GPU rental prices spike
Hourly rental prices for Nvidia Blackwell GPUs climbed to about $4.08, up from $2.75 two months earlier according to a compute‑pricing index. (alltoc.com) Reporting attributes the roughly 48% rise to surging demand from agentic AI and other intensive inference workloads. (intellectia.ai)
Renting Nvidia’s newest Blackwell graphics processors now costs about $4.08 an hour, up from $2.75 two months earlier, according to the Ornn Compute Price Index. (techmeme.com) That is a roughly 48% jump in eight weeks, based on the Wall Street Journal report cited by Techmeme and Bens Bites on April 13, 2026. (news.bensbites.com) A graphics processor, or GPU, is the chip cloud companies rent out to run artificial intelligence systems, and Blackwell is Nvidia’s latest data-center line. Nvidia says Blackwell is built for “AI factories” and is now in full production. (nvidia.com) Blackwell is tuned for inference, the stage where a trained model answers prompts, generates code, or carries out multi-step tasks for users. Nvidia says its Blackwell architecture adds a second-generation Transformer Engine with 4-bit floating point, or FP4, math to raise throughput while fitting larger models into memory. (nvidia.com) That matters because newer “agentic” systems do more work per request than a basic chatbot. Nvidia said in a February 2025 post that reasoning systems generate many additional tokens and need more memory, compute, and high-speed connections to respond in real time. (blogs.nvidia.com) Nvidia’s flagship GB200 NVL72 system links 72 Blackwell GPUs and 36 Grace central processors in one liquid-cooled rack. Nvidia says that setup can deliver 30 times faster real-time inference for trillion-parameter large language models than the prior Hopper-based generation. (nvidia.com) Cloud supply is still catching up. CoreWeave said on February 4, 2025 that it was the first cloud provider to make Nvidia GB200 NVL72 instances generally available, and Nvidia later said CoreWeave customers were gaining access to thousands of Blackwell GPUs. (coreweave.com) (blogs.nvidia.com) Ornn, the company behind the pricing index, said on April 2, 2026 that its compute benchmark had been added to Bloomberg Terminal. The company said the index tracks GPU-hour pricing across cloud and on-premise markets. (prnewswire.com) The price spike shows that the bottleneck in artificial intelligence has shifted from training one giant model to serving heavy usage after launch. For companies building agents, coding tools, or search products, the meter is now running at the inference stage. (news.bensbites.com)
