This is CBS Eye on the World. Here's John Batchelor.

Artificial intelligence, NVIDIA, and then Deepseek. We're telling a story here that is an illustration of how technology works in the 20th and now 21st century.

The breakthrough, and then I welcome John Cochran of Stanford University, a professor of economics,

who is also an esteemed fellow at the Hoover Institution writing his substact blog, Grumpy Economist.

He's anything with Grumpy.

He's cheerful.

But right now we're turning to the matter that made all of Wall Street frown.

And as far as I know, the oligarchs of the universe frowned when a startup, not even a monetized shop, a research development called Deep Seek out of China presented evidence

that the billion dollars you might need or 100 million minimum you might need to train

the chips that NVIDIA turns out to turn to answer the questions that the public has about artificial intelligence,

that that might be reducible by magnitude that I can only guess at because my minus is very

weak. John, a very good evening to you. It was a hundred million, maybe a billion, say the

lords of the artificial intelligence. And then it comes in about five to six

million, maybe some more pricing on top because of research, but certainly not a hundred million.

Was this, if I read your column correctly, this was always going to happen. It was the question of when.

Good evening to you, John. Good evening. Yes. That's the way things work.

We invent a new technology. We figure out, wow, it can do amazing things at prodigious cost,

and then the cost gutters get to work and figure out how to do it cheaper and cheaper and

cheaper. And that's, you know, since the first computers had vacuum tubes and look what's happened to computers.

Artificial intelligence and machine learning, the way they work is to be just computing power is cheap.

So we'll just throw lots of computing power things.

Well, computing power is cheap until you scaled that up into the $100 million range. That focuses the mind on can we do this, in this case, with cleverer software

...