What an Anthropic IPO Could Mean for the Bay Area

Airdate: Wednesday, October 28, 2026 at 9 am
Anthropic has filed to go public in what could become the largest IPO in history. The San Francisco-based AI company’s 335-page filing offers the first detailed look at its finances, growth ambitions and stated risk factors, including the possibility of catastrophic incidents. We’ll talk with San Francisco-based tech and business reporters about what Anthropic’s filing can tell us about the future of the AI economy and what the expected huge influx of money could mean for the Bay Area.
Guests:
- Berber Jin, AI enterprise reporter, The Wall Street Journal
- Sri Muppidi, business and technology reporter, New York Times and DealBook
- Mat Honan, editor in chief, MIT Technology Review
Episode Transcript
This is a computer-generated transcript. While our team has reviewed it, there may be errors.
Alexis Madrigal: Welcome to Forum. I’m Alexis Madrigal. Anthropic is one of those San Francisco stories that’s hard to believe. A group of researchers, including two siblings, a former journalist, and a handful of others, broke off from OpenAI to found their own AI company in 2021. Five years later, the company has grown its revenue faster than any company ever. To take one example, in the second quarter of last year, the company brought in $787 million — not bad for a startup. But in the second quarter of this year, Anthropic brought in $11.6 billion. That’s nearly 15 times year-over-year growth. To put that in perspective, Google, which was the hottest tech company on Earth from 2002 to 2005, grew revenue from $400 million in 2002 to $6 billion in 2005. Sure, there’s been inflation, et cetera, but Anthropic’s growth has been absolutely absurd, unprecedented, and, for many people, unexpected.
Oh, and they made almost a billion dollars in the second quarter, their first profitable quarter. So here we are now: this company, which believes itself to be building toward artificial general intelligence, has money pouring in and revenue from corporations that use its tools. But because it needs to keep competing at the frontier of AI, it needs even more money to keep growing, and hence the IPO, which could raise $100 billion — the most ever.
Here to break down what the IPO tells us about Anthropic, the state of AI, and the tidal wave of wealth cresting over our fair city, we’re joined by Berber Jin, AI enterprise reporter at The Wall Street Journal. Welcome.
Berber Jin: Thanks for having me.
Alexis Madrigal: We’re joined by Sri Muppidi, who’s business and technology reporter at The New York Times and DealBook. Welcome, Sri.
Sri Muppidi: Thanks so much.
Alexis Madrigal: And we’ve got Mat Honan, editor-in-chief of MIT Technology Review. Welcome, Mat.
Mat Honan: Thanks, Alexis.
Alexis Madrigal: Was Casey Newton busy?
Mat Honan: Yes.
Alexis Madrigal: Yeah, Kevin Roose and Kate Snow are building something new. So, Mat, you’ve been in the tech journalism game here in San Francisco a long time, and you’re looking at this Anthropic document. What are you thinking?
Mat Honan: I mean, first of all, I’m thinking this is the longest S-1 I’ve ever seen. You know, you see a lot of these things over the years, and they’re always dense, but there’s often really entertaining stuff in there. This one was so long that I didn’t — I, at best, could skim it.
Alexis Madrigal: He’s got a summary from Claude.
Mat Honan: I actually — I was about to say, I used Claude to get a summary of some of the high points there. But it was, you know, it’s a staggering document, and what it means, I think, is kind of unprecedented for San Francisco. I think we’re going to sit here and talk today about what’s going to happen in San Francisco, but one of my takeaways from that 335-page document is that nobody knows what’s going to happen in San Francisco or anywhere else because of just the scale of it. The scale of it is, you know, like when we talk about the Uber IPO or Slack or Facebook or anything else that’s come before, they’re all so much smaller.
Alexis Madrigal: Sri, can you just walk us through some of the numbers here? You know, how much is Anthropic hoping to raise? What could the company be worth, that kind of stuff?
Sri Muppidi: Totally. So the company is aiming to raise, or is expected to raise, about $100 billion as part of its IPO. And the valuation that’s been touted around is between $1.5 trillion and $2 trillion. That really comes down to what investors expect the company to be worth.
Alexis Madrigal: I love it. It’s just like, it’s so much money, and yet that’s a $500 billion difference. If you were like 1.5, 2, you know, it’s like a house in San Francisco. You’re like, “You know…”
Sri Muppidi: Totally. What’s the over-asking price in terms of how much the company can actually go raise in the public offering? And so I think that’s really striking, just because Anthropic already has $130 billion-plus in cash and other investments that it can turn into cash. And so the fact that the company could raise, or is hoping to raise, $100 billion more really shows the size and scale, as we were just chatting about. And I think that really comes down to this figure of more than $807 billion in commitments that the company has in terms of being able to build out data centers, as well as have more computing power to actually run and build its AI.
Alexis Madrigal: That’s like, yeah, if you looked at a table of all the people they owe money to and you added it up, it’s like $800 billion.
Sri Muppidi: Exactly.
