Nvidia at SC26: Agents are the new scientists, and they need a lab

What happened
At the SC26 conference, Nvidia introduced several new technologies aimed at scientific research using AI agents. The company launched the Nemotron 3.5 science model, the Vera CPU rack, the Rubin HPC GPU, new CUDA-X libraries, and quantum computing integrations. These products are designed to support AI agents that can autonomously conduct scientific experiments and iterate on hypotheses. The announcements represent a shift in focus from maximizing raw computational speed to enabling faster cycles of hypothesis testing and discovery.
Key details
- Nemotron 3.5 Super VL model comes with LitmusBench v0.2, an open dataset of nearly 50,000 chemistry-reasoning questions.
- Vera CPU rack contains 192 Vera CPUs, about 300TB of memory, and Quantum-X800 InfiniBand, with shipments by Dell, HPE, Lenovo, and Supermicro expected in early 2027.
- Rubin HPC GPU features over 50,000 CUDA cores and sparse-data acceleration, with a planned release in 2028.
- New CUDA-X libraries (cuPhoton, cuPotential, cuSpa) target fields such as astrophysics, drug discovery, materials science, and single-cell biology.
- Nvidia announced a modular quantum co-design platform with IBM and Fermilab, and quantum scheduling support in Slurm.
Why it matters
Scientific researchers and organizations using high-performance computing can now leverage AI agents to automate and accelerate the research process, reducing the time from hypothesis to result. Companies and labs adopting these new Nvidia tools may see lower costs and higher accuracy in specialized scientific tasks, especially when using domain-tuned AI models. The new products also give organizations more options for running AI workloads on infrastructure they control, including CPU-only and quantum-integrated systems.