AI's Sputnik moment
15:35 Share

This week we saw a huge leap in AI technology. It's caught the tech titans in the US by surprise and freaked out investors since it threatens the underpinnings of the entire economy. DeepSeek, the Chinese-owned AI company, says it's made huge strides with far fewer resources than its Western competitors. On this weekend episode of Reuters World News, our tech correspondents tell us how this happened and what happens next.

I'm your host, Jonah Green. Save on Cox Internet when you add Cox Mobile and get fiber-powered internet at home and unbeatable 5G reliability on the go. So whether you're playing a game at home or attending one live,

You can do more without spending more. Learn how to save at Cox.com slash internet. Cox Internet is connected to the premises via coaxial cable. Cox Mobile runs on the network with unbeatable 5G reliability as measured by UCLA LLC in the U.S. to age 2023. Results may vary, not an endorsement. Other restrictions apply.

I'm joined now by tech correspondents Jeffrey Dastin and Anna Tong. Hello, you two. Thanks for joining me today. Thanks for having us. Thank you. Who is DeepSeek and how did they do this? DeepSeek came out of a hedge fund actually in China. It was a quantitative hedge fund focused on using mathematical algorithms for trading as opposed to human analysis. And it had been around since 2015 when its founder,

someone named Liang Wenfeng,

said to his followers on China's WeChat that he was going to focus on building AGI. So AGI, for its artificial general intelligence, it's the stated goal of the major US AI providers like OpenAI and Anthropic. And so what was really, really interesting about this announcement in April 2023 that Liang would create this AGI-focused company we now know as DeepSeek,

is that he hadn't come from the major tech players in China. He hadn't come from Alibaba or ByteDance. He was at an elite university when he was 17. He majored in electronics communication engineering. But he had come from obscurity relative to, for instance, the Jack Ma's of the world. And here now, fast forward almost two years later, he...

with DeepSeek has completely rocked the world of AI and global markets at the same time. What's crazy is kind of like the whiplash that we're experiencing in the economy right now. Just two weeks ago, Sam Altman, Larry Ellison, and Masayoshi Son stood up with President Trump and said they were going to invest up to $500 billion into infrastructure that was essentially only for open AI. And then just a week after that, DeepSeek released its models, which then made the

The market and business people think that you didn't really need to have a huge investment in AI infrastructure. And that caused NVIDIA to lose hundreds of billions of dollars in market value. So I think that just like the amount of whiplash that the market is experiencing that AI investors are experiencing right now is crazy because it's gone from like, we need $500 billion of investment to maybe we don't need anything at all to now...

a lot of uncertainty around what's going to happen. And I think a lot of investors right now are kind of looking around to try to figure out what they should do, if they should pull back from AI investments or if they should double down because deep seek means that AI will be more freely available to everyone. So I don't think anyone knows what's going to happen. And so how are the other AI companies responding to all this? The response of the USA AI companies was interesting because they acknowledged that

how powerful DeepSeek's model was and how cost-effective, while at the same time downplaying the concerns that one might have from this. So to give an example, the CEO of Anthropic, one of the top AI companies backed by Google, backed by Amazon, his name is Jario Amodei, and he published a very long blog post basically saying that what the US has to do, and we can talk about this later, is just keep

its strategy with export controls of chips and preventing them from getting to China. Easy thinks they're working, actually. But the subtext to all this, what was the actual response to DeepSeek, was that we were, according to M.O. Day, at this interesting crossover point where it's actually possible with not a ton of money to produce really high-quality chips.

results and that that would change and you actually will need a ton of money and high-end chips going forward so he was he was basically saying that deep seek had done something very impressive engineering wise

And it had done clever things with various AI research techniques to very efficiently make a model that could do impressive reasoning. And that was almost to be expected based on the way this technology was progressing, the cost savings we've seen month after month. And he was kind of comparing DeepSeq's models to his own, Anthropic's own models from month

months prior and saying, this is exactly what we would expect to happen. We would expect it to be done so well and so cheaply. And we're at this crossover point where that's possible now. But if you fast forward some number of months, some number of years, you actually will need huge numbers of high-end chips and DeepSeq won't be able to keep up. So he actually presented this image of confidence

common certainty, of course, that's doing what it is. And really what surprised him was just that it had come from China, not that someone had produced results like DeepSeek's at all. So I think that is indicative of maybe how the US companies are working are confident, but still concerned from a geopolitical risk standpoint.