Alexis Madrigal: And I love, in the S-1, they say at some point, it may seem crazy that we’re spending all this money, but it’s because we really believe that there’s exponential growth necessary in what we spend so that we also can get exponential growth in what we make. There’s also this interesting figure, and maybe, Berber, I don’t know if you want to take this one. The company says they lost $80 billion in the first half of the year, but a lot of people have — some people have quoted that number. But it’s this really kind of odd convertible-debt thing where, essentially, Amazon invested. Can you talk a little bit about that? They didn’t really lose $80 billion, did they?
Berber Jin: No, they didn’t. Actually, Sri might know the specifics of this better than me. But my understanding is that that $80 billion net loss isn’t part of the core business. So it’s not like they’re losing $80 billion from their core operations, building and selling Claude. That would be very bad and would make the company much less attractive as a business to invest in. So I think it’s a little bit technical. They raised a lot of money through convertible notes, which eventually convert into equity but haven’t yet. And so, for some tricky, complicated, wonky accounting reasons, that shows up as a net loss. And the reason why that number is so inflated is because, actually, as Anthropic’s valuation has gone up, I guess the liabilities or the losses tied to those convertible notes have also gone up. So I think a lot of investors are also kind of not really looking at that. There are a lot of debates about profitability — how do AI companies calculate profitability? Is Anthropic kind of taking shortcuts to claim this profitability number from Q2? Because it is very impressive.
Alexis Madrigal: Mm-hmm.
Berber Jin: Um, but I think — I think I feel like it’s safe to say that that $80 billion—
Alexis Madrigal: Yeah, yeah. Well, and even they say at the same time, right, that they have something like — they were saying 85% gross margins for serving up Claude and Claude Code specifically, right? They’re able to make that much money there. But there are some questions about that, too, right? Sri, do you want to talk about how — what does this business look like?
Sri Muppidi: Totally. So the questions about gross margins — that basically means revenue minus the direct cost to produce that revenue. And so Anthropic accounts for that as what’s called inference cost, so essentially running its AI models to actually serve customers. So 85% gross margins is actually great for an AI lab, particularly because the company had, I think, negative margins even just two years ago, and very, very low margins even just last year. And so the fact that it’s been able to really improve its efficiency shows that they’ve been focused on this. And just for context, to put it alongside other software companies, that’s in line with software-as-a-service companies, anyway. So I think that’s quite impressive.
I think what folks are really indexing on now, in terms of where the company can go in the future, is that Q2 was really impressive. But the company actually shared some of its unaudited Q3 financials at an investor day a few weeks ago. And revenue growth increased, I believe, 45% between Q2 and Q3. But it shows that it actually decelerated compared with earlier this year. And so that kind of raises questions for some investors about whether Anthropic can actually continue sustaining that growth, particularly as you think about the threat of Chinese open-weight models.
What I’m hearing anecdotally is that businesses, especially large enterprise companies, are starting to adopt more open-weight models or turn to OpenAI for its Astra model, just because it’s gotten more competitive. OpenAI is also noting that this is an opportune moment to maybe cut prices, and so they’re gaining competitive share that way. I think there are a lot of questions around whether Anthropic can continue growing, just because the scale, as Mat was alluding to, is so large.
Alexis Madrigal: Law of large numbers, as I feel like they like to say. Mat, what do you think about the big question here? People have been trying to ask, “Is AI for real? Is this a boom that’s going to bust?” What do you think this document says about that larger question — where the whole industry is likely to land and its overall importance?
Mat Honan: I think you can answer the “Is AI for real?” question in two ways. One is: Is the technology real? Clearly, it is. There’s no question. If you’re doubting that AI technology is real at this point—
Alexis Madrigal: It’s a question from—
Mat Honan: —a few years ago.
Alexis Madrigal: Yeah. I don’t know what to tell you. Is the stock market overinflated? Is there a bubble? Is that the question about whether AI is for real? Is AI really going to be able to generate these returns? You know, I mean, I hate to be cautious in this way, the way that I was at the top of the show, but I don’t think anybody knows. I mean, I think anybody who’s really projecting what you’re going to do with the — to use the SpaceX term — largest addressable market on Earth—
Alexis Madrigal: Which is all human labor, apparently.
Mat Honan: It’s a great unknown, and I guess I personally, just because I’ve been reporting for however many years now, tend not to like to put my chips on the table about that kind of stuff. But if I had to guess, it seems like a not-bad business.
Alexis Madrigal: I mean, it certainly seems more real than a year ago. If we were having the same conversation and had somehow gotten access to these numbers last summer, we’d be like, “Hmm, they seem to be losing a ton of money.” They’re growing fast, but of course it’s not that much money against the idea of this massive — you know, even back then they were talking about a valuation that was hundreds of billions of dollars. What do you think, Berber, about the larger question?
