MARY LOUISE KELLY, HOST:
These first days of the second Trump administration are bringing big changes to a lot of things, including the business of artificial intelligence. Day 2 brought the announcement of Stargate - that's a $500 billion AI infrastructure project. It's a private venture, though President Trump unveiled it at the White House and is throwing his support behind it. Then there's all the hullabaloo about the Chinese startup DeepSeek, and today, OpenAI, the company behind ChatGPT - also a big part of Stargate - announced it is partnering with the U.S. National Laboratories to supply the government with advanced models for things like cybersecurity, energy infrastructure and nuclear security. Well, Chris Lehane is OpenAI's chief global affairs officer. He's here in D.C., where OpenAI made this announcement today. Chris Lehane, welcome.
CHRIS LEHANE: Mary Louise, thanks for having me.
KELLY: Start with this new partnership with the National Labs. In a few sentences, why is this a big deal and what is the time frame for it bearing fruit?
LEHANE: Yeah, we do think this is a really significant announcement. In short form, it is OpenAI bringing its leading-edge innovation, leading-edge AI technology, into the United States National Labs. It will be critical as those labs move forward in the development of research and development. It will support national security imperatives. But at the end of the day, what it's really bringing is our leading-edge tech, the innovation that we're building, and making sure that we're doing it in collaboration, in partnership with the U.S. government...
KELLY: Yeah.
LEHANE: ...The labs, so...
KELLY: Let me turn you to Stargate, this huge, new $500 billion joint venture that has just been announced.
LEHANE: Yeah.
KELLY: OpenAI is a key player in Stargate. As simply as you can explain, what is Stargate and why do we need it?
LEHANE: Stargate is infrastructure, is destiny. At the end of the day, the U.S. and the People's Republic of China are in a race, a competition, to see who ultimately leads in AI. Right now, the U.S. has a lead. It's not a huge lead, but we do have a lead. Ultimately, what will determine who wins this competition, who wins this race - and the stakes could not be bigger - it really is, are we going to build the world on democratic, free AI or is it going to be authoritarian, autocratic AI? And what will decide that is something called compute. And compute is, if you add up the following pieces - chips, data, talent and energy - and if you put that full stack together, that represents infrastructure. It's a 21st century AI infrastructure. And so what we announced last week with President Trump, with our partners at Oracle and SoftBank, is a $500 billion investment in U.S. AI infrastructure to generate that compute, which will help ensure that the U.S. maintains its lead in AI.
KELLY: It sounds lofty. Will Stargate do something, create something useful to the average American?
LEHANE: Yeah. For a typical American, first of all, there's going to be hundreds of thousands of jobs created just by the construction of this. We've already begun a project in Abilene, Texas. There will be other states that will get pulled in. But what this also means, because there will be access to compute, is that the technology that's being generated by AI will be accessible in a much broader way to many Americans. Think about education - bringing AI into schools to give teachers tools to help educate kids even better. Think about health care, the ability to make sure people have access to health care that will potentially have significant impacts and the ability to diagnose ailments, challenges, diseases at a much earlier stage. Think about this in science. The deep research capacity is going to allow for incredible advancements.
KELLY: Does it need to be this expensive? Five hundred billion dollars is a lot of money, and DeepSeek, the Chinese AI startup, has prompted a conversation about whether foreign rivals are doing things a lot cheaper than American companies.
LEHANE: So first of all, I think when you think about DeepSeek, I do think you have to acknowledge it's a pretty impressive tool that was released. Now, we released a version of this four or five months ago, and we'll continue to release updated versions where we're demonstrating that OpenAI is helping the U.S. to maintain its lead. I think we'll also learn a lot more information over the next couple of weeks that may reshape our thinking about exactly what we know about the model.
KELLY: But at this point, you see it as a worthy competitor?
LEHANE: Oh, yeah. I mean, I think - yeah, no matter what, like, I think this makes really clear that there is a real competition and a real race. And so when you talk about the $500 billion, I think really important to understand this is not U.S. taxpayer money. This is not public money. This is private sector money because there's an understanding that as AI continues to move forward, there's going to be more and more and more of a demand for the AI, which then requires more and more and more compute.
KELLY: Just to stay with this week's news...
LEHANE: Sure.
KELLY: ...'Cause there's a lot of things to keep up with here.
LEHANE: Yeah.
KELLY: DeepSeek, the Chinese startup, which, I will note, dethroned ChatGPT - your AI app - this week as the most downloaded free app in the U.S. Apple Store. Your company says it is investigating whether DeepSeek inappropriately used OpenAI data. Where does that stand?
LEHANE: Yeah. So we're in the process of reviewing to understand totally what may or may not have occurred. There is something that happens in the AI world called distillation, and distillation is a complex idea. But, you know, effectively can people send a lot of stuff at your models and be able to take information out of it and then use that information to replicate something else? There's also a version of distillation...
KELLY: I did want to ask...
LEHANE: Yeah.
KELLY: ...What is the difference between a rival company like DeepSeek possibly accessing your data and you training your own products like ChatGPT on other people's work without their permission?
LEHANE: And so we - so it's important to sort of bifurcate, you know, where you're doing distillation and someone is taking information and exactly replicating what you do versus distillation. Like, we provide things through our API and other sources so developers can actually distill our tech and build off of them. Those are two very different things. I mean, maybe a way to think about it is I go to the library, borrow a book. I can read that book and return the book. I have certain information - great. Something else, if you go to the library, take the book, never return the book, put the book out under your name, right? Those are sort of the two different ways to potentially think about this. But, yeah, I think...
KELLY: Scarlett Johansson would have thoughts on this.
LEHANE: Yeah. But I think the big point, I think, in all of this is that there is a real competition between the U.S. and China.
KELLY: Have you tried DeepSeek?
LEHANE: I personally have not. I've seen it, and I think amongst the things that they did and I thought was smart was they put out what's called chain of thought, so you can actually see how the technology is doing what we call reasoning. You know, OpenAI released something called o1 back in September that has the same tech. We actually think ours continues to be ahead, and we're going to be releasing some additional models, including o3, which comes out on Friday. So I think important to keep in mind, you know, we as the U.S., because of OpenAI, continue to maintain a lead. We need the infrastructure because infrastructure is destiny here if the U.S. wants to win this race.
KELLY: Chris Lehane, chief global affairs officer at OpenAI, thanks very much.
LEHANE: Mary Louise, thanks so much for having me.
(SOUNDBITE OF MUSIC) 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.
