我们是否正身处AI泡沫？很快就会见分晓。

In a few weeks, we’re finally going to get some answers on whether we’re in an AI bubble. That’s because the frontier lab Anthropic is going public.

San Francisco-based AI giant Anthropic filed papers today.

It might raise $100 billion at a $2 trillion valuation and rescue a whole slew of companies floundering in the debt market, or it might raise a lot less and send the entire stock market into a tailspin. Which is to say, a lot is riding on this IPO: the near future of the American economy, the medium-term future of the tech industry, and, given that the risk factors in Anthropic’s S-1 include catastrophic and existential risk to humanity, maybe even the long-term future of the human race.

At a minimum, the Anthropic IPO will mark the end of an era. The venture investment-driven, hype-accumulating startup phase of the AI century is nearing completion. But what comes next? In today’s episode, we’re going to try to wrap our minds around the most important company in the economy, at least for now, and look at what’s at stake in its big moment. Are we in a boom or a bubble? Will tech’s big bet on data centers work out or fizzle? And can Silicon Valley’s most militant AI believers be a normal public company, or are we entering a new era of AI hype and mysticism?

I’m Max Read. This is Hard Fork. We are live in studio with—

We’re not live.

Okay, that’s a really good start.

Yeah.

We are here in studio with Erin Griffith. It’s so good—

Hey.

—to see you in person.

Yeah, likewise.

There are 2 big developments in the Hard Fork universe this week. First is that you finally followed me on Instagram.

Oh, yeah.

We can get to that later. Maybe in segment B.

Okay, yeah.

There are a lot of referendums that we’re going to talk about today, and I think that was an important one. The second thing I want to talk about is that Anthropic announced its intent to go public—

Mm.

—which means the rubber is finally hitting the road after however many years it’s been of hype and doomsaying and argument. You’ve been reporting on Anthropic and the AI world for a long time, and I wanted to pick your brain about this company today. This feels like a huge inflection point for AI, tech, the stock market, the economy, and humanity.

Yeah, the stakes are only that high.

Yeah, exactly. So what’s riding on this?

Everything. I think our headline was “Nearly Everything.” Practically speaking, this is the first pure-play LLM, or large language model, company to go public and open its books so we can look at the underlying business of this AI boom. It’s been going on for 4 years, attracted trillions of dollars in investment, and sort of reoriented our entire economy around itself.

Anthropic, alongside OpenAI and arguably Gemini from Google, are the companies driving this. They’re the pillars. I’ve heard many metaphors: you could call them the bottom piece of the Jenga tower. I’ve heard them described as the tip of a pyramid, with an entire ecosystem of companies supporting it. Either way, they’re the quintessential AI companies, and now, for the first time, we have a look at the underlying business model so much money has gone into.

Are the investors you’re talking to nervous about the IPO? Are they worried about what might come out of it?

Last week, they were definitely nervous.

Yeah.

Now we’ve had a day or 2 to look at these numbers and see what’s there. Some of them have leaked along the way, so we had a sense, but this is the first time we’ve seen the full picture of what Anthropic’s business looks like in a 300-plus-page document.

So let’s go there. What are the big, important things we can take away from the S-1?

First of all, I guess it’s okay to say the business looks kind of healthy. It’s arguable; people have made the case that it’s not. But generally speaking, people have been pleasantly surprised to see that this company is capable of making money.

When we’ve seen how much money these companies have raised over the years and how much they’ve been spending to build their AI products, I think the general consensus has been that these companies can never make money. To be clear, Anthropic posted an $80 billion net loss, so let me unpack that a little bit. An $8 billion net loss seems insane, but it’s actually a weird accounting quirk: they have to count some convertible debt they issued to some of their shareholders as a net loss. That will essentially be cleared in the next quarter, or by the time they go public, so it’s not a reflection of the health of their actual business.

What we’ve actually seen is that, in the first half of this year, they had an operating loss of around $600 million to $700 million. That’s shrinking from the same period last year, when it was over $2 billion. They’re shrinking the losses of the actual business they’re operating.

On the flip side, they’re still shelling out enormous sums of money, and they’re on the hook for enormous sums to build out their AI infrastructure: the data centers, the chips, the compute power, and the very expensive stuff they need to keep training new models and support the old models. Dario has talked about how the company is sometimes on a knife’s edge when it’s making these decisions and signing these deals.

They’ve signed so many contracts in anticipation of demand, and it’s a very high-risk game. If the demand isn’t there, they’ll be sitting on all these contracts, and many of them are non-cancellable. If the demand is there and they don’t have enough, they’ll lose business. Underlying all of this is how much AI businesses and people are going to use, and whether the company has laid the groundwork for that.

Right. It’s interesting because, on the one hand, the S-1 seems to confirm that there’s an enormous amount of demand. We’re talking about $16.3 billion in revenue for the first half of 2026.

And that’s an enormous amount of growth, too.

Right.

We’re seeing a company grow 14x. That is insane, especially for a company of this size. I hate to say “unprecedented,” but it feels like we haven’t seen a company grow like this before.

Right. So we can say there’s demand for this, right? But the question you’re raising is that we don’t actually know how much there is. It might be more than we’ve already seen; it might actually be less. We’re building a certain number of computers, to put it very simply—

Yes.

—no matter what—

We’re building computers.

We’re building a lot of computers out there in the desert, and Anthropic has committed a ton of money to them.

Mm-hmm.

So is that the biggest risk they’re facing?

