Nvidia chief executive Jensen Huang put one of the biggest phrases in AI back into circulation in March 2026. On Lex Fridman’s podcast, the host described artificial general intelligence, or AGI, as a system able to start, grow and run a company worth more than $1 billion. Huang replied, “I think it’s now. I think we’ve achieved AGI.” The line travelled fast because it sounded final. The fuller exchange was narrower.
Huang qualified the claim straight away with a much smaller example. An AI system might create a simple web service or app, go viral for a short time, charge a small fee and then fade. He later put the odds of 100,000 such systems building a company like Nvidia at effectively zero. Read plainly, he was describing flashes of broader ability.
Flashes like those may be the first credible glimpse of something broader than AI built for one task at a time. But they fall short of stable performance across many kinds of work. A paper on how to measure intelligence argues that strong results on known tasks do not prove general intelligence. In its account, intelligence shows when a system can learn and adapt under new conditions, instead of repeating patterns it has already absorbed.
That puts reliability at the centre of the debate. The open issue is whether that broader ability is dependable, sustainable and consistent. For a business, that separates a system that impresses in a demo from one it can trust in real operations, where inconsistency means delay, rework and risk.
The next limit is the lack of a body. AI keeps improving at writing, reasoning, analysis and planning. Yet it cannot move through the world, pick things up or find its way across a factory floor unless it is placed inside a machine.
Researchers are working on that step. A 2023 study showed that a model trained on vast amounts of text and images from the web could improve robot control. Robots using it handled unfamiliar objects better, followed commands they had not seen during robot training and did simple reasoning while carrying out physical tasks.