Berber Jin: Yeah, I would tend to agree with what Mat said. I think, based on this document, Anthropic is an extraordinarily good business. And I think the fact that they were able to turn an operating profit in Q2 defies a lot of the narrative that people have had about just how AI businesses work. The narrative around the labs in particular is that they are money pits, that they’ll never achieve profitability. They spend all this money training the models, but it’s very unclear whether they’ll ever get an ROI from it. So I think the fact that Anthropic’s business is so good is something that, a year or two ago, you’d be very surprised by. I think everyone would be surprised to see the numbers coming out of the company.
Another statistic that really popped out to me in the S-1 was that they’ve raised $141 billion, and they still have around $120 billion on their balance sheet. Those are extraordinary numbers. Normally, we would expect them to be burning through the money they raised.
Alexis Madrigal: Exactly.
Berber Jin: And that’s what OpenAI is doing, by the way. We can talk about how much of a better business Anthropic is than OpenAI right now. But I do think there are a lot of weird things about the business that should give some people pause. For example, I think they said 5% to 7% of their revenue comes from users who violate their terms of service. They’re alluding to Chinese companies that are just distilling their models, trying to hack into the models to train their own models. A lot of their revenue comes from one or two big customers, so there are still a lot of question marks that should make us cautious about putting all our chips into Anthropic.
Alexis Madrigal: We’re talking about the Anthropic IPO with Berber Jin, AI enterprise reporter at The Wall Street Journal; Sri Muppidi, business and technology reporter at The New York Times and DealBook; and Mat Honan, editor-in-chief at MIT Technology Review. Are you more scared of a bust or a boom? 866-733-6786. Email [email protected]. Stay tuned.
Welcome back to Forum. I’m Alexis Madrigal. We’re talking about the Anthropic IPO. We’ve got a little roundtable of technology reporters here from The New York Times and DealBook. We’ve got Sri Muppidi, Berber Jin, and, from MIT Technology Review, Mat Honan. And we want to hear from you. You know, you’re a Bay Area person listening to this. Are you more scared of the tech boom here, or the tech bust that could come at some point in AI? 866-733-6786. That’s 866-733-6786. Email [email protected]. You can find us on social media, Bluesky, Instagram, and Discord. We’re KQED Forum. So, Mat, do you think we’re going to feel this IPO?
Mat Honan: I mean, I think we’re already feeling it a little bit in real estate. There’s been a bunch of great reporting about the ways it’s — something, including in the Journal. Maybe, Berber, this is one of your stories. Forgive me if it was. But there was a story recently about how there are 600 or so employees who’ve already been able to pull money out, and they’ve been investing it heavily in real estate. If you look at some of the home prices, look at how much they’ve gone up year over year, you’re definitely already seeing it. They’re definitely really feeling it. Like, what’s that going to mean for the corner stores and the small businesses in the city? I don’t know, but it’s showing up in real estate.
Alexis Madrigal: Yeah. I mean, normally, when I’ve talked about IPOs in the past, there are all these lockups on the options and all this stuff that’s happening. But it seems to me like Anthropic has made it unusually easy for their employees to trade, basically from the very beginning, right?
Sri Muppidi: Yes, exactly. So typically there’s like a 180-day lockup where companies’ early employees or others in the company have to wait. You might see an adjustment in the IPO price after, for example, that first pop. What’s unusual about Anthropic is that some shares can actually be traded right away, and then it’s staggered across a set of 37- or 60-second periods, I believe, in terms of when others can also start trading. And so that makes it possible for employees and other shareholders to start trading immediately. You’re going to see the impact of that right away.
What’s expected is that the company would actually start trading in the week of November 16, I believe, so right before Thanksgiving. Maybe if you’re making any big purchases — if you’re trying to buy a house, maybe do it right now. But you’re already seeing that impact, as Mat was saying, because there were these share sales that employees had already participated in, and they were able to sell some of their existing shares — for example, to buy a house. So I think you’re starting to see a trickle of it already, and it will continue to get worse.
Alexis Madrigal: I mean, when we’ve done real estate shows, it seemed like the major impact was at the ultra-luxury level. You know, it’s hard to feel bad for people who might be getting priced out of Pacific Heights mansions. But I suppose in this case it’s going to spread to some — what do we think, 1,500 or 2,000 people or something like that around here? Do we know the exact number of employees in San Francisco?
Sri Muppidi: I think it’s about 3,000, but you’d have to fact-check me.
Alexis Madrigal: And most of them are here, though, right?
Sri Muppidi: Most of them are here. It’s San Francisco headquarters. The other note I’d add to that is that rent prices have also gotten so crazy because Anthropic and other AI companies are all in San Francisco now. So when, for example, a young 20-something is moving to San Francisco and sees that rents are like $6,000 for a two-bedroom, they might as well just buy an apartment if the mortgage costs about the same as rent. What I’m hearing, even anecdotally, is that people are willing to buy not just those uber-rich mansions, but also a two-bedroom apartment or condo, because rents are so crazy as well.
