Physical AI News (Japan)

Physical AI, robotics and compute — reported from Japan.

日本語
· Product launches· 日本語で読む

NVIDIA Announces AI Agents for Scientific Research and HPC Platforms at SC26

NVIDIA’s Ian Buck kicked off the company’s weekly program at SC26. He introduced tools for AI agents that plan experiments, run code and check results on researchers’ own systems. The announcements also included Nemotron for Science, the Rubin HPC GPU and an expanded Vera CPU platform.

Key facts

  • Ian Buck kicked off NVIDIA’s weekly program in Chicago, the host city of SC26.
  • The AI agent tools introduced plan experiments, run code and check results on researchers’ own systems.
  • The announcements included Nemotron for Science, the Rubin HPC GPU and an expanded Vera CPU platform.

What happened

NVIDIA’s Ian Buck kicked off the company’s weekly program in Chicago, where SC26 was held, and announced tools for AI agents for scientific research. The tools plan experiments, run code and check results on researchers’ own systems. Buck also announced Nemotron for Science, the Rubin HPC GPU and an expanded Vera CPU platform for high-performance computing.

Why it matters

NVIDIA presented tools for AI agents that carry out tasks from experiment planning through results checking on researchers’ own systems, highlighting capabilities that extend into scientific research workflows. The materials do not specify the computing infrastructure required or how results are checked, so the scope of their application in research settings will depend on the release of detailed specifications.

Analysis

A notable aspect of the announcements is that NVIDIA presented AI agent capabilities for scientific research alongside GPU and CPU platforms for HPC in the same special address. The AI agents are said to handle tasks from experiment planning through code execution and results checking on researchers’ own systems. However, the materials provided do not show how Nemotron for Science, the Rubin HPC GPU and the expanded Vera CPU work together. The announcements should therefore be understood as presenting both research software and computing platforms, rather than as a single integrated product or operating environment. For research settings, a key point is that the agents are described as not only running code but also checking its results. The materials do not explain the procedures or criteria for these checks, or at what stages researchers intervene. The extent to which the tools could contribute to reproducibility or error detection cannot be determined from the feature description alone; actual use cases and validation methods will need to be examined. The description that the tools run on researchers’ own systems also offers a clue to how they may access computing resources. This is a separate issue from the U.S. BIS regulates remote use of advanced AI chips, but the location where research tools run should be distinguished from the regulatory conditions that apply when computing resources are accessed remotely. The materials provided do not establish whether the announced tools fall under regulated forms of use. Questions to follow include how each tool relates to the computing platforms, supported models and system requirements, availability and the specific mechanisms for checking results. Without this information, it is not possible to assess whether researchers can deploy the tools in existing computing environments or whether they require specific NVIDIA products. More important than the number of names announced is which processes run on researchers’ systems and how the tools return results that can be verified.

Details

According to NVIDIA, the AI agent tools introduced can plan experiments, run code and check results on researchers’ own systems. The materials provided do not include specifications, availability dates or terms of use for the tools. In the same special address, NVIDIA announced Nemotron for Science, the Rubin HPC GPU and an expanded Vera CPU platform. The materials provided do not include performance figures, configurations or prices for the products.

Sources

Translated from the Japanese original: /news/nvidia-sc26-de-kagakukenkyuu-muke-ai-eejiento-to-hpc-kiban-w-011b27. The Japanese version is authoritative if they differ.