Definitely. I think this is what everyone’s looking for. The S-1 we saw did not have third-quarter numbers in it. They’ve already previewed those numbers to their investors, and our reporting and others have shown that growth is actually slowing in the third quarter.

That’s because, in the first half of the year, Claude Code was growing like crazy. They basically took the entire industry by storm, and everybody was using their coding product. Now competitors like OpenAI are starting to catch up. Companies are trying to pull back on how much they’re spending on tokens because they realized token maxing got out of control. They’re also getting competition from open-source models or cheaper options.

I think in the third quarter they grew 44%, which is a pretty significant slowdown from gangbusters growth to just great growth. But is that going to continue on that trajectory? In that case, the growth story isn’t as exciting, and perhaps demand isn’t going to be through the roof forever. That’s one question investors are looking at.

In general, they’re looking to see how Anthropic explains that slowdown. If Anthropic comes in and says, “The whole market is pulling back,” that’s very bad for this whole ecosystem—the Jenga tower or the pyramid I’m talking about—with all the demand, all the investments everybody is making, and all the debt everyone is taking out to build this stuff.

If the reason growth is slowing is just competition—if OpenAI or open-source companies are eating market share, but demand is still growing—then great, party on. The AI boom continues to boom.

So they’re looking for new avenues for growth, potentially. They’re trying to sell different stories about what the next big product is.

Yeah, and I think we’re going to see this on the road show. I’ll be watching that pretty closely. I think that’s going to kick off in probably a week or so.

Before you start, what is a road show?

In a week or so, Anthropic is going to go on this tour where they meet with all the public-market investors and tell their story. They call that a road show. The bankers take them around on a little private jet, and they’re meeting with insurance companies in Iowa or whatever, telling them about the machine god. That’s going to be amusing on its own.

But they need to tell a new growth story, I think, because they’re going to say, “Okay, we had this coding product that’s incredible, but it’s already kind of showing slowing growth. What’s our next growth story?” They believe AI is going to solve many, many, many problems in the world. One of the stories they’re pitching to investors now is this bio thing: AI’s going to cure cancer, AI’s going to develop new drugs. That’s a big story they’ll probably be pitching. And then they have other products, too, like legal tech or other ways AI is going to disrupt the economy beyond just coding, which has been a blockbuster hit for them.

Right. One thing I want to talk about with the S-1 is the 93-page risk factor section.

Yes, yeah.

You’ve read a lot of S-1s before. Are they—

Yes.

Are they—

Every word of every one, just to be clear.

Right, yeah, obviously—

I read all 90 pages.

—in your capacity as a reporter. Are they usually 90 pages long?

They’re usually pretty long, and it’s a lot of legalese. It’s a lot of ass-covering, basically. It’s anticipating all the potential things that shareholders could get mad and sue over and saying, “We warned you.”

Right, exactly. And there are a lot of those for Anthropic. I’m just going to read a couple of them right now.

They’re very unique in their specificity, let’s just say.

It’s funny reading them if you’ve been following AI for a while. It’s like getting six years of Twitter and blog posts and Substack compressed into a document of legalese, because they’re listing everything from the resource constraints on building out compute to the various AI safety problems to your basic business problems with customer satisfaction.

One that caught my eye, for example, is: “Concerns have been raised both within our organization and by other researchers that sufficiently advanced AI systems could, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures, pose catastrophic or existential risks to humanity.” This is a pretty rare kind of thing to do. Another one that was really impressive—

But wait, can we just go back to that really quick? One thing that the great Matt Levine of Bloomberg pointed out was that this is written in the passive voice. It’s not Anthropic saying, “We think that our products...” It says, like, “Some people may be saying.”

Right. “Within our organization,” some people. I don’t know if it’s our founder and president, Dario Amodei, or anybody, but some people have raised this question.

Yes. Who could it be?

Yeah. Another one I like is: “In addition, considerations relating to model safety, interpretability, alignment, and welfare may lead us to limit or delay the deprecation of older models.” Just to translate that a little bit, when they’re talking about model welfare, they think the models might be alive, and they’re not ever going to shut down some of the product lines because they want to make sure the models get taken care of.

What? I missed this one. Wait.

I have to credit Joe Weisenthal at Bloomberg for picking this one up. They’re not saying this out loud, but they’re saying, “Considerations relating to welfare may lead us to limit or delay the deprecation of older models.” This is a little like Apple saying, “The iPhone 12 is alive, so we’re just going to keep making it forever,” or something like that.

Oh my God, imagine trying to explain that to some analyst at Goldman Sachs and being like, “We need to put this in here.”

The thing I wonder about a lot is how much people who spend their time talking to chatbots all day start to buy into and believe this stuff. You’ll actually have activist investors saying, “We have to make sure that we’re taking care of model welfare,” or whatever else. This is a future coming soon.

Well, the activist investors won’t have a whole lot of ability to do much because of the governance of the company.

Yes, let’s—I actually want to get into this. But before we get there, let’s start with a quick word about the founders, because this is a company with a very specific philosophy around AI risk and AI safety. It’s not just at the core of the business; it’s kind of their origin story. I’m sure a lot of people are familiar, but will you give us the 20-second version of where Anthropic comes from?

Sure. This is kind of the story of AI, too. Google’s working on AI, Elon Musk and Sam Altman see it, and they’re like, “Man, I don’t think they care enough about the future of humanity.” So they split off and start OpenAI, and they’re like, “We’re going to build AI, but we’re going to do it for the good of humanity. We’re the principled ones.”

