STEVE INSKEEP, HOST:
We're going to try to understand the business plan for Anthropic. The AI company that produces Claude is preparing to go public, selling shares on the stock market. They released a prospectus this week, trying to explain how it could make sense for a company that has made billions of dollars to be worth trillions of dollars. Casey Newton is the co-host of the new NPR podcast Machine Gods, all about AI. So we've called him in. Hi there, Casey.
CASEY NEWTON, BYLINE: Hey, Steve. Thanks for having me.
INSKEEP: OK. So when we talk about a company that might be worth well north of $1 trillion, what does Anthropic say they plan to do that would make that pay off?
NEWTON: Yeah. So what Anthropic does is they sell intelligence. So far, that intelligence is mostly going toward writing software to writing code, and Anthropic has been able to sell intelligence for those two jobs quite successfully. The question and the sort of investor bet is can you expand that intelligence to many other sectors of the economy, and could you maybe even eventually start to automate away some entire jobs? Because if you can do that, then I think there will be a plausible argument that Anthropic would be worth trillions of dollars.
INSKEEP: I want to talk through the mechanics of that, though. As everybody is learning, you have to build data centers in order to have the compute - the computing power to do more and more with artificial intelligence, and that is costing trillions of dollars. What's on the line here?
NEWTON: Yeah. Well, as you note, you do need data centers to run AI models, and those have become very unpopular politically. At the same time, there are enough of them that now have already been built, and there seem to be enough in construction that there will be a really healthy amount of compute for the next few years.
And one of the things that we learn in the Anthropic prospectus is just that they've been able to sell that compute at a really good gross margin. You know, for every dollar worth of Claude they're selling, they're keeping about 85% of that. So if that remains the case, you know, they may be able to keep their books balanced and earn a healthy profit.
INSKEEP: Oh. You just told me some news here because there've been a lot of stories over the past year or so about how that compute was getting more and more expensive, to the point where people wondered if it would even be worth it for some tasks.
NEWTON: That's right. And many observers, I was reading this week, have been surprised at just how good the margins Anthropic has been able to make are. Now, where I would say there is some room for worry is that the reason that Anthropic has been able to sell at such a great margin is that it has arguably had the best model for most of the past year or two. There are many competitors in this market, and their models are getting better continuously.
INSKEEP: There, nevertheless, are some mind-blowing phrases in this prospectus. I was reading the letter by the founders at the top of it, and they admit in the letter - they say in the letter, we're going to spend money. We're going to invest money in such enormous amounts that it will seem, quote, "irresponsibly large at first glance." Wow.
NEWTON: Yeah. It's - you know, it's a really crazy thing about the AI business is that you have to buy compute many years in advance - those data centers that you just mentioned. They're ungodly expensive. They take a long time to build. And you're buying them assuming that demand will be there, but, of course, you know, you can't guarantee that in the moment that you're buying it. So I think that was a signal to shareholders, like, hey, if you want to buy our stock, you need to be in for a kind of long ride here because we are going to have to make that calculation.
INSKEEP: Meaning that if there's a bubble, if things go down, there could be some pain along the way here.
NEWTON: Absolutely. And, you know, the CEO of Anthropic, Dario Amodei, has talked about that because, you know, believe it or not, the company was getting criticized at the start of the year for not buying even more compute. And one of the things that Dario said was, look, like, if I had made that bet and been wrong about the amount of demand for what we make, we would be bankrupt right now.
INSKEEP: OK. I want to ask about another phrase in this letter from Amodei and others. They say that if everything works out, they could usher in, quote, "an age of miracles."
NEWTON: (Laughter) Wouldn't that be nice?
INSKEEP: Yeah. But, like, what if it only turns out to be a really good business and not an age of miracles? Is everybody going to lose their shirt?
NEWTON: (Laughter) Then I think there will be a lot of disappointed shareholders, yeah, and people selling their stock, among other things.
INSKEEP: Casey Newton is the co-host of the NPR podcast Machine Gods. Thanks.
NEWTON: Thanks, Steve.
A MARTÍNEZ, HOST:
And a disclosure - Casey Newton is engaged to marry a person who works at Anthropic. 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.