Mat Honan: Alexis, I live in it. This is totally anecdotal. I live in the Outer Sunset, though. These neighborhoods are full of these things called doldrums, which are, you know, like this guy — Doldrum built, like, four guys on the first floor. They’re all pretty much the same floor plan, you know, two bedrooms with the garage downstairs. When I moved in there, they were selling for under a million dollars. There have been several in my immediate neighborhood that have sold recently for $900,000 over asking. And you can’t say, “Oh, well, is that because of Anthropic?” But it certainly seems like it. I mean, these are not high-end homes.
Alexis Madrigal: What’s wild is, you and I, in years past, were talking about the doom loop. I don’t think either of us were believers in this so-called doom loop, that San Francisco was going down in flames. And here we are, just a few years later, and this has happened now. Do you think this is a net good for the city, to get this massive injection of money into what is a tiny percentage of the city’s population?
Mat Honan: Well, it’s not going to show up in our list — the city’s not going to get a whole lot of tax revenue from it, right? There’s no income tax, and it’s questionable to me how much of that is actually going to go to the city. They are going to have this huge philanthropic division — LLC, what is it? Is it separate?
Sri Muppidi: Long-term trust.
Mat Honan: That’s going to be giving away all this money to philanthropic causes, but that’s probably not going to be local stuff, right? That’s going to be national, international, global health.
Alexis Madrigal: Right, right. More Gates Foundation-y level and scope, yeah. So how’s it going to benefit the city? I don’t know. Yeah. All right, let’s ask Jay in Palo Alto. What do you think, Jay?
Jay: Hey, good, good morning. Thanks for taking my call. I love that you guys talked about the housing issue a little bit, but I think that one thing — there have been several news articles about this recently — that’s united the whole country is that everybody hates data centers, right? And that’s exactly what this is. When we try to have this objective conversation about how cool it is, the ins and outs, and talk about it in a very matter-of-fact tone, I think there’s a tremendous disconnect between these facts you’re talking about and the fact that this represents something that almost everybody hates. I think we’re doing a real disservice when we neglect that. Thank you.
Alexis Madrigal: Thanks for that, Jay. I mean, if you look at the polling on data centers, it’s sort of the inverse of support for marriage equality. It is just a fast, hard move against data centers. This actually does — I mean, they address it in the risks in the S-1, but I think maybe it’s a little underplayed, right? Because, at least from my read on the data center industry, they’re already having a lot of problems. Do you want to talk a little bit about that, Berber?
Berber Jin: Yeah, I think it’s been really interesting to see, in the last few months, how data center build-outs have become a real political hot-button issue. It’s really turned into a liability for candidates who have supported data center build-outs in the midterms. It’s a great point to raise because they’re not building data centers in San Francisco. They’re building them in the rest of the country, oftentimes in communities that don’t really see the benefits of AI up front.
I think there’s this weird thing where it is a huge risk, right? Because if Anthropic can’t get the political goodwill to build data centers, they don’t have the compute needed to build and serve their models. But they’ve also been behind this very weird messaging around AI. I feel like they’ve marketed it as scary technology that already is having all these cyber risks and could potentially cause extinction. They have a lot of employees who are doomers. So there’s this weird cultural disconnect where they have a vested business and financial interest in trying to get the country to come around to this vision of building hundreds of billions, trillions of dollars’ worth of data centers. But they have this weird ideology around—
Alexis Madrigal: They’re like, “Don’t worry, we hate AI so much that we must build it and become trillionaires from it.”
Sri Muppidi: Exactly.
Alexis Madrigal: And it might kill you.
Sri Muppidi: And in case anyone else builds it—
Alexis Madrigal: Yeah. I mean, they do try to explain this in the S-1, right? They call it — and this has been, I do think, fairly consistent about this idea. It’s just a question of whether this idea makes any sense at all — what they call the “race to the top.” Okay, we have to build it first because then we will have built it the right way, with the right safeguards, and it won’t be Chinese tyranny running things, and it won’t be Sam Altman — whatever, even San Franciscan Sam Altman. What do you think? Does the “race to the top” idea make sense, Mat? How have you evaluated that?
Mat Honan: I think the “race to the top” idea is sort of a self-serving idea. I really do. And I think there are people at OpenAI who would tell you that they’re racing to the top as well. I think there are people at Google who would say they’re racing to the top — DeepMind, for sure. And I don’t know. To go back to that caller’s point about data centers, do you remember, probably around 2010 to 2012, there were all those protests against the Google buses?
Alexis Madrigal: Yeah.
Mat Honan: The Google bus is a physical thing that you can stand in front of, right? And I feel like the data center thing is much the same. You don’t want it in your neighborhood because it’s going to make your rates go up. It’s going to be loud. All that kind of stuff is true. But it’s also something that you could stop. And I think a lot of people right now feel sort of powerless and uneasy about the advent of the age of AI. You can stand in front of a data center and make it not be there.
