NVIDIA used its special address at SC26 in Chicago on Monday to introduce Nemotron for Science, an open toolkit for building AI agents that plan experiments, run code and check results on a researcher’s own systems. Ian Buck, the company’s vice president of hyperscale and high-performance computing, also covered the Rubin HPC GPU, a wider Vera CPU platform and a quantum collaboration with IBM and Fermilab.

The message was that simulation stays the base of scientific computing, with AI, quantum and now agents working next to it. Here is what was announced and what is still open.

What is Nemotron for Science?

It is a package for research agents, not a single product. According to NVIDIA, it combines three parts:

  • The Nemotron 3.5 Super VL open model
  • A new open chemistry dataset
  • Post-training recipes for life sciences and chemistry

Researchers can run it locally on a DGX Spark, a workstation or their own supercomputer. Buck stressed ownership: “You own your data. You own the model.” At Argonne National Laboratory, he said, an agent built with the toolkit took a single prompt, planned and tested hypotheses, chose which simulations to run and identified two new binding sites.

Which science agents are already running?

Buck pointed to three examples, all described by NVIDIA and not independently tested:

  • GROMACS: an agent from KTH Royal Institute of Technology repairs broken GROMACS setups on its own. In one overnight run on a single DGX Spark it fixed 81% of test cases.
  • SETI Institute: at the Allen Telescope Array, an agent took a one-line prompt to plot a methanol line, then retuned for Doppler shift by itself.
  • Nsight AI Porting Agent: NVIDIA’s own agent ported a routine from the Weather Research and Forecasting (WRF) model to GPUs, speeding it up more than 5x using about 150 million tokens.

Why do agents need CPUs, and who chose Vera?

Buck explained that agents lean on CPUs to run tool calls between model steps: run, evaluate, gather results. DKRZ, the German Climate Computing Center, has selected Vera CPUs for its HLRE-5 supercomputer, expected in early 2028. Buck said it joins Vera systems already announced at the Texas Advanced Computing Center.

What is Rubin HPC?

Rubin HPC is a new GPU in the Vera Rubin platform, built for science, engineering, analytics and AI. Per Buck:

  • Precisions from FP64 to FP4
  • Sparse-data acceleration
  • Seventh-generation NVIDIA NVLink
  • More than 50,000 CUDA cores, about three times as many as Hopper
  • Expected in 2028

What is new in quantum computing?

NVIDIA, IBM and Fermilab announced plans for a quantum computing co-design platform for error-correction and applications research. NVIDIA DGX B200 systems will be paired with IBM quantum computers, including those powered by IBM Quantum Nighthawk. The partners call it a proposed platform, and The Next Web reported that no start date or cost was given.

On the software side, NVIDIA says its new CUDA-Q decoding algorithm, TCBP-JOBS, cuts error-correction research computation from 12 weeks to 45 minutes, and its release claims up to 13,000x faster decoding. Buck said the goal for decoding latency is to fall from about 300 microseconds to 10. He also said a simulation on Jülich’s JUPITER supercomputer showed that magic state cultivation can cut the operations needed for one logical step tenfold, and that the work is being added to CUDA-Q.

NVIDIA’s release adds that IBM and RIKEN modeled protein-ligand complexes of up to 12,635 atoms, against a previous benchmark of 303, in work that is a finalist for the 2026 ACM Gordon Bell Prize. NVIDIA is also adding a quantum interface to the Slurm scheduler, with no date given.

What about networking?

Buck showed one of the first deployments of NVIDIA Photonics co-packaged optics switches, at Lambda. He said they are 15x more reliable on bit errors and link flaps. He also noted that NVIDIA now offers more than 1,000 CUDA-X libraries, looking back 20 years to his first public CUDA talk at an SC06 tutorial in Tampa, Florida.

Prices

None of the sources list prices. Rubin HPC is expected in 2028, and the quantum platform is a proposal. NVIDIA’s source text does not say how Nemotron for Science is licensed or priced, so check NVIDIA’s own pages.

Who it suits

Research groups with their own hardware. Nemotron for Science is aimed at labs that want agents running on a DGX Spark, a workstation or a local cluster, where data and models stay in-house.

Computational chemists and life scientists. The chemistry dataset and post-training recipes are the parts built for this field.

HPC center planners. The Vera CPU choices at DKRZ and TACC, the 2028 Rubin HPC timing and the Slurm quantum interface point to how centers may schedule agents and quantum jobs.

Quantum error-correction researchers. The TCBP-JOBS decoder and the IBM and Fermilab platform are for this community, not for general readers.

What is still unknown?

The agent results, such as the 81% fix rate and the 5x speedup, come from NVIDIA’s account of the talk, and the sources give no test details. Pricing and licence terms for Nemotron for Science, a start date for the quantum platform and a date for the Slurm interface are all missing.

What should you do next?

Watch NVIDIA’s SC26 sessions and its newsroom for the technical documents, and check the Nemotron for Science page for access terms before planning a pilot.