Then within OpenAI, Dario and all of his friends are looking around and saying, “Wait, Sam doesn’t care enough about the future of humanity. We need to build safe AI before they can.” So they split off and do that. And that has continued down the line. That’s where Anthropic is. Its roots are truly in AI safety, and it’s probably the most extreme on that spectrum, in terms of the fears inside the company, at the top of the company, and around it.

Well, let’s talk about how this relates to governance, because they have a very unique governance structure.

I’m losing my voice now, and hopefully they can use some AI to make my voice sound less gravelly.

Mm-hmm.

Anthropic is following a trend that’s been happening for the last 15 years at tech companies: founders want to retain control even while selling shares. The classic form of governance means that if you sell shares, those shareholders get more control. This goes back to Mark Zuckerberg and his multiple classes of shares, but Anthropic has taken it even a little bit further.

They have 5 classes of stock. Their 7 founders have 50.1% of the voting shares. They’re going to be in charge of electing all the directors to the board, and then they also have this independent trust that controls a majority of the board seats. It’s a little convoluted, and I think no one’s going to push back on this because, over the years, every time startups have tried to do this, investors have made a big stink about it—“Oh no, we hate this”—and then they just go along with it anyway. There hasn’t been a line in the sand. Even with SpaceX, they were a little bit more extreme about that. Some of the exchanges or indexes tried to put their foot down, and then at the end of the day, they were just like, “Ah, fine.”

So this isn’t necessarily new in Silicon Valley, but it seems distinctly connected to Anthropic’s AI safety philosophy at its heart. We should say, I’m not sure if we said it out loud, but Anthropic is the company that believes most strongly in the possibility of total AI annihilation, Armageddon, and additionally believes most strongly in the possibility that LLMs are alive, conscious, or sentient in some way.

And they lay out that they believe their founders are best suited to shepherd this into the world. It’s a little self-aggrandizing when I was reading it. I was like, “Wow, you guys are special.” And I guess you are, because you’re at the helm of a potentially $2 trillion company.

They’re even able to name their successors. They have a secret list of successors that they all have to approve if somebody leaves the board, so they’re setting this up to be in control of it forever.

They don’t want to lose control of the AI or of the company. Yeah. I want to return to the philosophy, to the question of philosophy: beyond this governance structure, to what extent are there philosophical commitments around AI safety? How do they inform the company itself or its business decisions? Or do they?

Well, I think in the beginning it was a huge piece of the company because it was the only place where they could get money.

Mm-hmm.

Regular Silicon Valley investors viewed the company as a science project for the first couple of years, and the only place they could find money was from people like Sam Bankman-Fried and Jaan Tallinn, people deeply enmeshed in the EA philosophy and that world. Their first product was very specifically and purposefully an enterprise business product. They weren’t building something for consumers. They weren’t necessarily trying to build an AI companion, and that has proven to be an incredible business.

Then when the VCs got word of that, they were like, “Oh, great. Let’s dump all the money in.” Most of those VCs who came in later are looking at this as an enterprise software, SaaS business, and they’re not really that hung up on the philosophy. They’re like, “Yeah, we like that we’re backing the good guys.”

Right.

But they don’t really buy into the stuff that’s a little bit weirder about it.

The woo-woo stuff that they say.

They don’t. I mean, is there a word for AI woo-woo?

Ooh, no, I don’t think there is, really.

We should come up with one.

Yeah.

We should come up—

We’ll think about it.

Like the Hard Fork signature.

Yeah, exactly. It’s very interesting because you could look at all the decisions they’ve made, from the governance structure on down to the specific things they’re doing, and tell a story where it’s all about their philosophy. It’s all about their fear of AI or their belief in AI.

But you can also tell a story that this is just a very straightforward, smart Silicon Valley company. Its founders want to retain control. It wants to be very efficient. It wants a clear path to profitability through enterprise sales, not trying to become the next Facebook or whatever. And it seems like investors, they don’t—

I mean, that’s certainly what the investors I talk to say. I remember when Jacob Coxon quit and everyone was freaking out about AI killing us all. I mentioned this to some of their investors, and they were sort of like, “Oh, yeah, hacking, that’s a big risk. We’re definitely worried about cybersecurity, but I don’t really know what you’re talking about on the other stuff. We’re not worried about that.”

Let’s zoom out a little bit. This feels, as we were saying before, like an inflection point. We’ve spent the last half decade in this big-time hype moment. VCs, executives, and researchers are selling this crazy story of a—

Mm.

—technology that’s going to change everything. Now we have a big document with a lot of real numbers in it, and pretty soon we’re going to know what the stock price of Anthropic is. How would you characterize this? What is happening right now with the IPO? What questions are being asked, and what questions are being answered?

I mean, it is sort of a demystifying moment because we’ve seen these companies in the private market get trillion-dollar valuations in such a short time, and it’s kind of being set by a handful of VCs, right? Yes, VCs believe in this company, but they also believed in the metaverse and a lot of crypto stuff, and that has not worked out.

This will be the first time we have Wall Street, public-market investors, pension fund investors—all the professional money managers—modeling out the financials of this company and deciding whether it’s worth investing at a $2 trillion valuation, or maybe $1.5 trillion. So that’s part of it: we’re actually going to see some real evaluation of the company. Although I guess you could also argue that Wall Street sometimes invests on hype and momentum, too, as we’ve seen in many stocks. So who knows? But in general, this is a time when we’re finally getting to look inside the business underlying the AI boom and decide whether or not it’s a valuable contributor to our economy, because there are a lot of people out there who think it’s not.