Alexis Madrigal: Yeah, you can go to the permitting hearing and slow it down. You can do all these things. It has been truly, truly amazing to look at that. One of our listeners, RJ Wrights, asks — and Berber, maybe we’ll come to you on this — can you talk about Anthropic’s commitment to social causes, specifically around income disparity? Of course, in the AI and automation conversations of 10 years ago, people were talking about this stuff all the time: universal basic income and all these kinds of things. Where does AI — where does Anthropic — stand on these things?
Berber Jin: Well, it’s an interesting question, because a lot of the business proposition behind Claude is, “We will have Claude be able to save companies a lot of money by automating your workforce.” Anthropic Works has this big project with a bunch of private equity firms to implement Claude in legacy industries, like accounting, to cut head counts and improve efficiency.
So, to start off, I think it’s important to be clear that the financial interest of this company is to herald an economic revolution that will result — they will not contest this — in massive transformations to the labor force. I think Anthropic has also been intentional in trying to find ways to offset that. They produce a lot of papers about how economic transformation will come, and they also have a lot of money earmarked for philanthropic causes. I think — I forget the name — but there’s a vehicle that exists where some of the money they have is specifically directed toward reducing the impact they will have on society.
As a normal denizen of the world, it’s a little bit uncomfortable, right? Because you’re handing this company so much power to say, “We are automating away your jobs, but we will also find ways to try to smooth it over.” And I think that goes to that sense of powerlessness Mat was talking about, where they just have so much power and control. I feel like a lot of people feel like it’s kind of out of their hands, the concentration of influence this company has. Yeah, it’s really quite extraordinary.
Alexis Madrigal: You know, one of the reasons I think people have come to believe in Anthropic as a business, if not in all these other things, is basically Claude Code, right? So, Sri, I want to come to you on this. This is Claude Code, and then the sort of harness they’ve put around their product so that you can ask Claude to do all kinds of things and build all kinds of things. Anything a computer can do, it could do it for you, basically. And this has turned out to be kind of the magic of the business, right, Sri? Because they’ve rolled this out to companies, and it has not only transformed Anthropic itself — where I think they say 90% of the lines of code committed to their code base are made by Claude Code — but also all these developers and business software makers.
Sri Muppidi: Totally. Claude Code has been pivotal to Anthropic’s growth. They released it last year at some point. While OpenAI, the bigger rival at the time, was focused on consumer AI with its ChatGPT product, Anthropic really bet big on serving businesses. By focusing on Claude Code, they were able to automate software engineering work, which allows enterprises to do more with AI but also cut labor and save on costs.
I think Anthropic was smart about focusing on that. Over time, OpenAI has tried to follow by focusing more on Codex, for example, its competing coding product. But I don’t think Anthropic expected Claude Code to be so popular. At the start of this year, it was a little flat-footed in terms of how much computing power it needed to serve Claude Code. A lot of folks were complaining about their usage limits being affected if they were using the chatbot, and developers were scrambling to get more tokens to use Claude Code.
Eventually, Anthropic went on a deal spree to strike so many deals. That goes back to the $807 billion commitment they have for more computing power, chips, and data center work. Claude Code really was — and is — a huge driver for Anthropic’s business.
But now, as part of their S-1 and what they’ve been pitching to potential investors, they’re saying, “Hey, we’re going to save the world not just through software engineering, but we’re going to focus on life sciences and help you discover new drugs and cure cancer.” I think that’s also part of what Berber was saying earlier: the rejiggering of its mission to be maybe more of a safety-conscious company, because they’re countering the doomerism that their founders and employees feel. But also, “Hey, we actually have an IPO and need to raise $100 billion. How are we going to do that?”
What’s also interesting, when I was reading the S-1, is that there was a phrase where the risk of humanity ending was written in the passive voice, versus, “There have been claims…” I don’t remember the exact wording, but it was very much like people, or it’s like—
Alexis Madrigal: Yeah, it’d be like, “See if all of our executives…” Yes, exactly. Say that we’re going to have a lot of—
Sri Muppidi: It’s like, not Dario saying it, but it’s like, “Claims have been made.”
Alexis Madrigal: Yeah, yeah. I mean, the strange thing — and I think it’s a real question for the business in an interesting way — is that Claude Code reached ubiquity among software developers. It seemed like more or less overnight, from the end of 2025 into 2026.
But, you know, my wife and I have a small business. I used Claude Code to automate the back end of this thing, and it works really well. I did it one time, used some tokens, and sort of that was it. I think one of my questions is whether they’ve had to start talking about biology and some of these really data-intensive realms because you could automate most white-collar work with a free plan, practically. These tools have gotten so good that you don’t actually need to pay Anthropic all that much money to do most of the white-collar information work that people do in normal jobs that aren’t coding.
So I think there’s a real question. You’ve seen this incredible wave of adoption inside companies by the coders, but do the rest of us, Mat, even need as many tokens as it would take? They could totally wipe out white-collar work, and it still might not be a good business if you have to pay $800 billion for data centers and chips.
Mat Honan: I’m sorry, I’m not quite sure—
Alexis Madrigal: Oh, the question is just, do you think there are going to be enough people who need to use these tools the way coders have?
