Jensen Huang says AGI has arrived, citing OpenAI’s newest model trained on ~100,000 of Nvidia’s chips

Jensen Huang just drew a line in the sand and planted Nvidia’s logo on it. On Sunday, the company boss said OpenAI’s newest ChatGPT model had been trained on Nvidia chips, then went further and declared that artificial general intelligence had arrived. That is a very big claim for a very messy term. AGI is the label people use for software that can do most of the thinking work a human can do, and nobody has agreed on when, or even whether, it truly shows up.

The hardware story is doing a lot of the work here. Nvidia once sold chips eight to a board; now it sells systems with 72 chips packed into one box, wired to behave like a single giant brain. That design is expensive and only really makes sense when someone is training something enormous. Huang says OpenAI’s newest model was trained on about 100,000 of those boxes. He briefly posted 300,000, then deleted that and replaced it with the smaller figure. Nvidia hasn’t explained the change.

The money is impossible to ignore. Nvidia said it sold $89 billion of AI computers in three months, more than double a year earlier. Huang’s line on the earnings call was blunt: “Now, compute is revenue.” That is the whole business in one sentence. Sell more computing power, watch the numbers climb, and hope the market doesn’t get too distracted by the philosophy.

OpenAI is not backing Huang’s language. Its president, Greg Brockman, did not go as far. And the partnership between the companies is not as grand as the headlines make it sound: the $100 billion deal they announced in 2025 was never signed, and BeInCrypto reported in June that OpenAI was quietly buying fewer Nvidia chips. Huang still ended his post with another big number: 400,000 more chips, coming soon. That is what markets will track. The AGI talk is just the garnish.