Yeah. So what are you looking for in the weeks after the IPO?

Whether or not AI kills us all. Big question still on everyone’s mind. I doubt that’s going to be talked about too much on the roadshow.

Yeah, yeah.

I’m sure there are some advisors at Goldman and Morgan Stanley who have been telling Dario, “Let’s reel it back on the old extinction talk.”

I mean, should I invest in a company that has a 10% chance of killing everybody?

I mean, it’s in the risk factors, so you can’t say they didn’t warn you.

There was a Wall Street Journal headline arguing—the headline is, “The Bull Case for AI Armageddon—For Investors Mulling—”

The bull case.

Exactly.

The bull case.

“For investors mulling the Anthropic share offering, a 10% chance of extinction may be a 90% win for your portfolio.” How can you argue with that, really? If you’re going to die, you’re going to die, but if you can become a billionaire—

Die rich.

Yeah, exactly. Erin, thank you for coming into the studio and talking to me about Anthropic.

Mike and Jasmine, we were just talking about Anthropic’s S-1. Mike, what was your big takeaway from the filing?

I thought it was way front-loaded with rosier sort of paintings of AI at the top. You guys tell me, but I feel like Dario Amodei is painted as the biggest doom-and-gloom guy of AI in the Valley, and my perception was that everyone around him was like, “Hey, you need to pump the brakes on this a little bit,” especially as we’re trying to get the average person excited about it.

The whole top was like, “Hey, by the way, we do want to cure cancer. We do want to create these amazing drugs,” and then they back into, “Oh, and it has a 10% chance of blowing up the entire world,” basically. But I don’t know.

Yeah.

Someone told him to jazz hands a little bit at the top, right?

Yeah. How much do you think it was worked over? I was interested that the adjective they use at the very top is, he says, “We believe that AI is one of the most consequential technologies in human history,” which is a very funny, neutral adjective, right? “Consequential” goes a lot of ways, and it’s not necessarily how I would be selling my product.

We should do a real close reading of all the language in this, because I do feel like Dario’s definitely a micromanaging control enthusiast, let’s say. Every word probably has a few different nested meanings in there, for sure.

Jasmine, what was your big takeaway from the S-1, if you had one?

Yeah, similar to what Mike said.

I was just thinking about how Anthropic has a very odd relationship to money and being a for-profit company.

Mm.

The founders spent a lot of time debating whether they should be a nonprofit or a for-profit when they broke off from OpenAI. Of course, they’ve invented some crazy, multilayered governance structure, as they all do, which is something they reinforce a lot in the letter: the founders have pledged 80% of their equity to charity. The employees have all this interesting matching.

There’s a famous culture-fit interview that all of the Anthropic employees go through, where they ask, “Hey, would you work here if your equity went to zero?” And they make sure—

Mm.

—that everybody says yes before they accept the job. Mostly, it just made me think about how weird they are about the status of being a sort of profit-seeking corporation.

I know. I can only imagine what the meetings with the bankers were like. I assume the meetings with the investors were very similar.

How do you tell a labor union that we’re going to be post-money in, like, three? I think that’s a hard sell.

This is coming in the midst of a little bit of anxiety and, let’s call it, a jittery moment in the market. The elections are next week, and I really want to talk to you guys about what we’re looking at and what we’re waiting to see. Jasmine, what are you tracking for the midterms next week?

These are interesting midterms because I think it’s the first election where people’s feelings about AI and data centers could plausibly have shown up. Even though we’ve seen a lot of discourse and anger about data centers in particular building up over the last year, I did some reporting on it over the summer in Wisconsin and Michigan, 2 states where data centers were actually a really big factor in the primaries. In the majority of states, I don’t think they’re shifting people’s votes yet.

They’re an 80/20 issue in the sense that people feel very negatively about data centers and AI. This is bipartisan, and in that sense it seems like a unicorn, but it’s also not quite salient yet. I’ve looked at a million polls, and AI usually ranks 22 out of 39 issues.

Oh my God.

Even in states like Ohio or Pennsylvania, where data centers are a big issue, usually between 1% and 5% of voters will list it—

Mm-hmm.

—as their number one thing. We’re seeing data centers really impact your mayoral election, with recall campaigns and city council—

Right.

—reshuffles, but—

Yeah.

—I don’t think a big Senate race is going to be determined by data center sentiment.

Right. It’s a municipal issue. I mean, this came across in the reporting you were doing: this is going to get—

Yeah.

—a lot of town managers voted out, and—

Yes.

—council boards, selectmen—

Mm.

—or whatever else. But it’s interesting, because at least in the summer, it felt like this was a focus of some Senate campaigns. Some—

Mm-hmm.

—Senate campaigns were, in fact, putting a lot of resources into being anti-data center. Do you think that was wasted money, or is it just not ultimately the issue that matters?

I think it really depends. There are races that are close enough where, with rural voters in particular, you can get a lot of normal rural Republicans or low-turnout voters to turn out if there’s a data center in their backyard. I think the races people are really concerned about include Sherrod Brown in Ohio, who might stage a comeback in particular because of data centers, or Greg Abbott, who had a big motivation to restrict approvals and connections to the electricity grid, even though he’s been very, very pro-build.

He’s been welcoming data centers for years. But because of a potential threat from both his Democratic opponent and all these rural Texan farmers who are like, “What’s up with these power lines and data centers?” he actually did a pseudo-moratorium because he was worried about the midterms. I think there are a few elections in purplish states where data centers could swing things.

