Andrej Karpathy: only ~5,000 people have top-tier AI access

Former OpenAI researcher Andrej Karpathy says the biggest AI divide right now is not technical. It is access. On X, he argued that understanding of what AI can actually do is splitting apart, with billions of people still only skimming the surface while a much smaller group is seeing something far more powerful.

His rough numbers are striking. Karpathy estimated that about six billion people have barely interacted with large language models at all. Another one to two billion use free versions casually, mostly as a search replacement or a writing helper. Then comes a far smaller set: roughly 20 million professionals, especially in mathematics and software engineering, who are using frontier models to finish work that used to take weeks or months.

And then there’s the really tiny club. Karpathy said about 5,000 people have internal access to the most advanced systems, the ones that can coordinate thousands of AI agents across huge software projects, cybersecurity work and mathematical research. He said the outside world has only seen a preview of what those systems can do.

His warning was blunt. Human review and comprehension, he said, are starting to lag behind what these systems generate. That gap, he argued, comes from access, task complexity and the fact that AI is much better in digital areas like coding and math, where outputs can be checked.

The spending side is starting to follow the same pattern. Elon Musk backed Aaron Levie’s warning that autonomous agents will need serious computing power. Morgan Stanley put a possible $32.5 billion to $60 billion addition to the data-center CPU market by 2030 on the table, while also saying global data-center electricity use could more than double. McKinsey, meanwhile, said task costs can swing by as much as 30 times, which is a cheerful little sentence if you enjoy invoices.