Steven Sinofsky & Balaji Srinivasan on the Future of M&A, AI & Tech

For 4 years, it was just a desert. A bunch of companies silently died. The state blocked IPOs. They’re blocking M&As. There’s an all-out anti-tech assault.

DC is this zero-sum game. There’s something positive out there, so it has to accrue to the DC power base.

Figma managed to make its way through that, absolutely no thanks to the state attacking it. Then Lina Khan decided to take a victory lap on this, which, as I said, is like the assassin congratulating themselves for helping to elect Trump.

There’s been a lot going on in the world of M&A in the last month or so. There’s been the Meta–Scale AI deal. There’s been the Windsurf saga. There’s been the discourse—the Lina discourse—surrounding Figma. I wanted to bring both Balaji and Steven together to reflect and discuss at a higher level how we make sense of what’s happening in M&A land. Balaji, I know you’ve had some thoughts recently, so I thought I’d let you open.

There are 3 separate issues that are all kind of interrelated, and they’re all related to how U.S. capital markets are becoming tougher, while the internet capital markets are opening up. Those are the Windsurf, Scale, Character, and so on—the new kind of deal structure—then the Figma IPO, and finally the new GENIUS Act.

To briefly summarize: since Sarbanes-Oxley in the early 2000s, which was passed in the wake of Enron, the intent was to stop Enrons, but what it actually did was stop IPOs. The number of public companies has just declined, and the number of IPOs has declined. Tech companies started going private for longer, and then, with the FTC’s antitrust harassment of the last several years, that also started to cut off the M&A window.

The DOJ interfered, for example, with JetBlue’s acquisition of Spirit. Spirit went bust. Roomba had issues. A bunch of companies silently died.

Go ahead. Roomba.

Roomba. Roomba. Sorry Rob.

Robot company.

Yes. They said Roblox is fine. Sorry. Roomba.

You’re right.

So, first, the state blocked IPOs, and we had to go private for longer and build a whole private-equity kind of model. Then they started blocking M&As. Essentially, that caused—among other things—the assault on AI, with limits on the number of FLOPs, and the assault on crypto, with the SEC essentially doing lawfare against the whole space and debanking companies. There’s just an all-out anti-tech assault.

The amazing thing was that Figma managed to make its way through that and actually get to an IPO, absolutely no thanks to the state attacking it. Then Lina Khan decided to take a victory lap on this, which, as I said, is like the assassin congratulating themselves for helping to elect Trump. It was quite a remarkable statement, but it really gets to the heart of the way DC is a zero-sum game. There’s something positive out there, so it has to accrue to the DC power base because otherwise there’s not enough positive left over. It was this absorbing of the 1 glimmer of hope that’s out there and ignoring the long tail of carnage that they’ve recently caused.

That’s right. Basically, they take credit for the good things and the bad things. “Oh, well, that company must have sucked anyway,” or something like that.

The DC thing—I’m not sure this is apocryphal, but I remember Gates said something in the late 1990s or early 2000s, before the whole antitrust thing, which you guys actually had to go through. You had to go through the whole thing. It was something like, “He wanted nothing to do with Washington at all. He just wanted to code.” And some politician said, “Well, that’s fine, but we’re just going to hold hearings on you, and then you’re going to have to donate to us.” Do you remember that? I remember somebody wrote that up. Go ahead.

I think part of what you said brought up 2 things for me. One is that the whole thing about computing is that, except for the IBM antitrust case—which started in the late 1960s and pretty much lasted through me in high school—the whole world of computing just arose without government regulation, without government oversight, and even with the internet, with just government funding.

It’s a very weird thing if you’re in the business of governing and regulating: this giant, giant thing that swallowed the economy happened without you. What’s interesting is that the entire software industry, in a sense, happened that way, too. You never needed to be licensed to be a software engineer. There was never approval to sell software.

The most the government got involved was when it used to prevent mail-order laptops and desktop computers because they had radios in them.

They needed FCC approval.

Yes, they needed FCC approval in order to finally ship computers to homes. A guy named Dan'l Lewin, who ran the Computer History Museum, really cracked that while working for Apple in the early 1980s. That’s how the Macintosh made it to campuses: they figured out how to work around the FCC. Michael Dell had to do the same thing in the PC world.

You had this whole industry that swallowed the economy, basically, happen without any hearings. There were just no hearings. It’s remarkable when you think about it.

My framework for that—and I don’t use this for everything, but I think it’s a useful framework—is network versus state. The network—the internet—and the state, meaning regulations and government, as well as informal things that are aligned with the state.

Because the network is intangible, the scale of the internet and how quickly it grew is still something that, even today, people don’t fully grasp. There’s this thing, which is that it takes an enormous effort to see what’s in front of one’s own face. What’s in front of one’s own face? A phone. A screen.

The internet is simply the most popular thing in the world, perhaps in human history. It’s completely ubiquitous. It’s upstream of AI, upstream of phones, upstream of drones, upstream of everything on social media, and so on and so forth. It’s upstream of the election. Twitter elected Trump, then Twitter deplatformed Trump, and then X elected Trump. The internet is upstream of everything, and yet, because it’s invisible, we don’t think of it as a primary actor.

One of the most remarkable things future historians, when writing about this era, will say is something like: you need a license to cut hair, and you need a license to do this or that, but you didn’t need a license to own a computer—the most powerful device ever created. Thank God.

It’s a really good point. I’m not going to try to be the other side, but I can sort of defend that side. It is true: you needed a license to apply makeup in a salon or give a massage, but not to write software to change the world.