The other thing I’m tracking is that data centers are very concrete, whereas—

Mm-hmm.

—AI generally still feels so abstract to people that I’m not really sure. If data centers are barely swinging some elections, I’m not sure AI in general is going to be a voting issue yet, even though some people suspect it.

Yeah. It’s very funny. As you say, if you’re on Twitter or reading too much about this, it does feel like a life-or-death issue. Certainly it is for some people—some people writing introductory letters to S-1s. But it hasn’t really trickled down, except in an indirect way through the effect of AI on the economy.

Yeah.

Mike, what about you? Are you tracking anything in the election next week?

You talked about this a little bit in last week’s episode, with the head of policy at Meta leaving and imagining which new regime of Democratic operatives they might replace him with. I just want to see if there’s—

Thank you for listening, by the way. I really appreciate that, Mike.

No, of course. I’m a loyal Hard Fork listener. Of course I’m tuning in to my buddy, Kevin.

That’s right. That’s what we like to hear.

But—

We’ve got one. You’ve got one.

Two. I’ll get my wife to listen. I really want to see the theater of all the tech companies that have been trying to cater to Republicans suddenly being like, “Actually, we love the new Democratic, potentially Democratic regime in all of these states.” Because they potentially have a hold on Congress: “This is what we really believe in,” you know?

Yeah.

I think these swings are valuable because they show that these are corporations that don’t necessarily believe in anything except self-preservation a lot of the time.

Yeah. Are you seeing any indications besides, say—We’re talking about Joel Kaplan, who had been the head of policy at Facebook for 15 years and has been a sort of conservative whisperer. He announced his departure last week. Are you seeing any other movements that suggest a little bit of team-switching, or trying to make inroads among Democrats?

I think all of the companies are looking hard at their D.C. teams and saying, “Who do we need to elevate and put out front?” They’re never going to cut all of their conservative relations people or whatever. But it’s sort of like, “Actually, look at this guy, the guy who wears a blue pin instead of a red one,” or whatever. That’s all the companies. That’s just the game. I guess I shouldn’t be so cynical, but it is worth looking at and seeing if it works.

Why wouldn’t you be cynical? This seems like exactly—

I know.

—the right thing to be cynical about.

I know, right?

Yeah.

Do you feel like there’s a shift? I think we’re all aware of what felt like, I’m sorry to use such immaterial language here, a vibe shift in Silicon Valley a few years ago.

No, vibes are important, buddy.

Is there a vibe shift swinging back? I’m interested in whether any of you feel like that might happen, or whether the first vibe shift was an illusion.

I’ve seen headlines to that effect, but I think, like Mike said, it’s going to be a little more on the D.C. side of things. From my perspective, a lot of the VCs were really excited to be able not to care about things Democrats care about, like DEI. Culturally, it seems like we’re still going to be in that vibe.

Yeah, it doesn’t seem like that’s changed. Jasmine, we were talking before we were taping about a sense that there’s a question about how Republicans are going to react to this. They’re likely to lose the House of Representatives, and there’s going to have to be a shift in how they approach AI regulation in Silicon Valley.

Totally, yeah. This is one of the big open questions I have about the 2028 primary, too. Primary season is going to start basically 2 weeks from now. Are JD Vance and Marco Rubio going to break from Donald Trump’s extremely, extremely unpopular positions on AI regulation and data centers?

The majority of Republican voters do want to regulate AI. They would prefer slower development with more safeguards versus some kind of abstract, immaterial “beat China” narrative.

Mm.

And yet basically every Republican in Congress or every Republican leader has been unable to say any of that because the president sets the party line. If the Republicans take a beating in the midterms, I could imagine more folks deciding to break off and say, “Actually, our constituents are angry at data centers. They’re worried about kids’ safety,” if they’re social conservatives.

I’m starting to see a bit of a split in the Republican Party between the more traditional libertarian, free-market, “let ’er rip” types, who are still very excited about deregulation, versus some of the more New Right folks.

Mm.

I spent the weekend with some folks from American Compass, the New Right think tank, and they’re all thinking, actually, maybe after the midterms, this is a time for the Republican Party to get serious about AI regulation on kids, jobs, safety, things like that.

Were they saying AI or were they saying SI?

It was actually really funny. It was a Chatham House workshop, but there was a split in the room, and then you could tell there was a bit of—

Ah.

—internal tension between folks from the administration and other places who were saying SI, and the folks—

Mm-hmm.

—who were a little too staunchly like—

So good.

“We’re not caving to that.” So it was really funny. It was kind of great.

I’ve never seen a better metaphor for people talking over each other and not being able to communicate than not even being able to agree on what we’re calling this thing.

It was kind of amazing.

1. Silicon Valley Fights The Wealth Tax

Yeah. So one of the issues on the ballot next Tuesday is this tax in California on billionaires. Erin, can you explain what the proposition is, what the measure is, and why Silicon Valley is so interested in this tax on billionaires?

Yeah. It’s actually been pretty fascinating. Prop 40 is a proposed one-time 5% billionaire tax on wealth, including unrealized gains. That means private startup stock for billionaires in California. And as you can imagine, the billionaires are a little bit mad about it. What better way to fight back than to use their vast wealth?

They’ve gathered this group that we reported on. Building a Better California has gathered $333 million, which is so, so, so much to fight a state proposition, mostly from Sergey Brin. Then he brought in John Doerr and Patrick Collison, the founder of Stripe. John Doerr’s a very famous VC investor from Kleiner Perkins. Over the last couple of months, they’ve recruited this who’s who of Silicon Valley: very high-powered, famous VCs and founders, many of whom aren’t even billionaires. There are some mere hundred millionaires and even—

Temporarily embarrassed billionaires.

