NVIDIA has announced its next evolution in AI computing with the Blackwell Ultra platform, designed specifically to handle the demanding requirements of AI reasoning, agentic AI, and physical AI applications. Building on the Blackwell architecture introduced last year, the new platform includes the GB300 NVL72 rack-scale solution and the HGX B300 NVL16 system. The GB300 NVL72 delivers 1.5x more AI performance than its predecessor while increasing revenue opportunities for AI factories by 50x compared to Hopper-based systems.

"AI has made a giant leap — reasoning and agentic AI demand orders of magnitude more computing performance," said Jensen Huang, NVIDIA's founder and CEO. "We designed Blackwell Ultra for this moment."

The platform excels at test-time scaling, allowing AI models to access increased compute capacity for step-by-step problem-solving and more complex reasoning tasks. According to Nvidia, the HGX B300 NVL16 provides 11x faster inference on large language models, 7x more compute, and 4x larger memory compared to the previous Hopper generation.

Supporting this hardware is the new NVIDIA Dynamo open-source inference framework, also announced yesterday at GTC, which optimizes reasoning AI services by orchestrating inference communication across thousands of GPUs.

Major cloud providers including AWS, Google Cloud, Microsoft Azure, and Oracle, along with server manufacturers like Cisco, Dell, HPE, Lenovo, and Supermicro, are expected to offer Blackwell Ultra-based products starting in the second half of 2025.