Speaker 1: It's the breakout star of the A.I. world. What we can call the deep seek sell off is still really pressuring tech that just put Wall Street on notice today. These fears from China's deep seek A.I. sparking the sell off in the market. Chinese artificial intelligence company Deep Seek sending American tech stocks into a nosedive as their new open source A.I. tool has become one of the most widely used and highest rated in the world. Unseating open A.I.'s chat GPT as the top free app in the U.S. Apple App Store. Side by side the competitors look and feel very similar. Both can help you plan a five day getaway vacation or even provide an easy to make dinner recipe. But here's where they diverge. Deep Seek's integrated deep think feature gives users a glimpse into its thought process. We asked both chat bots to tell us a joke that was safe for work. Ten words long and wasn't cheesy. Chat GPT replied. Why did the scarecrow succeed. He stood out in the fields. Deep Seek replied with time flies like an arrow. Fruit flies like a banana. We'll give you a second to think about that one. Deep Seek also showed its chain of thought sharing jokes. The A.I. thought about making but decided didn't meet our criteria including a scarecrow joke very similar to chat GPT that it deemed too cheesy. Deep Seek would not provide answer to some topics deemed sensitive in China. When we asked it what happened in Tiananmen Square the A.I. declined to answer. Instead saying it was designed to provide helpful and harmless responses. Chat GPT did provide answers to the same questions. Deep Seek says they train their model in just two months for less than six million dollars. Seemingly leapfrogging its American competitors like open A.I. Microsoft and Metta who have spent millions training their models and doing it despite U.S. efforts to block high power A.I. chips from the Chinese market. They were able to really do this on what some people would call a shoestring budget. Thank you. Less than a week ago the Trump administration rolled out

Speaker 2: a 500 billion dollar A.I. initiative called Stargate. What we want to do is we want to keep it in this country. China is a competitor and others are competitors. We want we want it to be in this country and we're making it available. Satya Nadella CEO of Microsoft which is

Speaker 1: participating in Stargate issued a warning about China's A.I. capabilities in an interview with CNBC from the World Economic Forum to see

Speaker 3: the deep seek new model. It's super impressive. I think we should take the development out of China very very seriously. But Sam Altman CEO of

Speaker 1: open A.I. seem to suggest deep seeks product is a derivative of his own posting after the chat bots launched last month. It is relatively easy to copy something that you know works. It is extremely hard to do something new risky and difficult when you don't know if it will work. Deep Seek has not responded to NBC News requests for comment. Brian Chung joins us tonight. Brian we now know where you get

Speaker 4: your comedy material. But on a serious note is Deep Seek actually made in a much cheaper fashion than Chachi PT. Do we know the answer to

Speaker 1: that. Yeah. Well look when it comes to that five point six million figure that is self-reported from this Chinese company. So take that as you will. Now when it comes to the scale of it though I want to point out that is so much cheaper than how much it cost the other companies. Open A.I. cost them about a 78 million dollars to train their models and Google. It cost them almost 200 million dollars to train their Gemini. So again five point six million is just really a drop in the bucket compared to those other figures. Although I didn't I

Speaker 4: didn't realize you thought my jokes were that cheesy. I mean Brian the news triggered a massive stock sell off right of NVIDIA which everyone had been falling all day. It makes many of the chips that power American A.I. models. NVIDIA shares look at this dropping nearly 17 percent today. More than 24 dollars a share. This graph doesn't really do it justice. That translates to a loss of nearly 600 billion dollars in market cap. What Bloomberg is reporting is the biggest market value drop in U.S. stock stock market history. And some of the indices a large part of which are tech stocks also took a hit. You see them right here. The question to you Brian how worried should investors and people who have a lot of tech stocks in their 401k be right now. Yeah I've been getting more questions

