NVIDIA launches Nemotron Labs open models for enterprise and sovereign AI

NVIDIA unveiled Nemotron Labs, framing a family of open models as the foundation for enterprises and nations to build AI they can 'trust, control, and customize.' The sovereign-AI pitch targets governments and regulated industries wary of depending on closed frontier APIs — a direct appeal to the same data-sovereignty anxieties Hassabis and export controls have amplified. Nemotron 3 Ultra reportedly leads benchmarks at lower cost than top closed models when paired with LangChain's agent orchestration.

On the tooling side, NVIDIA published two concrete demonstrations. First, using Agent Skills, developers can post-train its Cosmos 3 vision-reasoning model above 90% accuracy in a single day with almost no manual effort — automating what used to be weeks of adaptation work. Second, an autoresearch workflow using RL Agent Skills with NVIDIA NeMo lets coding agents inspect repositories, set up runtimes, and resolve issues across long-running ML tasks.

The strategic thread is embodied and agentic AI as infrastructure. NVIDIA expanded its collaboration with Hugging Face and LeRobot, bringing Cosmos 3 to open robotics platforms — the same week Hugging Face shipped LeRobot v0.6.0. NVIDIA is positioning itself not just as the GPU vendor but as the full-stack provider for open, controllable AI.

The competitive context: this is NVIDIA hedging against the risk that closed-model labs (OpenAI, Anthropic) capture value at the application layer. By championing open models and enterprise control, NVIDIA keeps demand flowing to its hardware regardless of which lab wins. The caveat is that 'open' here still means the NVIDIA software and silicon stack — watch whether Nemotron gains real independent adoption versus staying a GPU-demand driver.