LEILA FADEL, HOST:
The small Chinese company that might be about to burst Silicon Valley's AI bubble. The company is called DeepSeek, and it even caught President Trump's eye.
(SOUNDBITE OF ARCHIVED RECORDING)
PRESIDENT DONALD TRUMP: The release of DeepSeek AI from a Chinese company should be a wake-up call for our industries that we need to be laser focused on competing to win.
FADEL: The product was made on the cheap and is said to rival tools from companies like OpenAI, which created ChatGPT. In response to DeepSeek AI's release, tech stocks plummeted around the world. So is this company and its creation a threat to the United States' multibillion-dollar AI industry? To dive into what DeepSeek did and what it means, we've got Matt Sheehan with us. He researches AI and China at the Carnegie Endowment for International Peace. Good morning, Matt. Thanks for being on the program.
MATT SHEEHAN: Good morning. Thanks for having me.
FADEL: So what did this company do that U.S. companies like OpenAI and Google, what they haven't done yet?
SHEEHAN: DeepSeek's model essentially matches OpenAI's most recent model...
FADEL: OK.
SHEEHAN: ...In terms of performance. But the surprise and the challenge comes from the fact that they did it so quickly, so cheaply and so openly. So quickly in terms of they were able to match OpenAI's performance within just a couple of months after the OpenAI model was released. Cheaply in terms of spending far less computing power to train the model, with computing power being one of if not the most important input during the training of an AI model. And openly in the sense that they released this essentially open source online so that anyone around the world can download the model, use it or tweak it, which is much different than the more closed stance that, ironically, OpenAI has taken.
FADEL: And why did we see stocks react this way and, really, the companies here in the U.S. kind of freak out?
SHEEHAN: The fact that DeepSeek did this so quickly, and specifically openly, releasing it open source, is really a challenge to the business models that a lot of people have imagined for AI going forward. That includes for the companies that are trying to build and then sell access to their models, and it also includes the stocks of chip companies, semiconductor companies, like Nvidia. The assumption beforehand was that you need tons and tons, you know, tens if not hundreds of millions of dollars spent on access to chips in order to reach this kind of frontier of AI performance. DeepSeek has shown that you can achieve that for much cheaper, and that's got people nervous about the stocks of companies like Nvidia. And it's also representing a challenge to companies like OpenAI, or you could say Google with Gemini, any other frontier AI company that's trying to sell access to its model globally.
FADEL: I mean, how did this Chinese company do this, especially given that the Biden administration had banned the best AI microprocessors from being sold to China?
SHEEHAN: Ironically, those things might actually be quite interconnected, in that the Biden administration's export controls on the chips used to train AI have essentially backed Chinese companies into a corner. They are not entirely cut off from access to these chips, but they have much lower supplies. And so it's forced them to get very creative in how they can squeeze as much efficiency as possible out of those chips. And in that process, they've done it much cheaper, which led to the result here.
FADEL: Do you think there are going to be some similar concerns from U.S. officials around this company over national security that we saw over TikTok because it is owned by a Chinese company?
SHEEHAN: Yes, almost certainly. And that kind of comes in from a few different angles. You have the quite direct concern about data privacy, about whether or not, you know, Americans interacting with, say, the DeepSeek app, whether or not their data is going to China and then could be accessed by the Chinese Communist Party. But you also have the more sort of macro level concern about what does this say about where the U.S. and China stand in the race or the competition to build the most powerful AI systems?
FADEL: Matt Sheehan is a fellow at the Carnegie Endowment for International Peace specializing in artificial intelligence and China. Thank you, Matt, for your time.
SHEEHAN: Thank you. Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
