NVIDIA AI Updates: May 20, 2026

1. NVIDIA and Google Cloud expand 100k-developer community with JAX path, Dynamo codelab, and Gemma 4 plus Nemotron pairing

NVIDIA. NVIDIA and Google Cloud expanded their joint developer community, now past 100,000 members, with a new JAX-on-NVIDIA-GPUs learning path, an NVIDIA Dynamo inference-optimization codelab, and monthly developer livestreams. The technical pitch lands on pairing Google DeepMind’s Gemma 4 models with NVIDIA’s Nemotron open models on Google Cloud G4 VMs powered by NVIDIA RTX PRO 6000 Blackwell GPUs, and using NVIDIA Dynamo on Google Kubernetes Engine to optimize mixture-of-experts inference at scale. The partnership also threads in responsible-AI work, with Google DeepMind’s SynthID watermarking integrated into NVIDIA Cosmos world foundation models for provenance on AI-generated content. Source

2. NVIDIA ships Verified Agent Skills as a signed, scanned capability layer for MCP-era agents

NVIDIA. NVIDIA introduced Verified Agent Skills, a portable instruction-set format paired with a verification pipeline aimed at giving enterprises capability-level governance over what AI agents can actually do. Skills flow from product-team repositories through human and automated review, security scanning via a new SkillSpector tool that targets prompt injection and tool poisoning, evaluation metrics, skill-card documentation, and cryptographic signing before publication to a public catalog that syncs daily. Each verified skill ships with a machine-readable skill card declaring ownership, dependencies, licensing, technical limits, and verification status, complementing NeMo Guardrails on the runtime side and giving teams a way to audit MCP-connected agent capabilities before deployment. Source