Nvidia's Agentic AI: Supercharging Scientific Supercomputing

Nvidia is pitching agentic AI for scientific supercomputing. The company highlighted these workflows and new supercomputing platforms at ISC High Performance 2026. Dion Harris, Nvidia's senior director of HPC and AI Factory Solutions, states we are seeing a massive inflection point with agentic AI. He describes agents that can call simulators, surrogate models, and tools within unified workflows. Nvidia's press materials also announce the Vera Rubin rack-scale platform. This platform delivers 7 Exaflops of AI for Science and 5 Petaflops of native FP64. A separate blog details Los Alamos National Laboratory systems called Mission, Vision, and Veritas. These systems are built with HPE and Nvidia hardware, including about 2,300 and 1,150 standalone Vera CPUs in planned configurations. New software components like ALCHEMI, DAQIRI, and cuPhoton are also part of Nvidia's scientific stack. This development could transform how complex scientific tasks are executed, moving AI from a question-answering tool to an autonomous system.