WSJ What’s News - Tech Stocks Tumble Sparked by China’s DeepSeek

Episode Date: January 27, 2025

P.M. Edition for Jan. 27. Some $1 trillion was erased from the U.S. stock market today as investors reacted to a cheaper, high-performing AI model from Chinese company DeepSeek. WSJ reporter Asa Fitch tells us what’s unique about the company and its technology. Plus, Journal markets reporter David Uberti joins to discuss what this might mean for tech stocks in the long run. And a top prosecutor opens a review of the Justice Department’s decision to charge Jan. 6 rioters. Alex Ossola hosts. Sign up for the WSJ's free What's News newsletter. Learn more about your ad choices. Visit megaphone.fm/adchoices

This episode is brought to you by Samsung Galaxy. Ever captured a great night video only for it to be ruined by that one noisy talker? With audio erase on the new Samsung Galaxy S25 Ultra, you can reduce or remove unwanted noise and relive your favorite moments without the distractions. And that's not all. New Galaxy AI features like NowBrief will give you personalized insights based on your day schedule so that you're prepared no matter what. Pre-order the Samsung Galaxy S25 Ultra now at Samsung.com.

More than a trillion dollars have been erased from the U.S. stock market. Tech stocks dropped after China's DeepSeek released a cheaper AI model. Plus, is this the correction in AI stocks investors were expecting? Today is one of the big first pullbacks against that theme that we've seen over the last year or two. And it has raised in a really big way whether the market has overvalued some of these stocks.

And a top prosecutor has opened a review of the Justice Department's decision to charge hundreds of defendants of the January 6th riots. It's Monday, January 27th. I'm Alex Osela for The Wall Street Journal. This is the PM edition of What's News, the top headlines and business stories that move the world today. An artificial intelligence model from the Chinese company DeepSeek has sparked a deep

freakout among investors. Nvidia's stock tumbled nearly 17%, wiping out about $593 billion from the company's market cap. Makers of AI infrastructure such as Oracle, Supermicrocomputer, and TSMC also suffered deep declines. That pushed several major US stock indexes down, though the Dow edged nearly 0.7% higher. The NASDAQ fell about 3%, and the S&P 500 declined about 1.5%. The reason? A Chinese artificial intelligence company called DeepSeek.

Last week, the company put out a paper that said it had created an accurate AI model that it trained for relatively cheap, about $5.5 million. Compare that to the $100 million to $1 billion it took to train a recent model from Anthropic, according to that company's CEO. The development has pulled the rug out

from under global companies riding the AI wave, including chip makers, infrastructure suppliers, and energy companies, as investors question the outlook for AI spending. This comes at the start of a busy week for earnings. Some of the biggest spenders on AI, such as Microsoft and Meta, are due to update investors. We'll see a bit later what explains the market's frenzy and what's to come, but first let's talk a bit about the company behind it all.

Asa Fitch, who covers semiconductors for the Wall Street Journal, joins us now. So Asa, how much of a surprise is this? Did this company just come out of nowhere? Well, it did and it didn't in a sense. For a lot of people, it did come out of nowhere. They'd never heard of this company DeepSeek. But DeepSeek has been around for a couple of years, driven by an entrepreneur in China who's actually a hedge fund guy

who does quant investing in China. What really has caught people's attention is the release of a new model last week by DeepSeek that could rival the performance of some of OpenAI's most sophisticated AI systems. So this was what really kind of led people to think, wow, there's something to this Chinese company and what they're doing and it could disrupt the entire AI ecosystem. What is unique about this model that they just published this paper on? The interesting thing about this model really is how little it needed

in terms of computing resources to do what it did. These companies like OpenAI, like the big tech companies, have been throwing billions upon billions of dollars around buying GPUs, these chips that are needed to train AI models. And what DeepSeek did suggested that you can do the same sorts of things with less. So that's kind of what has led to this kind of surge in interest in DeepSeek and the surge of worry around people who have invested in these companies at the forefront of AI. How did DeepSeek manage to come up with its product, given the U.S. sanctions on exports

of the most sophisticated chips to China? Well, that's complicated. They have a bunch of chips. First of all, they have around 2,000 of some of Nvidia's chips that are available in the Chinese market still. They're not the most powerful chips the company makes. DeepSeek also had a large cluster of Nvidia's earlier generation of chips before there were these export controls on China.

It's not really entirely clear how DeepSeek was able to train up this big model and do these things with apparently fewer chips. Some of the papers that have been published suggest that they spent much less, but there are a lot of gaps in those assertions. We're not entirely sure yet exactly how much computing power DeepSeq needed, but everybody pretty much is impressed with what they've been able to do nonetheless. What can we make of the fact that DeepSeq's code is open source? That's another thing that's really been impressive to a lot of people.

It's not one of these closed source things where everything's hidden behind a wall and they claim this performance and so on and so forth. People can kind of see what they did. They can check their work and they can see that they used all these techniques to do what they did and it seemed to have worked out. So that in itself is pretty notable. It gives them more credibility because it's all been done out there in the open. So given the sort of freak out in the broader tech sector, semiconductor companies today,

where do we go from here? Well, the real question is how much these techniques actually apply to other models. And if you look at this question about the broader impact, the answer really revolves around how much the rest of the world adopts the same strategies. I mean, it is open source, as we were talking about before, and that means that anybody else can use the same techniques. The work shifts toward actually deploying them. You know, if you have all these consumers in the world using these things, even if you don't need as much computing power to train these things, you're still going to need an increasing amount of computing power to maintain and to deploy them.