Yeah.

Yeah, right. And aspiring billionaires. They’re mad about this for a number of reasons. They think it’s poorly written, they don’t believe it’s going to be a one-time tax, and they don’t like how the money is going to be used. They don’t think the money is earmarked well enough to do the things the measure says it will do, and they basically don’t trust the politicians to do the right thing with it.

They’re really mad about it, and they don’t like the precedent it sets because California is known as a trendsetter. They’re worried it’s going to turn into a cat-and-mouse game where other states do it, especially because a lot of them have already left the state. But they’re mad about the principle of it.

The principle of the matter.

Yes. They believe this is going to hinder aspiring billionaires, founders, and builders from moving to California and wanting to get rich. There are a lot of reasons they’re fighting it, and they’re banding together, getting on weekly calls, and talking about their strategy.

One of the really interesting things we found in our reporting was that they’re aware of how it looks. There was a quote in one of the documents they’re circulating to other donors that says, “Billionaires are seen as having a net negative impact across issues.”

Brutal.

So they’re not trying to be the face of this. They don’t want to be out front. They’re not talking publicly about this like they might if there were some other issue, including data centers. They’re trying to go through other groups they’re partnering with. Planned Parenthood is a really big one they’re working with, because it’s opposed to the measure.

Why, sorry, why is Planned Parenthood opposed to it?

Yeah.

For some of the same reasons: they don’t think it’s sustainable. This one-time tax is supposed to make up for some of the Trump administration cuts.

Ah, I see.

Oh.

They don’t believe it’s setting them up to be sustainable going forward.

Right.

So they just think it’s poorly written. I think a lot of the people who are opposed to it aren’t opposed to the idea of a wealth tax, including some people I spoke with for this story. They just think this is a crappy one, and they wish it were better structured and made more sense.

But do you feel like they’re always going to say that? I feel like it’s always, “Well, this bill is poorly written, and if they could only legislate it better...” I just feel like I’ve—

Yes.

—heard that from Andreessen or whoever a million times. You know what I’m saying?

Totally. It’s almost sort of like respectability politics, where it’s like, “I agree with your message, but I don’t like how you’re doing it.”

Yeah. I was interested to learn that about Planned Parenthood, though, because I hadn’t realized this was more than just a cabal of billionaires plotting against it.

No, there are a bunch of groups. Ever since I wrote this story, weirdly, my phone has been blowing up with political text messages from different groups. A lot of healthcare groups have lined up against it.

Mm.

That’s interesting.

Yeah. From a Democratic point of view—this is a nonpartisan podcast, of course—from a Democratic point of view, it’s sort of nice to put one of these ballot measures on every year so the billionaires all get distracted and focus on that instead of putting money around the country the way Elon Musk is.

That’s pretty good game theory.

Yeah, there you go. Feel free to call me, Senate political action committees. There’s one other big thing I want to talk about today, and that is what’s being called Black October, the vuln apocalypse, the PHP vulnerability discovered last week.

Mike, I’m going to put you on the spot. Can you explain what happened last week and why it made life hell for everybody who works in information security? I mean, it continues to be hell.

Well, look, I think the way I like to think about it is that the entire internet and the infrastructure that computers run on is pretty Byzantine and built on a lot of old code that has a lot of holes and stuff in it, right? That’s just the nature of computing. You figure out what are called vulns, vulnerabilities, and you fix them over time. Microsoft has dozens of fixes to its main software a year. Now it’s ranging up into the hundreds because AI is finding more of them.

What happened, though, with Black October is not trying to get a Furby for your kids—

Mike, come on. Was that on the spot?

I’m gonna—

You gotta do better than that.

No, no.

Yeah.

Like, we’re—

I hope you workshopped that.

We’re on squishies now, Mike.

It’s dumpling squishies. I—

I was working—

My son did his—

Okay.

—his birthday wish list this week, and dumpling squishies, that’s the big—

Oh my God.

—that’s the big new thing.

See, I’m showing my lack of children on this podcast.

You’re showing your age is what you’re showing, brother. Sorry, Andrew.

But it is essentially a vulnerability that goes far wider than just any one little system tweak that you have to update your individual software on. This is stuff that folks in the industry said could break the internet.

It could affect how we get on websites and use them, and have our accounts taken over. It was much wider. My big takeaway is that this is the type of thing that’s going to be much easier and more commonplace to find in the age of AI, where you can have these programs scrape the internet and all that.

Right.

So we should say specifically what we’re talking about.

Yeah.

PHP is a programming language that—God, every time I start talking like this, I think about the dozens of men with huge beards who are going to be chortling and emailing me the meanest things they can say.

Do it.

I’m so sorry, but half of them are my friends. I’m thinking about specific people who are going to be texting me about how wrong I am. PHP is a programming language that’s used on a ton of different websites.

Yeah.

And as Mike is saying, it’s much easier to find vulnerabilities in these infrastructural languages with AI, which can run a bunch of different ideas and figure out what the exact problems are. When that happens, all of a sudden you have these really intense, really serious vulnerabilities that now need to be fixed—not from the top down, but across a bazillion websites.

Yeah.

I don’t want to make this podcast about me, but one day last week, I wasn’t able to record my score for a daily logic puzzle I do called Clues by Sam because Sam was fixing the PHP.

