Jensen Huang puts AGI on a five-year clock

Everyone has been trying to pinpoint when AGI will finally arrive. Now, the man selling the shovels for the whole revolution has stamped a date on it.

Speaking at the 2023 New York Times DealBook Summit, Nvidia CEO Jensen Huang said we could see artificial general intelligence within the next five years.

When asked whether AGI would arrive in the next decade, Huang didn't hesitate. He noted that depending on how you define it, the answer is yes.

Redefining the finish line

Huang grounded his timeline on a very specific definition. To him, AGI means software or computers that can pass tests reflecting basic intelligence in a way that is fairly competitive with a normal human.

Under that benchmark, Huang said systems will obviously be passing those tests within five years.

When interviewer Andrew Ross Sorkin asked if AGI meant AI capable of designing Nvidia's own cutting-edge hardware, Huang agreed. In fact, he revealed that Nvidia's flagship H-100 chips are already designed with the help of multiple AI systems.

According to Huang, modern chipmaking and software engineering at Nvidia simply wouldn't exist without AI doing heavy lifting today.

Fast progress, but reasoning is missing

Despite the rapid pace, Huang made sure to draw a clear boundary around what today's systems can actually do.

He admitted that AI is moving faster than he anticipated, but insisted it still shows no sign of mastering complex human cognition.

The biggest roadblock is multi-step reasoning. Humans excel at thinking through complex chains of logic, and Huang pointed out that AI simply cannot do that right now.

Huang is not alone in predicting a near-term arrival. Former Meta executive John Carmack estimated AGI would arrive by the 2030s and create trillions in value, while Google DeepMind CEO Demis Hassabis predicted human-level AI within a few years.

Yet others remain deeply concerned. OpenAI cofounder Ilya Sutskever has warned of automated weapons and cybersecurity threats, while investor Ian Hogarth cautioned that unregulated superintelligence could lead to human obsolescence.

Why it matters

Huang's five-year prediction is clever because it shifts the goalposts from sci-fi consciousness to functional utility. By defining AGI as passing standard tests and assisting complex engineering, he turns a philosophical mystery into a measurable milestone.

For developers and hardware engineers, the message is already playing out on the ground. When the world's premier chip designer admits its best chips cannot be made without AI assistance, operating alongside AI is no longer an optional perk—it is the baseline workflow.

At the same time, we have to look at the business behind the words. As the dominant seller of AI hardware, keeping the timeline urgent keeps cloud giants and startups trapped in an endless GPU buying spree.

Still, Huang's candid callout of multi-step reasoning highlights the real barrier ahead. Passing multiple-choice exams is one thing, but until models solve deep, chain-of-thought logic, true AGI might stay just out of reach no matter how much compute you throw at it.