Jensen Huang says we’ve achieved AGI, then immediately walks it back

Silicon Valley has spent the past few months tiptoeing around the term AGI, with founders trying to invent fresher, less-exhausting buzzwords.

Then the man selling all the shovels went on a podcast, declared the finish line had already been crossed, and immediately pulled the rug out.

During a Monday appearance on the Lex Fridman podcast, Nvidia CEO Jensen Huang dropped a major bombshell. When asked whether true AGI was five, ten, fifteen, or twenty years away, Huang didn’t hesitate: "I think it's now. I think we've achieved AGI."

The billion-dollar test

Fridman set the bar high before asking the question. His definition of AGI wasn't passing a benchmark test; it was an AI system capable of doing a founder's job by starting, growing, and running a successful tech company worth over $1 billion.

Huang pointed to the wild popularity of OpenClaw, the open-source AI agent platform, as evidence that the shift has already begun.

People are deploying autonomous agents for almost everything, Huang noted. He said he wouldn't be surprised if someone casually created an overnight digital influencer or a viral social app that feeds a digital Tamagotchi.

Walking it right back

Fridman pointed out that declaring AGI achieved would send shockwaves through the industry. Right on cue, Huang pumped the brakes.

Huang admitted that the current reality of AI agents is far less glamorous. A lot of people play with them for a couple of months, lose interest, and the usage dies away.

Then came his reality check: the odds of 100,000 of those agents building a company like Nvidia are exactly zero percent.

Within the span of a single conversation, Huang went from crowning today's tech as AGI to acknowledging that swarms of agents can't even begin to replicate a real enterprise.

Why it matters

In our view, Huang’s whiplash statement neatly captures both the hype and the ceiling of the current agent wave.

For builders and users, open-source platforms like OpenClaw prove that automating personal tasks is easier than ever. But Huang's walk-back touches on an uncomfortable truth: agent retention is notoriously poor. AI can whip up a clever Tamagotchi app overnight, but it lacks the strategic endurance and organizational grit required to run a real business.

Across the wider industry, AGI is no longer just philosophical bar talk. It is a contractual trigger. In multibillion-dollar deals like the one between OpenAI and Microsoft, legal definitions of AGI dictate intellectual property rights and revenue shares, which is why most tech executives now actively dodge the term.

Huang, however, has nothing to lose by playing both sides. Proclaiming AGI gives AI adoption another adrenaline shot, while joking that 100,000 agents still can't build Nvidia protects his own executive prestige. Either way, whether those 100,000 agents fail or succeed, they will still need millions of Nvidia chips to run.