Mat Honan: Oh, I don’t know. I mean, I will say, just on a personal level, I’ve now started paying for Fable upgrades, which is their most capable model, which was surprising to me — that I actually hit the point where I wanted to do that. And, like you, I’ve done a ton of personal projects and things like that.
I think, again, a lot of what I’m talking about today is anecdotal, but our development team — our team of devs — has become so much more productive because of Claude Code. I do think businesses are going to be willing to pay for that. It’s clear that all kinds of businesses, such as your small business, are going to be able to use these tools in the same way that a larger customer is. And Anthropic clearly has some really large enterprise customers, right?
Alexis Madrigal: Right, and that’s where, like—
Mat Honan: Most of the revenues are right now.
Alexis Madrigal: We are talking about Anthropic as a business, and we’re also talking about what the Anthropic IPO could mean for the Bay Area and what it says about the state of the AI industry. We’ve got MIT Technology Review Editor-in-Chief Mat Honan. We’ve got Sri Muppidi, business and technology reporter at The New York Times and DealBook, and Berber Jin, AI enterprise reporter with The Wall Street Journal.
We’re going to get to more of your calls. I promise, in the next segment. What about you? Are you planning to invest in Anthropic? Why or why not? You can give us a call, 866-733-6786. Email [email protected]. How do you feel about this economy that’s solidifying here in the Bay Area? You can find us on Bluesky, Instagram, and Discord. We’re KQED Forum, all those places. I’m Alexis Madrigal. Stay tuned.
Welcome back to Forum. I’m Alexis Madrigal. Here we are talking about the Anthropic IPO. We’ve got MIT Technology Review’s Mat Honan, the New York Times and DealBook’s Sri Muppidi, and Berber Jin from The Wall Street Journal. Let’s bring in — oh, we just lost the caller I was going to go to. All right, here we go. Let’s bring in Michael on data centers. Welcome.
Michael: Hello, thanks for taking my call. What I wanted to bring up is this notion that data centers are evil, data centers are awful, and we should rise against the data centers. That’s really an inaccurate view. Data centers are no worse than any of the other things that we seem to be okay with: pollution from cars, pollution from nuclear power, pollution, pollution, pollution. What people don’t seem to realize is that many states bent the rules to get data centers into their states because they needed the revenue, right? So I think this is basically a straw man. It really is.
And if you think about what you’re getting from these data centers, you’re giving power to people who normally wouldn’t have it. Here’s what I mean: take medical technology and medical information. It used to be, “Okay, I’ve just got to listen to this doctor and hope that they’re right, because I have no other resource to validate what they’re saying.” You can actually use Claude to do that. “Hey, here are my test results.” You can feed them into Claude. Claude will give you plain-English explanations of what this means and the kinds of questions you need to be asking your doctor.
Alexis Madrigal: Well, I want to take the data center question, though, because I think it’s an interesting point. I think part of it is, in exchange for the data centers, you also get to lose your job. That seems to be sort of the rhetoric, you know? People previously didn’t have a lot of opposition to data centers. Santa Clara County, for example, has tons of data centers, and there was virtually no opposition to them. Mat, do you think that’s because — essentially, people are using the tools of, say, environmental justice to raise questions about pollution or other things, but what people are really objecting to is, as you put it, this is a physical manifestation of the AI-driven economy?
Mat Honan: I mean, that’s definitely my opinion. A lot of it is about the physical manifestation. I do think people have legitimate concerns about their rates. They have legitimate concerns about power plants being built. To power these data centers, a lot of the energy sources they’ll have to use may even include natural gas.
Alexis Madrigal: One is natural gas, yeah.
Mat Honan: And I think people have legitimate concerns about that. The data center pushback is older than the AI boom if you think about it from a purely power or energy perspective. Texas did a lot to bring in data centers during the crypto boom, right? Then they had a bunch of brownouts, and they started peeling back the rules they’d put in place to make it easier to bring in data centers. And they’re back because there are concerns about energy usage that are—
Alexis Madrigal: Yeah, I mean, I think if you look at Virginia, where a huge chunk of data centers are, and you look at their power prices, I also think people are responding to the other idea of power — the other definition of power. In a lot of these Virginia places, so much of their property tax revenue is now dependent on these data centers, which makes you wonder, what are they going to do politically in that circumstance? Let’s bring in Todd in San Francisco. Welcome, Todd.
Todd: Hi there. Since Anthropic and these IPOs are going to generate a lot of money, and there’s already been a lot of money generated in the Bay Area by investments, I’d like the cognoscenti of the AI world on your show right now to speculate about why these millionaires and billionaires are so much different from San Francisco’s previous set of millionaires and billionaires — like the gold miners who struck it rich. Famously, cultured gold miners would build opera houses and import famous authors to read out loud. But living in the Bay Area, one would have to say they’re rather hermetic. What’s up?
