SourceYouTube

Published March 10, 2026

CNBC Television

Hosts:Carl Quintanilla

This is an AI-generated summary. The source video includes demos, visuals and context not covered here. Watch the video → · How our articles are made →

In Brief

Nvidia is reportedly building NemoClaw, an open-source platform for AI agents, according to a report by Wired. The platform has already been pitched to Salesforce, Cisco, and Google (0:14). What makes this strategically striking is that NemoClaw would run on any chip, not just Nvidia's own hardware — a direct break from the company's two-decade strategy of locking developers into its CUDA software ecosystem. OpenClaw, the viral open-source AI agent project that inspired the name, went viral earlier this year and has been closely followed by both Silicon Valley and Chinese developers (0:48). Nvidia's developer conference GTC 2026, running March 16-19, is expected to be where CEO Jensen Huang addresses the move. For more on OpenClaw's rise, see 90,000 Commits, One Developer: The OpenClaw Story and OpenClaw and the Age of Personal AI Agents.

What happened

CNBC reporter Deirdre Bosa, citing a Wired report, revealed that Nvidia is building NemoClaw: an open-source platform for enterprise AI agents (0:36). AI agents are software programs that complete tasks step-by-step on your computer without you touching the keyboard. Unlike a chatbot, which answers questions, an agent can independently research, write, book, or execute actions on your behalf.

The "claw" name comes from OpenClaw, the open-source AI agent framework created by developer Peter Steinberger that went viral earlier this year and accumulated over 180,000 stars on GitHub. OpenClaw quickly became a reference point for the kind of autonomous, locally-run agent software the industry is converging on.

NemoClaw has reportedly been pitched directly to Salesforce, Cisco, and Google (0:21), suggesting Nvidia is positioning it as an enterprise-grade solution from day one. The platform is described as open source, meaning the underlying code would be freely available for anyone to use, modify, and build on.

Context and background

For nearly 20 years, Nvidia's competitive advantage has been CUDA (1:12). CUDA is a software platform that lets developers use Nvidia's Graphics Processing Units (GPUs), the specialized chips originally designed for graphics that turned out to be ideal for AI. Because so much AI code is written specifically for CUDA, switching away from Nvidia hardware became painful and expensive. That dependency is what gave Nvidia its dominant market position.

NemoClaw would break that playbook because it is designed to run on anyone's chips, including those made by competitors (1:23).

CNBC's Bosa drew an explicit comparison to Microsoft under CEO Satya Nadella. Nadella inherited a company whose profits depended on locking users into Windows, then made a bold bet in the opposite direction: he put Microsoft Office on the iPad, embraced Linux, and took Microsoft's software cross-platform. The logic was that if your product is genuinely the best, you do not need walls to keep customers in — you just need to be everywhere (1:28). That shift helped take Microsoft from a few hundred billion dollars in market capitalization to over three trillion dollars today (1:47).

Nvidia faces growing pressure on the chip side from Google, AMD, Amazon, and Broadcom, all of which are building custom AI chips to reduce their dependence on Nvidia hardware (2:07). Rather than defend its existing hardware moat, Nvidia appears to be racing to own the software layer above it.

Why this matters

The enterprise AI agent market is growing fast, but adoption has hit a wall. Businesses can see the productivity potential, but the tools are currently unpredictable. They lack standard guardrails: the safety and compliance controls large organizations require before trusting software to act on their behalf (1:01). That is the gap NemoClaw reportedly targets.

Nvidia's business logic for giving the platform away for free is straightforward: more agents running everywhere means more compute consumed everywhere (1:57). And compute, the raw processing power required to run AI models, is exactly what Nvidia sells. If NemoClaw becomes the standard framework that enterprises use to build and run AI agents, Nvidia wins regardless of whose chips are running underneath.

For context on how AI agents and inter-agent protocols are evolving at the infrastructure level, see A2A and MCP: The Two Protocols AI Agents Need.

What we are tracking next

• GTC 2026 (March 16-19): Jensen Huang is expected to address NemoClaw at Nvidia's developer conference. An official announcement would confirm the Wired report and reveal more technical details (2:22).
• Enterprise partnerships: Whether Salesforce, Cisco, or Google publicly confirm their involvement will signal how serious the early-stage pitches were.
• Competitor responses: How Google, AMD, Amazon, and Broadcom respond to Nvidia moving up the stack into open-source agent software.

Glossary

Sources and resources