Jensen Huang is not waiting for academic consensus. The Nvidia CEO has declared that artificial general intelligence—the long-pursued holy grail of AI research—is already here.

“I think it’s now. I think we’ve achieved AGI,” Huang said during a March 23 appearance on the Lex Fridman podcast. Coming from the man whose company supplies the computational backbone for nearly every major AI lab on the planet, the statement carries particular weight.

Redefining the finish line

On August 27, during Nvidia’s Q2 2026 earnings call, Huang doubled down on the claim. “For many tasks, we could say that we’ve already achieved AGI,” he told analysts and investors.

He went further, dismissing traditional AGI benchmarks as “kind of senseless.” Rather than measuring whether AI can match or exceed human cognition across the board, Huang argues that the real metric should be whether AI systems can generate what he calls “profitable tokens”—essentially productive economic output from real work.

The framing matters because AGI has never had a settled definition. Since the term was popularized in the 2000s, researchers have debated whether it means human-level performance across all cognitive domains, across economically valuable work, or something narrower. Huang’s position effectively shifts the goalposts from capability to commercial utility, aligning the milestone with what businesses can deploy today rather than with laboratory benchmarks.

400,000 AI agents and counting

Nvidia is building toward a workforce model in which more than 400,000 AI agents operate alongside roughly 40,000 human employees. The company envisions these agents handling tasks across engineering, operations, and business functions.

On the hardware side, Nvidia has discussed GPU deployments involving more than 1.5 million chips in a single generation, powering what the company calls “AI factories”—purpose-built facilities designed to produce intelligence as an industrial output.

A convenient disagreement with OpenAI

Huang’s definition of AGI puts him at odds with OpenAI. The Sam Altman-led company has maintained a more formal definition, describing AGI as systems that outperform humans across a majority of economically valuable tasks.

The disagreement is not merely academic. OpenAI’s AGI definition is embedded in its corporate governance structure, with specific provisions tied to what happens when AGI is achieved. Notably, OpenAI’s landmark compute agreement with Microsoft applies only to pre-AGI systems, meaning a formal AGI declaration carries direct commercial consequences for that partnership. For OpenAI, declaring AGI means triggering contractual and organizational changes. For Nvidia, declaring AGI means selling more hardware to companies racing to build on top of it.

Huang credited OpenAI’s contributions to reaching this moment, acknowledging the role its models have played in demonstrating what is possible with scaled compute.

For readers tracking the debate, the practical test of Huang’s claim will come in enterprise adoption: whether AI systems credited with generating “profitable tokens” translate into measurable productivity gains at scale, and whether other major AI labs accept a task-by-task definition of AGI or hold to more comprehensive ones.