Report: Nvidia developing AI agent platform "NemoClaw"

Nvidia is following the hype surrounding the AI agent OpenClaw and is developing a platform called “NemoClaw” specifically for companies, Wired reports, citing several people familiar with Nvidia's plans. The platform is intended to enable companies to deploy AI agents that take over tasks for their workforce.

As it is said to be an open-source project, the code will not only be available to companies. However, Nvidia wants to provide corporate partners with additional tools for security and data protection to reduce central risks when using autonomous AI agents. In addition, according to the report, they could gain early and free access to “NemoClaw” if they participate in its development.

The chip manufacturer has already promoted “NemoClaw” to various major software providers. The Wired report names Salesforce, Cisco, Google, Adobe, and CrowdStrike as potential partners. However, it is unclear whether concrete cooperation has emerged from the talks.

Strategic opening at Nvidia?

According to the report, it is likely that potential partners will be able to use the platform regardless of whether their AI agents run on Nvidia chips.

The abandonment of a dependency on Nvidia GPUs is rather unusual for Nvidia. Until now, the company's software ecosystem has relied heavily on the proprietary CUDA platform, which effectively ties developers to Nvidia hardware and has given the company a significant competitive advantage.

The open-source approach is also remarkable, even though there are already examples of it: For instance, in December, Nvidia released Nemotron 3 Nano, an AI model that is almost entirely open source.

“NemoClaw” could be another step towards openness, possibly motivated by the increasing success of open AI models. Many startups and developers use such freely available models to test new applications or build their own products on them.

In addition, major AI providers are increasingly developing their own chips or having them developed to become more independent of Nvidia in the long term. Nvidia's partnership with the startup Groq, which develops so-called inference chips, also fits into this picture. These are less intended for training AI models and more for enabling already trained AI to respond as quickly and energy-efficiently as possible in everyday use.

The Wall Street Journal recently reported that Nvidia could present its inference chip based on a Groq design at the upcoming GTC developer conference. It is possible that there will also be concrete information about “NemoClaw” then. The conference will take place from March 16 to 19 in San Jose.

OpenClaw shows opportunities and risks of autonomous AI agents

The AI agent OpenClaw attracted great attention in the AI scene at the beginning of the year. The open-source project by developer Peter Steinberger differs significantly from classic AI chatbots, which usually only react to individual inputs and generate answers. OpenClaw, running on a local computer, can independently execute multi-stage tasks and control various programs or online services.

At the same time, this approach also harbors risks, as the powerful AI agents with extensive access rights can potentially cause damage if they malfunction or are manipulated. The Chinese cybersecurity authority has therefore recently advised authorities, state-owned companies, and banks against installing the AI agent on work devices. OpenAI hired Steinberger in February, but OpenClaw remains open source.

(mack)