Alexis Madrigal: This is really — I mean, Berber, why don’t we go to you? Anthropic specifically was deeply aligned with a movement. I don’t know if I want to call it a philosophical movement, but I suppose it is: effective altruism. In fact, one of their earliest backers, Sam Bankman-Fried, was one of their largest investors at the point when he went down. So what do we think? This seems to deeply inform the culture of Anthropic. Can you give us a little summary of effective altruism and its application to Anthropic?
Berber Jin: Yeah, so effective altruism is deeply ingrained, particularly among the co-founders and the very early set of employees, who also happen to be the richest in the IPO. EAs care a lot about how to maximize the amount of good in the world, which leads to a lot of conclusions that most normal people would consider esoteric. For example, if we want to minimize suffering in the world, we should all be vegans. And you take that argument to the extreme — a lot of anti-EA people like to make this argument to make fun of EAs — and you get shrimp farming, because billions of shrimp are farmed every year and they can experience pain.
So apply that to AI. A lot of the early employees, a lot of people who became very interested in AI, became interested in AI because they felt like here’s a technology that could cause extinction, and extinction would be the worst kind of outcome. It would maximize the amount of suffering in the world, so we have to prevent it.
Anyway, to apply that to the question, which I think is really interesting, about what values will animate this new billionaire class: I think an interesting litmus test will be to see how much of their money goes toward funding cultural and artistic institutions in San Francisco and the Bay Area, and toward funding civic institutions that I think a lot of normal people could really get on board with.
The early employees at Anthropic — a lot of them have pledged sometimes half of their wealth to philanthropic causes. But because effective altruism is so deeply ingrained in their worldviews, the chatter right now among employees is, “We should donate this money to fund AI safety nonprofits that are focused on how to use AI to prevent human extinction. We should use it to back animal welfare nonprofits.” Berkeley’s going to get rich.
There is this interesting countercultural element where a lot of these people are in Berkeley. One dig against EAs is that they famously don’t care about art and culture. A lot of people who are anti-EA will make that argument. But I think it will be interesting to see how they spend this money. Will it just go to these kind of wacky causes? How much of it will go toward institutions like SFMOMA or supporting the opera? That will be a way to see the relationship between this new billionaire class and the rest of San Francisco. What that looks like will depend in large part on how these employees and founders choose to spend the philanthropic wealth they’ll gain from the IPO.
Alexis Madrigal: I mean, having followed some EA stuff — not closely, but sort of a little bit from the outside — it feels like a real kind of hothouse-flower thing. This started out as a very small group of people all talking to each other about this stuff, and now, of course, it’s become a global, massive discussion about whether the kinds of ideas they had about the world and AI make any sense at all. But, of course, they will have massive wealth, and they also have the proof of this company they’ve built on their side of the ledger, I suppose.
Mat, what about you? We’ve seen plenty of weird ideologies flow through San Francisco. I would not have bet, 12 years ago, that effective altruism would be associated with one of the fastest-growing capitalist companies ever. Think about the philosophical underpinnings of this place.
Mat Honan: I mean, you want to talk about the rationalists a little bit when you talk about effective altruism, a lot of which is headquartered out in the East Bay. I do think it’s had a profound effect on the way a lot of people, especially those working in tech, think about the world and about what’s the most effective way to make the world a better place.
I think Berber’s exactly right, though. It’s really unclear whether the people who have come up in that scene or movement — whatever you want to call it, philosophy — are going to donate to the opera versus places that are more globally oriented, or AI safety organizations. I don’t know how much we’re going to see of this in the Bay Area. And I think if they wanted to generate an enormous amount of goodwill, they could effectively solve the housing problem in San Francisco instead of making it worse. There are all kinds of ways to put that money to use locally.
Alexis Madrigal: Yeah, yeah. We’ve got a couple of comments here. Eric writes in to say, “As with the microchip boom of the ’90s and the dot-com boom, it’s real. We might see a bit — a lot — of volatility as profitability catches up with valuation, but the smart money has already spoken.”
Then Steve on Discord has this question. News came out recently about China’s AI companies really suffering, with half a trillion dollars of valuation gone. Steve notes that — what do you think is going on there? It’s seemed like two different worlds to me that kind of come together, I suppose, in the American market. But, Berber, do you think the fact that these Chinese AI companies are struggling in their stock market says anything about Anthropic or what’s happening here?
Berber Jin: That’s a really good question. The whole Chinese AI ecosystem really puzzles me. I mean, I think — maybe this isn’t a direct answer to your question — a lot of the Chinese companies, it’s unclear to me what their modus operandi is.
What we do kind of know is that they somehow don’t have enough — they can’t raise as much money as Anthropic and OpenAI. I’ve always been confused about why the government doesn’t just subsidize and pour tens of billions of dollars into these companies. They’re famous for distilling Anthropic’s models. They’ll use the outputs from Claude to train a cheaper, hacked-in version of Fable or Opus or the Anthropic-class models, which Anthropic hates. But then they also give away a lot of their technology for free. They make it open source, so none of the companies really have a business model, which is why I think there’s been all this volatility in the Chinese stock market with those names.
