As artificial intelligence tools get increasingly agentic, Nvidia is setting the pace by designing high-powered systems that keep data centers humming.
Its signature new product, the GB300 NVL72 platform, is a two-ton system that packages 72 of the company's Blackwell Ultra GPUs and 36 Nvidia Grace CPUs in a liquid-cooled rack, combining the parallel compute capacity of GPUs with the orchestration power of CPUs to enable the heaviest AI workloads.
The platform is “the most ambitious system we've ever built,” says Dion Harris, senior director of AI infrastructure at Nvidia, which reached a $4.5 trillion market cap in 2025. Integrating innovations in networking, liquid cooling, power management, and software, the GB300 NVL72 functions as a plug-and-play, rack-scale AI supercomputer. “The engineering challenges were extraordinary,” Harris says, noting that the company…