Blackwell GPU demand spikes
Hourly rental prices for NVIDIA Blackwell GPUs jumped to about $4.08, a roughly 48% rise over two months as agentic AI demand grows. (intellectia.ai) At the same time, vendors are validating Blackwell support across edge systems, with Premio adding NVIDIA RTX Pro Blackwell GPU compatibility for industrial and rackmount AI solutions. (natlawreview.com)
Renting an NVIDIA Blackwell graphics processing unit in the cloud now costs about $4.08 an hour, up 48% from $2.75 two months ago as demand for AI agents tightens supply. (techmeme.com) Blackwell is NVIDIA’s newest family of AI chips, built for the heavy math behind training models and running inference, the step where a model answers prompts in real time. NVIDIA says its GB200 NVL72 system links 72 Blackwell graphics processing units and 36 Grace central processing units in one liquid-cooled rack. (nvidia.com) NVIDIA said on February 4, 2025 that CoreWeave became the first cloud provider to make Blackwell generally available with GB200 NVL72 instances. NVIDIA said those systems were aimed at “AI reasoning” workloads, which use multiple model passes and more tokens than simpler chatbot responses. (nvidia.com) That helps explain the price jump. NVIDIA says Blackwell is “the engine behind AI factories for the age of AI reasoning,” and cloud operators have been marketing the chips for agents that plan, call tools, and generate longer outputs than older assistants. (nvidia.com) The supply push is no longer limited to giant data centers. Premio said on April 13, 2026 that it had validated NVIDIA RTX Pro Blackwell graphics processing units across industrial and edge systems, including machine-vision computers, rugged industrial computers, and 1U rackmount edge artificial intelligence servers. (premioinc.com) Premio said the supported lineup runs from the RTX Pro 2000 Blackwell to the RTX Pro 6000 Blackwell Max-Q Workstation Edition. The company said the top configuration offers up to 24,064 CUDA cores, 3,511 artificial intelligence TOPS, and 96 gigabytes of GDDR7 error-correcting memory for on-premises generative artificial intelligence and real-time inference. (premioinc.com) Cloud vendors have also been widening the product stack around Blackwell. CoreWeave said on July 9, 2025 that it was the first cloud platform to make NVIDIA RTX Pro 6000 Blackwell Server Edition instances generally available, alongside its GB200 NVL72 and HGX B200 offerings. (coreweave.com) The result is a market where the newest chips are spreading in two directions at once: hyperscale clusters for model builders and smaller edge boxes for factories, robotics, and local inference. When both markets pull on the same generation of hardware, hourly rental prices become a visible signal of how tight compute has become. (nvidia.com (premioinc.com))
