NVIDIA released open-source physical AI agent skills across Omniverse, Cosmos, Isaac, Metropolis, Alpamayo, and Jetson. The company points to manufacturing gains including 67% faster training and deployment at Pegatron and a 17% detection-rate improvement at Delta Electronics.
NVIDIA says Vera is now in full production and can complete agentic workloads 1.8x faster than x86 CPUs. OpenAI, Anthropic, SpaceXAI, ByteDance, CoreWeave, and OCI are among the names tied to adoption or evaluation.
NVIDIA released Cosmos 3 as an open physical AI omnimodel with Super and Nano variants. Its technical post points to six synthetic datasets, Hugging Face checkpoints, and GitHub recipes for domain adaptation.
NVIDIA’s open humanoid reference design combines Unitree H2 Plus hardware, Sharpa five-finger hands, and Jetson AGX Thor T5000 compute. The 75-DoF system is aimed at making humanoid research more comparable across labs.
NVIDIA is packaging a 550B-parameter MoE model with agent tooling instead of treating the model as a standalone release. The pitch is concrete: up to 5x faster inference, up to 30% lower cost, and availability beginning June 4.
NVIDIA is targeting the hidden cost of LLM serving experiments. Its DynoSim post says the Rust simulator can screen deployment choices before GPU validation, with a blog example replaying 23,608 requests about 1,500x faster than real time.
The expensive part of LLM inference is often the experiment itself. NVIDIA says DynoSim replayed a 23,608-request trace on an Apple M4 MacBook Air in 2.41 seconds, about 1,500x faster than the 60.1-minute serving window it modeled.
NVIDIA Labs released SANA-WM, a 2.6B parameter open-source world model capable of generating up to one minute of 720p video. The relatively small model size and open-source availability make it a significant contribution to accessible video generation research.
Anthropic on May 6 signed an agreement to use all compute capacity at SpaceX-xAI Colossus 1 in Memphis — 220,000+ NVIDIA GPUs and 300MW. The deal doubles Claude Code rate limits and removes peak-hour caps for paid subscribers.
NVIDIA AI has released Star Elastic, an innovative architecture that packs 30B, 23B, and 12B reasoning models into a single checkpoint, enabling zero-shot slicing to dynamically switch between model scales without separate downloads.
Intel ($INTC), AMD ($AMD), and Micron ($MU) all posted double-digit gains in the week ended May 8, 2026, while Nvidia ($NVDA) lagged the group. Analysts framed the rotation as a structural shift toward CPU and memory chipmakers as beneficiaries of the AI inference cycle, with Apple's preliminary Intel Foundry Services agreement as the primary catalyst.
NVIDIA unveiled Nemotron 3 Nano Omni on April 28, 2026 — an open 30B-A3B hybrid MoE model unifying vision, audio, and language with a 256K context window and 9x higher throughput than comparable open omni models.