That was WSJ reporter Asa Fitch. Asa, thank you so much. Thank you. Coming up, markets have been rattled. Where might they go from here? That's after the break. DeepSeek's impact on the stock market today isn't just limited to tech companies. But what investors want to know is, will the slump stick around? Here to talk about it is David Uberty, who covers markets for the Wall Street Journal. David, what about this has

got the market so spooked? Well, the market has been riding so high over the last year or two, in large part because of a lot of this AI hype. And the big question surrounding it all has been how to price all of this. What is the actual value that AI brings to the broader US economy? And compared to earnings, you know, through corporate earnings statements and such, today is one of the big first pullbacks against that theme that we've seen over the last year or two. And it has raised in a really big way whether the market has overvalued some of these stocks. If there is in fact a cheaper, more efficient way to do some of these things that AI models have been doing, that is why we're seeing a bit of this correction

here in trading today. And furthermore, we're seeing sort of the broad tentacles of the AI trade throughout the US economy. It's not just, you know, the big tech firms. It's also the so-called picks and shovels of the AI booms. Obviously, Nvidia is the poster child for that. But companies that produce servers, that produce data centers, that produce cooling systems and power systems for data centers, nuclear power producers, even natural gas futures, those have all been affected by the sort of like

changing dynamics of this trade. So it remains to be seen how it will play out in the next couple of days. What explains this broader reach throughout the market? It has really gone to show how wide-reaching AI has been as a theme within Wall Street's sort of investment thesis and the broader US economy, which is very much digitizing and very tech-focused over the last few years in particular as well. And as I said earlier, a lot of these stocks detractors from them have been saying for some time that they've been overvalued. So in some ways, the market was sort of primed

for a moment like this, at the very least for them to be tapping the brakes to see what happens next with the next round of earnings reports. But it could also be a more fundamental sort of reevaluation of how these things are valued. So is this the correction that some investors were expecting, or does it look a little different? That's certainly what we will sort of understand over the next few days or so. Major companies are reporting earnings over the coming weeks. They will give us some insight into that and sort of answer these questions from the C suite for sure.

You know, more broadly, the bigger question is because markets have become so top heavy with a lot of these AI related stocks, if this does become this broader correction, you will see a broader reorientation of what's sort of happening on Wall Street. Even today with this sort of stock market sell off, you also saw a rally in government debt treasuries, which sort of suggests that investors are sort of fleeing for safety, at least in this particular moment in time. So it's an open question as to whether it will continue. That was Markets reporter David Uberti. Thanks, David. Thanks for having me.

Speaking of tech, the new Trump administration is shaping up to have profound effects across the tech industry, from regulation and policy, to AI investment and development, to several high-stakes antitrust cases against tech giants Google and Meta. In a special series that kicked off today, our tech news briefing podcast will be looking at what Trump means for tech. In the first episode, WSJ Global Tech Editor Jason Dean talked about some of President Trump's policy changes that could affect Silicon Valley. I mean, you can see some of them already. He's been very vocal about embracing policies that favor cryptocurrency and artificial intelligence. It wasn't like the Biden administration was anti-AI, but they weren't quite as fulsome

in their embrace. And obviously they were quite tough on the crypto industry. And that's a sector that venture capital firms in Silicon Valley have invested enormous sums in and have a lot at stake. I'm sure they hope that these antitrust suits they're facing might go away, or at the very least that there won't be

stricter enforcement or continued strict enforcement of antitrust regulations so that there might be an ability to do more deals, which can be beneficial both for the big companies and for the investors in startups who could see exits on their investments and have really been constrained in that over the last four years. That was Wall Street Journal's global tech editor, Jason Deane. And you can listen to our special series, What Trump Means for Tech, every day this week on our tech News Briefing podcast.

In other news, we exclusively report that a top prosecutor has opened an internal review of the Justice Department's decision to charge hundreds of January 6th defendants with felony obstruction offenses. People familiar with the matter said that acting U.S. Attorney Ed Martin, who was appointed by President Trump, asked prosecutors in an email today to turn over files, documents, notes, emails, and other information related to the cases. Justice Department spokespeople had no immediate comment. Trump pardoned virtually all of the January 6th defendants last week. A Rwanda-backed rebel group has entered the Congolese city of Goma, threatening a bloody

new chapter in a 30-year conflict set off by the 1994 genocide. The city of Goma sits on Congo's border with Rwanda and is a long-time international aid hub and safe haven for civilians. Its residents have reported gunfire and heavy shelling. Rwanda's foreign ministry denied that its forces were involved in the conflict. An extended battle for Goma could become an early test for the Trump administration and its willingness to intervene in a war involving one of America's closest allies in Africa. And that's what's news for this Monday afternoon.

Today's show was produced by Pierre Bienamé and Anthony Bansi with supervising producer Michael Kasmitis. I'm Alex Ossola for The Wall Street Journal. We'll be back with a new show tomorrow morning. Thanks for listening.