Microsoft CEO Satya Nadella shared images from a data center showing that the company had received Nvidia’s first mass-produced Vera Rubin systems on Aug. 21, 2026. Nvidia subsequently confirmed that Vera Rubin had entered full production. The platform is the company’s next-generation rack-scale AI computing system after Blackwell. Its NVL72 configuration combines 72 Rubin GPUs with 36 Vera CPUs. Nvidia says that, compared with GB200 NVL72, Vera Rubin can reduce inference costs per 1 million tokens to about one-tenth and cut the number of GPUs required to train mixture-of-experts models to one-quarter. Nvidia has also said the platform can deliver up to five times faster inference and 3.5 times improved training compared with Blackwell, while CoreWeave has reported a 10-fold increase in AI throughput per megawatt. Microsoft plans specialized deployments at next-generation superfactory sites in Wisconsin and Atlanta and joins Google Cloud, CoreWeave and OpenAI as an early Vera Rubin adopter. Nvidia moved the platform into full production in June 2026, roughly two months before Microsoft’s delivery. The company is positioning Vera Rubin as a building block for trillion-GPU architectures and what it calls "agentic AI factories," while Amazon Web Services has not appeared in early Vera Rubin announcements and may be emphasizing its own Trainium silicon.