DeepSeek, AI and the markets

Katie Martin (0:06)

Pushkin. Big curveball hit global markets this week in the form of DeepSeek, a new alternative to ChatGPT that has sprung out of China. It was, we're told, dead cheap to build, it requires fewer fancy microchips than the dominant US versions, and seems to work pretty well. Its emergence whacked a massive $600 billion off the market value of chip maker Nvidia on Monday. Even for the biggest company on the planet, that is really quite a lot of money. Today on the show, we're asking, why is this such a big deal? This is Unhedged, the markets and finance podcast from the Financial Times and Pushkin. I'm Katie Martin, I'm not an AI robot, I'm a human markets columnist here at the FT in London, and I'm joined down the line from New York City by fellow human, Robert Armstrong, from the Unhedged newsletter. Rob, how's it going?

Robert Armstrong (1:03)

It's so great to have a day where markets go a little bonkers, and it's not about American politics. It's about something else.

Katie Martin (1:12)

There's still time. The week is young.

Robert Armstrong (1:16)

Yeah, we're not going to run out of bonkers anytime in the next couple of months.

Katie Martin (1:20)

Not soon.

Robert Armstrong (1:21)

Yeah.

Katie Martin (1:21)

Here's the thing. DeepSeek has been around for quite a while, right? The tech community has been abuzz with it for weeks, but the simple souls of like financial markets, like traders and investors and stuff, and frankly, you and me only really got the memo this week. So we're not techies and we're not going to pretend to be techies. But why are markets in a tizzy about this?

Robert Armstrong (1:44)

Several days ago, let's call it a week ago, DeepSeek said, we have this new model that was cheap to build and works really well. And everybody was like, yeah, tell me another one, broadly speaking. And then they released it, meaning you could download the model and run it on your own computers. And it slowly dawned on people in the AI industry. This thing really works. And that message took a day or two to get through to markets. But the message did get through yesterday, boy. And it's interesting listening to people talk about it. There is remarkably little noise from people whose net worth has radically decreased as a result of the appearance of this model. Very few of them are calling BS on this thing. Instead, they're saying, wow, this actually is very impressive. We're working with this. We're trying to reverse engineer this. This is very fascinating indeed.

Katie Martin (2:46)

Yeah, I've seen some of the work that it spits out. The economist Stephanie Kelton, she asked it, what would happen to Canadian GDP in the event that Trump brought in 25 percent tariffs on Canadian exports, and it spat out a very credible answer actually, and came up with a number that matches what a lot of actual humans have come up with.

Robert Armstrong (3:08)

We're going to have so much time for our hobbies, Katie, when this thing takes our sad little jobs. I'm going to take up quilting, I think is what I'm going to do.

Katie Martin (3:21)

I'm thinking paddle. So last week, Angela Zhang, Z-H-A-N-G, she's an expert in Chinese tech. She wrote in the FT about how DeepSeek is like the real deal, and here's a really interesting wrinkle that she was writing about. The reason I spell out her surname is that listeners, you got to check this thing out because it flew under the radar when it came out last week, but she is dead right. She was saying, first of all, DeepSeek is not a one-off. There's a whole bunch of Chinese AI developers out there doing really interesting stuff and coming up with high-quality products. The other is, one of the reasons this has all happened is that the US has starved China of good quality powerful US chips, and so Chinese developers, Chinese engineers, Chinese techies and nerds basically have had to work really hard and innovate really quickly to come up with new solutions of their own.

Robert Armstrong (4:16)

Especially on the front of efficiency. They've essentially been deprived of computer power, and so they've just been forced to think harder about efficiency. It is an important lesson in how technology evolves and how hard it is to keep it in a box, as it were.

Katie Martin (4:34)

So, OpenAI's ChatGPT cost whatever $11,000,000,000 to build. And OpenAI is still constantly saying, we need more bazillions of dollars, possibly even trillions of dollars to keep building this out and create the energy platforms that we need to make this work. Yadda yadda. DeepSeek is saying, yeah, we did it for $5.6 million bucks. And you're like, you did? Yeah.