Nvidia announces NemoClaw at GTC — Jensen Huang calls OpenClaw 'the most important software release probably ever' and ships Nvidia's enterprise layer for it

At the GTC 2026 keynote on March 16, Jensen Huang devoted significant stage time to the "claw" ecosystem — the open-source AI agent frameworks that have exploded in adoption since January. Huang called OpenClaw "the most important software release probably ever," describing it as the orchestration layer that turns LLMs into persistent agents capable of spawning sub-agents, accessing cloud systems, maintaining memory, and decomposing complex tasks. He predicted OpenClaw would reshape enterprise IT the way Linux and HTTP did in prior eras — turning SaaS into what he called GaaS (Governance as a Service), with AI agents acting as the actual service layer.

Nvidia is declaring where the next platform battle is. GPUs won the training era. NemoClaw is the opening move for winning the inference-and-orchestration era — specifically the enterprise slice of it, where security, compliance, and auditability are buying criteria. By building NemoClaw on top of OpenClaw (now owned by OpenAI), Nvidia is positioning itself as the enterprise deployment layer regardless of which frontier model a company uses. The hardware-agnostic framing is deliberate: Nvidia doesn't need you to buy their GPUs to use NemoClaw, but NemoClaw is designed to run best on DGX infrastructure, which they sell.