It’s all kind of perplexing to me, how the Chinese AI ecosystem works. I don’t think American policymakers or lab leaders have a great sense of it either. So that’s how I would answer the question. It’s all very confusing to me.
Alexis Madrigal: Yeah, no, I don’t disagree. And I think it has confused American AI leaders a lot. When you started to see these Chinese models come out with training numbers that were teensy-tiny on the cost side, it made it seem like, well, why are Anthropic and OpenAI spending $800 billion over the next five or 10 years when these Chinese model makers seem to be able to do more with less? Which I think remains an open question. It feels like there are some variables missing there. James in Larkspur, welcome.
James: Yeah, hi. Good morning. I wanted to shift the conversation just a little bit to the control of information that’s coming out of Anthropic, for instance. Traditionally, on the internet, we were shown sites that we could go to and ultimately make our own decision about whether the information was good or bad.
What’s happening now with AI is that they’re synthesizing all of that information, writing it, and feeding it to you. We have to remember that this is really based on algorithms that are originally being written and controlled by a certain entity. To release information, to control it and manipulate it, leads to what’s happening, for instance, right now with a pump-and-dump scheme on one hand, where they’re saying only we can control this information. It’s rogue computers that only we know how to tame the monster. So they’re generating a great deal of scare tactics and technological advancement to raise money through these IPOs.
Caller: Mm-hmm.
James: But at the same time, what we have to be careful of is that, ultimately, at the end of the day, they will be decreasing — as they are now — the sites that you can see on the internet, and increasing their interpretation of that information and controlling it. Lastly—
Alexis Madrigal: Wait, James, I want to take this, though, because we only have a couple of minutes, so I want to throw this to Sri. This is really interesting. It’s basically changing the shape of the internet, and I think that’s really true. Now, instead, Google would have presented some set of options for most people, and you could have moved freely through that, although most people didn’t. Now you’re filtering all that information through Claude or through — what do you think, not just as a business proposition but as a cultural proposition, Sri? Is that a good idea or not?
Sri Muppidi: Totally. You’re already seeing it, as we just discussed, where businesses and advertisers are trying to cater to being surfaced through these chatbots. I’ve heard anecdotes about people trying to add so much slop to Reddit because chatbots index Reddit answers really quickly to surface suggestions or whatever it is. So there’s now a whole set of startups trying to do AI-like SEO so that their companies will be recommended.
These labs, Anthropic and OpenAI, are striking partnerships with Walmart or whoever so that those businesses are recommended first. I totally agree that it will shape how information continues to be shared. Businesses are recognizing that and trying to invest money so they aren’t left behind.
I think it will be an issue from a societal standpoint as well, just because AI is now being shared with students on their Chrome laptops or whatever. It’s really educating our next generation of students and workers. So I think it will be quite interesting to see. I don’t have, like, a—
Mat Honan: Yeah. Something about the way we get information and things that was really interesting to me: there was a study from MIT last year, or in the past year, on financial literacy. A lot of people are now going to a chatbot and saying, “Where do I put my $1,000?” or “How should I invest my money?”
What this study found was that people who have a higher degree of financial sophistication, and can ask a more financially sophisticated question — a more financially sophisticated prompt — get much better advice. If you just go in and say, “Where do I invest?” you’re going to get some really basic advice, and you’ll get advice that will yield you lower returns.
They also found that it was very gender-specific. Women were getting basically worse financial advice — or worse long-term financial advice, I should say — being told to put more money into savings accounts than men were. I think that’s one of the real things we haven’t really started to think about or address yet: your ability to interact with the chatbot is going to become increasingly important. You’re going to have to develop a literacy about how you interact with these things.
Alexis Madrigal: I have a somewhat heterodox position on this, only in the sense that if I look up rose quartz or some mineral, you just can’t find the facts about a mineral on the internet anymore because so many people want to sell you rose quartz for ritualistic purposes. I just think the baseline level of what you were finding on the internet has gotten so bad that I feel like Claude’s synthesis of it is better than what most people would do for themselves. Sorry, unpopular opinion from the room.
Nancy writes in to say, “Looking forward to the burst of this AI bubble. I’m so sick of hearing about thousands of people losing their jobs because of software forced upon us that hallucinates. Going into the city makes me depressed because every billboard is about someone’s new AI tool. It’s so dystopian.”
There it is — counterarguments to the whole show. We’ve been talking about what an Anthropic IPO would mean for the Bay Area. We’ve had Mat Honan, editor-in-chief of MIT Technology Review. Thank you, Mat.
Mat Honan: Thank you.
Alexis Madrigal: Sri Muppidi, business and technology reporter at The New York Times and DealBook. Thank you, Sri.
Sri Muppidi: Thank you.
Alexis Madrigal: And Berber Jin, AI enterprise reporter with The Wall Street Journal. Thank you.
Berber Jin: Thank you.
Alexis Madrigal: I’m Alexis Madrigal. Stay tuned for another hour of Forum ahead with Mina Kim.