Yeah, that was kind of the heart of the panic this week, right? If a company like this can do so much with so little, then that kind of overturns all of our assumptions about

how valuable a company like NVIDIA is or how much you need to invest in a company like OpenAI. But what you're saying is that maybe that's not the case going forward. We saw NVIDIA's stock drop 17% at one point on Monday. And certainly the question that people were asking was, do we need to spend so much money on NVIDIA's highest-end chips if this company in China, without access to them, was able to reproduce

these impressive results without those chips. And obviously, Nvidia stock has paired some of those losses now. And I think it is an open question. Going back again to what some of the USAI leaders are saying, this is a temporary moment in time. It's the case that going forward to keep scaling up these

these models and keep getting better, more intelligent results that get us to this future that the deep seek itself claims it wants, which is artificial general intelligence, really human equal intelligence, then you will need still more powerful NVIDIA chips and lots of them, tens of billions of dollars, if not a hundred billion dollars or more. And that actually suggests that NVIDIA still has a good business going forward.

And Anna, you've been talking to a lot of sources about this. What are investors thinking right now? For sure. Like, you know, Masa Yoshisan from SoftBank is certainly doubling down on OpenAI. So he's not only investing in Stargate, which is the joint venture that's going to be building data centers and purchasing chips, but he's also making an up to $40 billion investment into OpenAI's foreclosures.

for-profit company, which changes on whether an open AI can convert from a non-profit to a for-profit company. So there's certainly investors like him that are really, really doubling down. On the other hand, we're hearing also kind of rumblings from investors that maybe think that they are overexposed in AI and want to dial it back. So I think we're hearing both right now. Have you heard if...

any of the American companies are changing course at all on their strategies or it doesn't seem like what the Anthropic CEO is saying that they're just staying the course and using this as a way to get people to

want to spend even more on American AI to stay ahead. Next week, there is going to be a slew of tech earnings, and there were tech earnings this week too. What we're hearing so far is that companies are staying the course. So in Meta's earnings, Mark Zuckerberg said, doubled down and said they are going to continue investing in AI infrastructure. Same with Microsoft.

So I think that thus far, companies are at least publicly saying that they are excited about DeepSeek. And that means that there is going to be more AI for everyone. Because right now, getting AI out into the world, into the hands of consumers has been extremely limited by the number of GPUs that are available. And so there's a lot of excitement now that with the efficiency gains that DeepSeek has proved, that there will be more AI.

AI available for everyone. Can you explain what GPUs are? A GPU is a computer chip that specializes in running AI and also training AI models. And it stands for Graphics Processing Unit because NVIDIA created this chip actually for video games. And a bunch of AI researchers realized more than a decade ago that the way it

processed computations in parallel, different from how computer processing units, TPUs, did it, was actually perfect for the kind of AI that the world and those scientists are now developing. This news comes at a very interesting time. The U.S. is trying to ban TikTok, a Chinese-owned company,

And that's been put on hold by President Trump, despite years of lawmakers warning that the app is a threat to national security and that China is using it to corrupt the minds of our young people.

And now we have China potentially dominating the AI space. Right. So there have been reports of Yang Wenfeng, the DeepSeek founder, meeting with senior government officials in China. And clearly his presence there is an indication that China, as it has said for years, understands the huge economic value of AI. And what's interesting, if you talk to or hear from

from the American AI founders. It's not that they don't want China to have the benefits of AI. They want those to be distributed across the world. It's just that there is concern in the United States about the government model of China being in the driver's seat as opposed to the democratic model in the U.S. So there is this idea, however alarmist as it might be, that if the U.S. isn't the first to get to AI...

and in particular to get to this artificial general intelligence, this way more powerful type of technology that we don't have yet, then there could be great benefits to China's military. There could be a reshaping of the geopolitical world order, that sort of thing. So there's this belief that the US has to get there first. And this is the first real challenge to the likelihood of that. And so I think that's why people

at least in government circles and national security circles in the U.S. are alarmed. And so what does this mean for the race to dominate AI? What does the U.S. do next here? What...

We're starting to see our calls for still more export controls on chips, which there already are a lot. But the idea is to maybe prevent some of those lower-end types of NVIDIA chips from getting to China, the ones that are already there, but preventing further sales of them, because clearly the DeepSeq model shows that even some of those can be impactful for building AI in China. And I think...

Beyond expert controls, there still is a lot of head-scratching, what's the defensive, what's the offensive move for the US? I think, as Anna mentioned, having Trump present or lead into this announcement of this $100 billion up to $500 billion target announcement involving data centers for OpenAI suggests that there's this desire, aside from just preventing China's access to chips, there's this desire to let the US industry run freely. So

I think the expectation is that some of the requirements and regulations suggested already by the Biden AI executive order, which Trump has gotten rid of, that will be more of the future, that there will be less hand-holding, less burdensome regulation just to allow U.S. companies to win that race as fast as possible. Thanks to Anna and Jeff for coming on and tech-splaining all of this to us.

Reuters World News is produced by Gail Issa, David Spencer, Sharon Reich-Garson, Christopher Wall, Jasper, and me, Jonah Green. Our senior producers are Tara Oaks and Carmel Crimmins. Our executive producer is Lila DeKretzer. Musical composition and sound design by Josh Sommer. We'll be back on Monday with our daily headline show. To make sure you never miss an episode, click follow on your favorite podcast platform or download the Reuters app.