← News

Agents & MCPSources: NVIDIA · 17 Nov 2026

NVIDIA puts AI agents at the center of its SC26 science push

At the SC26 supercomputing conference, NVIDIA announced Nemotron for Science, a Rubin HPC GPU, expanded Vera CPU plans and quantum collaborations with IBM and Fermilab.

In plain words

NVIDIA used its SC26 conference talk to promote AI agents: software that can plan experiments, run code and check results on researchers' own machines. The company announced an open model toolkit for science, a new GPU and CPU plans, plus quantum error-correction work with IBM and Fermilab.

What was announced

The main announcement was Nemotron for Science, which combines the Nemotron 3.5 Super VL open model with a new open chemistry dataset and post-training recipes for life sciences and chemistry. NVIDIA said researchers can run it locally on a DGX Spark, a workstation or their own supercomputer, keeping control of their data and models. As an example, NVIDIA said an agent built with the toolkit at Argonne National Laboratory took a single prompt, planned and tested hypotheses, chose simulations and identified two new binding sites. Buck also cited agents in use elsewhere: one from KTH Royal Institute of Technology that repaired broken GROMACS setups, fixing 81% of test cases in an overnight run on a single DGX Spark, and NVIDIA's Nsight AI Porting Agent, which ported a Weather Research and Forecasting model routine to GPUs with a speedup of more than 5x.

On hardware, NVIDIA announced the Rubin HPC GPU, expected in 2028, supporting precisions from FP64 to FP4 with sparse-data acceleration and seventh-generation NVLink, and said DKRZ, the German Climate Computing Center, selected Vera CPUs for its HLRE-5 supercomputer, expected in early 2028. In quantum computing, NVIDIA, IBM and Fermilab announced plans for a co-design platform pairing DGX B200 systems with IBM quantum computers, and Buck said a JUPITER simulation showed magic state cultivation can cut operations for one logical step tenfold. He also said a new CUDA-Q decoding algorithm, TCBP-JOBS, cuts error-correction research computation from 12 weeks to 45 minutes.

What it does not show

  • All figures and claims come from NVIDIA's own blog and special address; no independent testing or confirmation is reflected in the source.
  • Performance and availability details, such as the 2028 hardware timelines and speedup figures, are stated expectations or vendor-reported results, not verified deployments.
  • The source is a single NVIDIA blog post, so competing or corroborating perspectives are absent.

For security and IT teams

Scope
NVIDIA says Nemotron for Science is an open model that researchers can run locally on a DGX Spark, a workstation or their own supercomputer.

Sources

Source-support review completed 18 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Updates and corrections

No corrections so far.

First published 18 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.