NVIDIA's Vera Rubin NVL72 Goes into Production, Boosts Throughput 10x per Megawatt

NVIDIA announced that its new AI computing platform, Vera Rubin NVL72, has entered mass production and is already running at partners like CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. To meet global demand, NVIDIA has built over 350 factories across 30 countries, creating its largest-ever rack-level supply chain.

CoreWeave powered on and verified the first Vera Rubin NVL72 system in early June. Testing used the DeepSeek R1 reasoning model, showing that Vera Rubin NVL72 delivers 10x more tokens per megawatt compared to the GB200 NVL72, with the same interactivity goals. This performance boost will directly accelerate the deployment of agentic AI applications.