The real victim.

Well—

The real victim.

I’m just trying to bring this down to earth for everybody listening so they really understand what’s going on.

Imagine if you couldn’t do your Wordle every day. That’s the analogy you need to be using.

Yeah, exactly. Are you guys talking to people who are in the midst of this? I’m interested in whether, from the perspective of cybersecurity people or AI people in general, we think this is just the new normal: every month is 85-hour weeks for people who work in security, and all the AI companies are throwing up their hands and saying, “We told you so.”

I’d love to hear what you guys think, but I think the flip side, or the counterargument—and this kind of happened in the open-source versus closed-source debate a few weeks or months ago—is, “Well, guess what? AI is also going to enable us to solve all these problems, quickly defend against them, and fix them faster. These vulnerabilities already existed. It was only a matter of time before someone found them.” There are arguments from folks saying these fixes can happen. We just need to enlist the advanced tools we have now to fix them faster, rather than have random websites across the internet unable to do that in enough time. But I don’t know; that feels a little rosy to me. I do feel like we’re going to see this more often, and the old internet breaking much faster than before. That’s my point of view, at least.

Hmm.

One of the crazier news stories from this week was that the startup Cathedral, made up of former DOGE staffers—our nation’s best and brightest—was literally hacking into Anthropic’s Slack and IT systems.

Yes.

Anthropic is supposed to be the one building the best cyber defense internal models, which they can’t even release to the public. And yet their own technical systems weren’t secure enough to block this crew of young startup guys from accessing trade secrets in their Slack. Before, I would’ve thought, “Man, some municipal governments are really going to struggle if they get hacked by people with an open-source model.” But now it’s actually the best and most AI-forward companies in the world that are also vulnerable.

Yeah.

That’s a great point.

It’s going to be a very long few years, I think. I read a great piece that I want to recommend by Joshua Saxe, a former Meta AI guy. His argument is basically that, as bad as it seems, we’re still in a sort of defender-dominant window.

Mm.

Even though all these vulnerabilities are being discovered, his point—which I admire because I think it’s true—is that it takes cybercriminals a very long time to figure out how to do their jobs. Generally speaking, they’re not that smart, and they’re not that competent at running businesses. So we have a certain period where we can lock down defenses and, certainly at the level of cybercrime, protect against some of the worst vulnerabilities. But he says, on the other hand, this is not a wholly positive story. We’re still thinking about nation-state actors that have the resources and the will and desire to, as Jasmine says, hack into municipal government infrastructure and whatever else. I suspect this is not the last time we’re going to be talking about AI-discovered vulnerabilities on this podcast.

Great.

That’s right. Exactly. Go back to Hawaii, Mike. There’s nothing left there. The last thing I want to talk about is that it felt like it flew under the radar amidst a bunch of other news this week, but Microsoft offered buyout packages to 60,000 people. Erin, what did you make of that?

That is an enormous number. When you put it in perspective of the fact that Microsoft has about 230,000 employees, it just reminds you how enormous these companies are. But just to be clear, buyout packages don’t mean 60,000 people are going to take them.

Mm-hmm.

Ah.

They expect 3 to 5 percent of their employees to actually take them, so in reality it’s going to be at most probably 3,000 layoffs. But that’s still so many. To your point, Mike, over the last 3 or 4 years, we’ve just become so numb to these rolling layoffs from the giant tech companies because that’s just the new reality that started after the ZIRP era kind of deflated. All these tech companies that got super bloated because they thought they needed a million engineers decided, “Okay, now we’re just going to cut 1,000 people every six months or so.” Tech used to be this cushy job, especially at a place like Microsoft, which is very geographically focused, where you could have a 30-year career. There are a lot of people at companies like Microsoft, Amazon, and even Google and Facebook, which are a little bit younger, who have been there for so long. That is no longer the case. That may be a sign of what’s to come for the rest of the economy, because tech is always a little bit out ahead on workplace trends as well. That’s a little bit depressing.

The other thing I’ll add is something kind of darkly funny about the announcement: Microsoft said it was part of its efforts to give employees more control over their future.

Oh.

Oh.

What the—

Oh, boy.

What?

That’s not the kind of control I’m hoping for.

Yeah.

That’s not the control I want.

Seriously.

Really, really, really grim.

Oh, God.

Oh, God.

The silver lining, as far as your harbinger of what’s to come, is that a lot of this is because of their AI spend and infrastructure spend.

Yes. Yeah.

Perhaps Accenture, or some giant employer that’s not spending billions of dollars per quarter on AI, won’t do the same level of layoffs. They’ll just use the tools to justify different types of layoffs in lesser numbers.

Oh my God, Mike, Accenture is the worst example, because if there’s any company in the world that’s going to be fully taken out by AI, it’s consultants, right?

Yeah, you’re right.

You’re right.

I’m sorry. Good Lord. I’m going to get taken out by AI because I can’t even think of a good company.

No, actually, your mistake is what makes you human, Mike. That’s what people love.

Thank you.

That’s what people love.

Thank you. I live to serve.

When we come back, we’re going to be talking about stories that we’re tracking this week.

Welcome back. I’m with Erin, Jasmine, and Mike, and we’re going to be talking about stories that we’re tracking this week. Mike, what story are you tracking this week?

This is kind of nerdy, but that’s my thing. I found this memo from the VP and distinguished engineer at Amazon this week that most normal people are not tracking. I think it’s really interesting because it’s essentially a very long manifesto on the use of AI among software engineers inside the big companies.

