL72 rack links 72 Blackwell GPUs and 36 Grace central processors, with 130 terabytes per second of NVLink bandwidth inside the rack. That design is aimed at trillion-parameter model training and real-time inference, which is why Blackwell capacity is being chased so aggressively. (nvidia.com) The immediate pressure is coming from “agentic” artificial intelligence workloads, the industry term for systems that take multi-step actions instead of answering one prompt at a time. The Decoder, citing the Wall Street Journal, reported that this demand is pushing up rental prices across Nvidia’s lineup, not just Blackwell. (the-decoder.com) The same supply squeeze is showing up in service reliability. The Decoder reported on April 13 that Anthropic’s application programming interface availability was 98.95% over 90 days, below the 99.99% level many enterprise software customers expect. (the-decoder.com) Companies are also reshuffling products to free up compute. OpenAI’s help center says the Sora web and app experience will be discontinued on April 26, 2026, and the Sora application programming interface will end on September 24, 2026. (help.openai.com) Spot prices are not the same as long-term contract prices, and public cloud listings can vary widely by provider, region, and minimum cluster size. One market tracker published April 12 showed Nvidia B200 prices across providers ranging from about $2.25 an hour at the low end to materially higher rates elsewhere. (datastorage.com) For developers and companies buying compute by the hour, the signal is simple: the newest Nvidia capacity got more expensive in early 2026, even before demand for bigger artificial intelligence systems eased. (the-decoder.com)