Speaker 1: about what's going to happen to my NVIDIA stock than I got actually about deep sea today. But look it's it's colossal when you consider that just the market cap loss that NVIDIA took today that's the equivalent of like an eggs on mobile the entirety of that company. So it's really big here. But I think one argument here is that NVIDIA stock has basically been going to the moon over the last two years. So this may have been one excuse for those that wanted to sell to ultimately sell. But I do want to point out that NVIDIA is such a key part of this story because they're the ones that manufacture the chips that really drive A.I. So if you do it a lot cheaply you don't have to buy as much of it. So that's maybe one reason why the stock went down today. Brian Chong always

Speaker 4: great to have you on for more on the impact of deep seek could have on the A.I. market and the future of technology. I want to bring in matching and he's a fellow at the Carnegie Endowment for International Peace. He's done extensive research on China's work in artificial intelligence. Matt you're a great guy to have on tonight to explain this is deep seek better than chat GPT.

Speaker 5: Deep seek is essentially on par with what open A.I. released in September of last year which is the 0 1 model. That's their own reasoning model that takes the time to think through its steps as we're dead as was demonstrated there. So you could say in some sense that deep seek is just a few months behind open A.I. It's important to keep in mind though that open A.I. has already sort of released a private demo of 0 3. They're updated model which appears to be performing at much higher levels. So China's really deep seek specifically has really closed the gap. But it's hard to say that it's an exact equivalent. Is Sam Altman right. Did China

Speaker 4: basically do what they've done with luxury goods what they do with pretty much any product that America manufactures. They make a

Speaker 5: copy of it and sell it cheaper. Is that what this is. It's not so much a copy as it took inspiration from what open A.I. did. And what Sam correctly pointed out there is that when you sort of know the destination you know what you're trying to create it's much easier to get there even if you're the second person trying to sort of hoe that road. But they're saying it costs them less than

Speaker 4: six million dollars to make. They did this in two months. Tech companies are saying it costs billions to invest in A.I. infrastructure here. They're asking for billions from the government. So I mean is this a cheaper model that does the same thing.

Speaker 5: The five point six million dollar number is really not capturing the whole picture. You have to spend a lot more to essentially up your skills and develop your your training data and a lot of things before that five point six million number comes in. What they've done is very impressive and they have done it much more efficient much more efficiently than a lot of the U.S. competitors. But to just look at that number and say sort of the game is over there way out ahead or we don't need to invest is it's

Speaker 4: just not right. Deep Seek is from what I understand is working with inferior chips. Right. And this is part of the crackdown that the U.S. put on competitors. But by doing that did they essentially force China's hand to make something maybe more affordable maybe

Speaker 5: better. That's exactly right. U.S. chip controls have big picture just constrained the total volume of compute in China. They've made it more expensive. So when you're sort of backed into a corner and you have to deal with these constraints then you're going to put all your energy into efficiency. And that's really paid off for them. U.S. companies essentially have the luxury of just throwing more and more compute at the problem. And so that's the route that they've gone down that might pay off big time in the long run. But it has forced deep seek to be more efficient. And that's probably going to pay dividends when it comes to serving the model globally in the race. What happens here now. In the big picture the U.S. still has a pretty strong lead. And I think this might lead the U.S. government to really double down on a specific area of chip controls that could hamper China. That's controls on the inference chips the chips you use when you're running the model as opposed to training the model. If the U.S. does that that will be a very unfortunate knock on effect for deep seek and for China in general. Big picture I would still be betting on the U.S. long term. The compute advantage is durable no matter how efficiently they're able to work with the limited compute they have. They just don't have as much that is going to pay dividends for the U.S. over a 1 2 3 year span. We've got about 15 seconds. Should people

Speaker 4: be concerned about what they put into deep seek because it is controlled by the Chinese government. If you're using the excuse me if

Speaker 5: you're using the app then yes that data is going to deep seek. If you download the open source model and run it locally then that all

Speaker 2: stays on your computer. All right. Matt pleasure to have you. Thank you for all that. Thanks for watching. Stay updated about breaking news and top stories on the NBC News app or follow us on social media.