Mm-hmm.

What they essentially cop to is that token maxing is a terrible metric for measuring how you’re doing as an engineer. LOC, or lines of code—the amount of code you’re contributing to build these products—is also a messy metric. It basically means you’re larding up these systems with a bunch of stuff that someone’s eventually going to have to delete.

In this long memo, they essentially tell employees, “Look, what we care about are long-term outcomes. This is what really matters in the end, and this is what we need to be judged by.”

Mm.

The problem is you can’t really judge long-term outcomes on a quarterly basis when you’re trying to figure out who should still be employed here.

Yeah.

After thousands of words, we end up back where we started.

Yeah.

I’m very curious how they judge these engineers in the coming quarters.

That’s so interesting, because I feel like when the tokenmaxxing trend started, everyone with any common sense was just like—

Yeah.

—this seems absurd. Spending as much money as possible to write as much code as possible, playing the volume game—it was obviously going to end badly. We all knew it was a short-term bump, and then the tokenmaxxing, or whatever they called it—

Yeah.

—tokenmaxxing. The reverse of that trend happened so fast. By the time we could barely get our first story up—

Yeah.

—it was already reversed. But it seems like they’re still struggling to figure out where the balance is. Is that your upshot from this?

What better metaphor for the tech industry than spending billions of dollars to find out something is a terrible idea and ending up back where you started?

They’re in the arena trying things, Mike.

Yeah, exactly.

That’s true. I am a coward.

I have to say, you say this is nerdy, but this feels like the most relatable possible thing to me. Everybody has a job where they introduce a new metric, and you know the metric’s stupid, and you fulfill the metric, and then your bosses are like, “Actually, we’re going to do the opposite metric.” So you do that, and you’re like, “This is still stupid.” And then they’re like—then they’re sort of like, “Actually, our metric is the long-term health of the company.” And you’re like, “Well, I’m still just going to do my job, and let me know—”

Casey, you’re self-employed. What are you talking about? Are you—

You are your own boss.

I am also self-hating, so I think I can—

Jasmine, what are you tracking this week?

I saw this new research report that a team at Anthropic put out. They did a randomized controlled trial trying to understand how much experts defer to AI systems. They gave a bunch of professional statistical analysts access to AIs to give them advice on a decision in a scenario.

What they found was that the experts made the same decision as the AI over 80% of the time, even though the AIs were split half and half. They also used the AI’s arguments to justify their decision over 70% of the time, but only 15% of the time did they say, “Oh yeah, I was influenced by this AI agent.”

So we have this world, which I think is one we’re going to be increasingly in, where humans are often almost becoming wrappers on AI agents—

Whoa.

—and AI systems, where all we’re doing is being the faces and typing the thoughts. I feel like there’s all this concern about AI takeover and rogue AIs, and I think a lot of the time we’re just going to let the AIs take over because we’ll be outsourcing all our decisions and won’t even realize.

Oh, wow.

So that’s what I’m worried about.

That is a cheery thought. As somebody who professionally takes other people’s opinions and repeats them as though they were my own—

I just said—

—I do kind of sympathize. It’s just usually the people I’m stealing from are people, not artificial intelligence.

Oh, yeah.

People say that AI makes everybody a manager. Here we are.

Yeah, exactly.

That’s exactly—

Oh.

This is exactly what managers do.

Man.

Wow. Erin, what about you? What are you tracking?

I cover venture capital, as you know, and one of the most famous firms, Andreessen Horowitz, recently made a big change to the branding and the name of one of their sub-funds, their a16z crypto fund.

Mm.

They’re one of the biggest crypto boosters and have been for many years. Hard Fork listeners might remember Chris Dixon, their top crypto investor, coming on for an interview a few years back to talk about the importance of blockchain and crypto for the future of the internet.

This would make it look like they’re backing away from that promise, and that’s something that happens in VC a lot. VCs chase trends and change. It’s not just a trend-chasing business; it’s also a marketing business, because all money is kind of the same. In order to stand out and get entrepreneurs to come to you, you have to put a stake in the ground and say, “This is what we’re interested in,” or, “This is what we’re really excited about for the future.”

They were really big bulls on crypto. Despite the fact that crypto is super out of vogue right now, and something like 80% or 90% of all VC dollars are basically going to AI, they’ve been very careful to say, “We’re not backing away from crypto.”

I mean, it’s like—

“We’re just expanding our mandate.” And—

It’s just all Frontier now.

Yes. And this is something other crypto funds have also been doing. They’re not alone in this. There’s a fund called Paradigm that was started by a Sequoia investor. In 2023, they very famously removed some of the crypto language from their website, and people noticed and freaked out.

A big part of it is that there are probably just not that many people building investable crypto startups right now, because everyone wants to build in AI. So it’s kind of necessary. One of the crypto industry’s many problems is that without the hype, without the number going up, there are just fewer people interested in working in and building in the industry. To find opportunities to deploy billions of dollars into startups, you have to change your mandate.

When crypto is too toxic even for Andreessen Horowitz, that’s when you know the vibe has shifted.

Yeah.

Here at Hard Fork, we embrace our crypto nominative origin.

That’s right. Our roots in the blockchain.

Oh, my God.

We’ll be sticking with this name forever.

Wow. Well, guys, thank you for coming through. This has been a really fun discussion. I’m so glad to have you, and I hope you all come back soon. This has been Mike Isaac, Jasmine Sun, Aaron Griffith. I’m Max Read. Thank you for listening.