Jensen Huang doesn’t need a new chip. He needs a new trench.

Nvidia AI dominated the first era — CEO Jensen Huang is making sure he owns the next. He’s turning Nvidia from a chipmaker that helps drive the market cycle to an operating system for the future of artificial intelligence.

The change has largely gone unnoticed and has yet to be priced in by investors. But the clearest signal so far has come this week.

At Nvidia’s annual developer conference, GTC, Huang launched Nemocla, an open-source, chip-agnostic platform for building and deploying AI agents — the autonomous software programs at the center of the latest advances in the industry.

“Every company in the world should have an agent system strategy,” Huang said. “It’s a new computer now.”

New chip announcements grabbed most of the attention at GTC, but the NemoClaw launch was a more important strategic shift and shows what Nvidia is really up to.

Nvidia won the AI ​​training era by locking out users. Its chips and software ecosystem were so deeply embedded in how AI models were built that it was impossible for a competitor to switch.

But the industry is shifting from building and training models to running them, and decision workloads don’t require the same lock-in. Google, Amazon And Broadcom Everyone is building chips according to their own conclusions. The moat that made Nvidia the world’s most valuable company is thinning.

Selling chips, even the best chips, ultimately means selling in a cycle. Having a platform where those chips run is a more durable business. It is sticky, high margin and difficult to dislodge. That’s where Huang is committing the crime with Nemoclav.

NemoClaw is built on OpenClaw, an open source agent created by a solo developer that went viral earlier this year and is the fastest growing open source project in history. Open source means that anyone can download, modify and run the software locally on their own servers. That makes it powerful, but also dangerous, because there’s no company controlling what an agent can access on your computer.

Enterprises banned OpenClaw as security risks increased. Nvidia’s version adds guardrails — security tools, privacy routing, data controls.

“Open” sounds generous, but it’s strategic for Nvidia. Nvidia offers the layer that drives adoption and monetizes what’s underneath — the chips and computing power each AI agent needs to actually run. Microsoft Internet Explorer didn’t charge, and Google didn’t charge for Android, but they unlocked the adoption of monetizing it: Windows and Search.

Huang is following that playbook — he’s not charging NemoClaw. The product is the platform. Mark Zuckerberg spent years and tens of billions of dollars on Metaverse, trying to escape his dependence on platforms owned by Apple and Google. Huang is making sure Nvidia never ends up in that position.

The most aggressive part of Huang’s strategy is that it is a direct threat to some of his top clients. Today’s Nvidia relies on a few companies building the most powerful AI models: OpenAI, Anthropic, Google, and Meta. If any one of them becomes dominant enough, it will gain leverage to squeeze Nvidia on price.

NemoClaw, named after Nvidia’s existing NeMo AI framework, prevents that. One AI CEO, who asked not to be named to speak candidly on the matter, called it a classic “commoditize the complement” strategy. While enterprises can deploy AI agents through NemoClaw for free, OpenAI and Anthropic find it difficult to charge premium prices for their own versions. Open source allows hundreds of companies to build their own models and split the model layer running, large enough to dictate any rules. Nvidia will stay in the middle and GPU demand will skyrocket.

Nvidia is also stepping into a gap that no one else is filling, at least in America. Meta pioneered open source AI with its Llama models, but its Next Frontier model may reportedly be closed. Google and OpenAI keep their best models proprietary, and Anthropic has never released open weights. The open source bench in America has been thin since the AI ​​boom began.

Chinese labs, meanwhile, are only accelerating open source efforts. DeepSeek has proven that it can build frontier models for a fraction of what US labs spend. AlibabaFollowed by Bytedance et al.

Data from OpenRouter, which tracks real-world model usage, shows that four of the five most popular models on its platform this month are open source, and most are Chinese. OpenRouter’s ranks are limited to its own customer base, and developers with enterprise deals often use API tools from sample companies.

Can a chipmaker really become an operating system?

History suggests otherwise. By previous attempts Intel And IBM Didn’t go anywhere. But Huang has pulled off platform transitions before, pivoting Nvidia from gaming to crypto to cloud to AI training. Nvidia posted just 73% revenue growth last quarter. Its latest guidance of about $80 billion for the fiscal first quarter crushed estimates.

Networking alone is now a multi-billion dollar business for Nvidia, and it only existed three years ago. No CEO in the semiconductor industry has a better track record of foreseeing a shift and proactively repositioning to take advantage of it.

NemoClaw content requires enterprise deployment. Nvidia’s open source models are free but so far unproven compared to what the Chinese labs are shipping. And the Huang vacuum could close fast if Meta reverses course or if Google opens up its models.

Representatives for Meta and Google did not immediately respond to requests for comment.

The question investors should be asking is not whether Nemocla will work tomorrow. Is Nvidia still a chipmaker or an operating system? One sells wheels, the other compounds. The market will price in the former but if Huang pulls it off, it should be priced in the latter.

Watch: Jim Cramer’s interview with Jensen Huang