In short: Vera Rubin is the NVIDIA platform that includes the Rubin HPC GPU and the Vera CPU. At SC26 in Chicago in November 2026, NVIDIA presented them as the base for science, engineering and AI agents.
Why an agent needs a CPU
A chat answers you. An agent plans, runs code, looks at the result and tries again. Between two steps of the model, something has to run the tool and gather the results, and that is the CPU's job. NVIDIA's Ian Buck puts it this way: agents need a CPU that can run, evaluate and perform the operations. That is why NVIDIA is expanding its platform with Vera CPUs. DKRZ, the German Climate Computing Center, has already picked Vera for its HLRE-5 supercomputer, expected in early 2028.
The chip for science
Rubin HPC is a GPU built for science, engineering, analytics and AI. According to NVIDIA, it supports precisions from FP64 to FP4, adds sparse-data acceleration and connects over seventh-generation NVLink. It has more than 50,000 CUDA cores, about three times as many as Hopper. It is expected in 2028, so for now this is a plan, not a product you can order.
A quantum part is tied to the platform too. NVIDIA, IBM and Fermilab announced a plan for a quantum co-design platform to support error-correction research. According to NVIDIA, a new CUDA-Q algorithm cuts the computing work for error correction from 12 weeks to 45 minutes. The numbers are NVIDIA's.
Primary sources
NVIDIA: Ian Buck at SC26The visual is generated code-art, no third-party images.