Mistral AI said on March 16, 2026 that it is entering a strategic partnership with NVIDIA to co-develop frontier open-source AI models. A linked Mistral post says the effort begins with Mistral joining the NVIDIA Nemotron Coalition as a founding member and contributing large-scale model development plus multimodal capabilities.
Roche says it is deploying more than 3,500 NVIDIA Blackwell GPUs across hybrid cloud and on-premises sites in the U.S. and Europe. The company says the new AI factory backbone will support biological foundation models, faster drug discovery, digital twins for manufacturing, and broader AI use across diagnostics and digital health.
Microsoft used NVIDIA GTC on March 16, 2026 to widen Microsoft Foundry and Azure AI in three directions: production agent tooling, next-generation NVIDIA infrastructure, and Physical AI workflows. The company said Foundry Agent Service is now generally available, Nemotron models are coming to Foundry, and Azure is already powering on NVIDIA Vera Rubin NVL72 in Microsoft labs.
Nvidia's March 16, 2026 DLSS 5 reveal drew a widely shared critique from Digital Foundry's John Linneman, who praised the environmental lighting upside while warning that the character rendering looked far too invasive.
NVIDIA announced an open Physical AI Data Factory Blueprint on March 16, 2026. The reference architecture is designed to unify data generation, augmentation, and evaluation for robotics, vision AI agents, and autonomous vehicles, with GitHub availability expected in April.
On March 11, 2026, NVIDIA introduced Nemotron 3 Super, an open 120-billion-parameter hybrid MoE model with 12 billion active parameters. NVIDIA says the model combines a 1-million-token context window, high-accuracy tool calling, and up to 5x higher throughput for agentic AI workloads.
A high-signal LocalLLaMA thread on March 15, 2026 focused on a license swap for NVIDIA’s Nemotron model family. Comparing the current NVIDIA Nemotron Model License with the older Open Model License shows why the community reacted: the old guardrail-termination clause and Trustworthy AI cross-reference are no longer present, while the newer text leans on a simpler NOTICE-style attribution structure.
A LocalLLaMA thread amplified Phoronix coverage of GreenBoost, an experimental GPLv2 Linux module that adds a multi-tier memory path for NVIDIA GPUs. The design pairs a kernel module with a CUDA shim so large allocations can spill from limited on-card vRAM into pinned system RAM and NVMe-backed storage without modifying CUDA applications.
A March 15, 2026 LocalLLaMA post pointed to Hugging Face model-card commits and NVIDIA license pages showing Nemotron Super 3 models moving from the older NVIDIA Open Model License text to the newer NVIDIA Nemotron Open Model License.
NVIDIA and ABB Robotics are integrating Omniverse libraries into RobotStudio to bring industrial-grade physical AI simulation to factory automation. The companies say RobotStudio HyperReality reaches 99% sim-to-real accuracy and can cut deployment costs by up to 40%.
NVIDIA introduced Nemotron 3 Super on March 11, 2026 as an open 120B-parameter model built for agentic AI systems. The company says the model tackles long-context cost and reasoning overhead with a 1M-token window, hybrid MoE design and up to 5x higher throughput.
NVIDIA and Thinking Machines Lab said on March 10, 2026 that they will deploy at least one gigawatt of next-generation NVIDIA Vera Rubin systems under a multiyear partnership. The agreement also covers co-design of training and serving systems plus an NVIDIA investment in Thinking Machines Lab.