NVIDIA's "Lobster" NemoClaw Revealed

What NVIDIA is unveiling

It has been reported that NVIDIA (英伟达) will use next week’s GTC developer conference in San Jose to unveil NemoClaw, an open‑source platform for autonomous AI agents that Wired (连线) first described in a blog post. Reportedly the platform’s headline technical claim is hardware agnosticism: enterprises could run and distribute NemoClaw agents even on non‑NVIDIA chips, enabling businesses to automate internal workflows without being locked to a single GPU vendor.

It has also been reported that NVIDIA is actively pitching the project to major software firms, including Salesforce, Cisco, Google, Adobe and CrowdStrike, and that early access may be available to partners who contribute code to the open‑source project. The company’s playbook appears to be straightforward: lure software vendors with an enterprise‑friendly agent framework and a safety layer, then accelerate adoption through open collaboration.

Why enterprises — and regulators — will care

The move follows a wider surge of interest in so‑called “lobster” agents (小龙虾), open tools that run locally and can execute long sequences of tasks with minimal human prompts. Proponents argue these agents can dramatically cut labor and streamline processes. But it has been reported that large firms already worry about unpredictability and security: Wired noted Meta has banned OpenClaw on work machines after a security lead said an agent formerly deleted many work emails.

There’s also a geopolitical subplot. Analysts note that hardware‑agnostic software takes on new significance amid US export controls and trade frictions that limit access to top‑tier accelerators in some markets, including China. Whether NemoClaw will ease deployment across constrained hardware environments — and whether its promised enterprise safety layer will be enough to placate security teams — are the questions now being tested on the industry stage.