It’s interesting to think about the mindset of the regulator because, of course, all of the antitrust laws—whether you start with the Sherman Act or the Clayton Act—were based on a very tangible, distribution-constrained, resource-constrained world. You can’t own the railroad tracks, the railroad cars, and the coal or the ships that use all of them, because that’s this vertical-integration thing.

The PC industry just didn’t have any of those constraints, and the computing industry only had them for that brief time with IBM, when it cost tens of millions of dollars. IBM actually chose only to lease the computers, not sell them, which makes a ton of sense in the world of technology because owning a depreciating asset that doesn’t matter in a year is actually a bad idea.

All of those laws came about for that world. Even these crazy elements of merger-and-acquisition analysis, when they try to understand market share—which is not in the law; there’s nowhere in the law that says this—they hire people and compute the HHI index. That’s a way of taking all the players in a market. Right away, you presume that the market is well-defined and has a certain number of players in it.

Then they take each one’s share, meaning you can actually measure it, square the share, and add those all up, and divide by the number of players. They decide if it’s greater than 0.25, less than 0.75, or so, and then they say, “Monopoly.”

This has been the challenge in computing forever. Just take the question, “What is the market for word processors?” Is it the thing called a word processor? Is it everywhere you can type? Is it only if you print it? You very quickly can’t figure out even what the market share of email is. Is it the client? Is it the server? Does it depend on features? Is it mainframe-hosted, minicomputer-hosted, PC-hosted, or now cloud-hosted?

Steve Jobs put up a slide at the iPhone launch showing mobile-phone share, but he very deliberately chose to measure it by the manufacturer of phones so that it diminished the share of Windows Mobile.

But he also could have just done it by operating system, which would have made a completely different chart. But what was the right way to define it? When Microsoft was going through antitrust, we had 100% share of the Windows market.

Well, duh. And this is what you get into when you look at a deal—not even to pick on anyone, but to pick robot vacuum cleaners. What do you count as the share of robot vacuum cleaners? Do you count Optimus in the line—the share of robot vacuum cleaners—or is it only the spinning ones that cats sit on top of, that dock in your living room corner?

Because then, they've always struggled with this, but they don't admit it. They apply this sort of econometrics to the thing that makes it seem like it's this perfectly well-defined, well-reasoned thing. Even to this day, I don't think people would say it would be correct to define the phone market as iOS versus Android, because China would have something to say about that. Android itself is just not the same product across them.

Actually, I have 3 reactions to that, because I think there's a bunch of stuff I just want to shoot at. The first is on the definition of a market. Eric Schmidt also talked about this, but, for example, when the iPhone came out, people didn't think of it as a competitor in terms of being a camera. But it was one of the most popular cameras because it was ubiquitous, had essentially zero incremental cost, was internet-connected, and was programmable.

Even though the image quality was very poor and the number of pixels was low relative to a DSLR or whatever, it was a very popular camera, but it wasn't thought of as a camera on a primary axis. Often, the late, great Clayton Christensen framework—the whole disruptive innovation concept—is that something comes in and it's not really recognizable as a peer to the existing products in the marketplace, but it's better on some critical axis.

It gains adoption in that way, and then eventually it's a substitute, but it takes a while for people to even acknowledge that it's coming in there. Then you also have a fuzzy-set sort of thing where, for example, Google and Apple compete on operating systems, Google and Facebook compete on ads, and Facebook and Apple compete on headsets, and so on and so forth. There are lots of fuzzy-set overlap kinds of things.

And another point is the entire concept of, “Oh, look at all the tech monopolies.” Benedict Evans has this good saying: “Think about all the tech monopolies. There are so many of them.” Right, which is—

Really, the issue is they try to use these formulas sometimes as a substitute for judgment and sometimes as a mask for animus, right? Elizabeth Warren was saying she wants to build an anti-crypto army, but really she wanted to build an anti-tech army. That's just a particularly explicit example: she's in a tribe that's against our tribe.

And as you said, to put yourself in their shoes: at first order, it's people of the network versus people of the state. It took me a while to understand that, but fundamentally, what they want more than anything else is to get a piece of the state—to get a baton, to be able to be assemblyman of this or undersecretary of that, to have a piece of the state, this baton, and allocate capital and start doing things for the good of the world, making you do this and forcing you to do that.

It's all about coercive power over others, status, and the use of force, implicitly or explicitly. Whereas our framework is the total opposite. We don't want anyone to have power over us. We're not asking to tell anybody what to do who didn't consent to being in our organization. We just want a bare domain name like reddit.com, a field that we can build up on ourselves, and no one tells us what to do.

That doesn't mean there hasn't been an investment or something. Okay, maybe it's a board seat or something, but in general, you're just able to build on your own. These 2 kinds of things can coexist for a long time so long as we're just typing and doing math, and they're off regulating, bombing countries, or whatever it is.

But as the network grew and grew and grew and became state-level—

It's funny, right?

Yeah. As the network grew to tens of millions and hundreds of millions, and eventually billions of people, we're like, you know, this thing just grew to state level. And now, when these 2 things started to conflict, we felt, legitimately, that we're the CEOs, the founders of these companies. We built them from scratch. We should be able to have authority over what happens on these networks.

These guys started to see, “Wait a second. Their authority, whatever it is on paper, is actually being limited in practice.” Because, for example, let's say you're a taxi regulator. The actual regulator of taxis is Uber or Lyft, because they have real-time tracking and star ratings on both sides. Or you're the FCC or something, which issued licenses. Well, the actual regulator of speech, or what have you, is YouTube or Facebook.

So our expansion—our peaceful, invisible internet expansion—started implicitly taking market share away from them and sapping regulatory power. The empire struck back. They attacked us hard. And I can actually, if I squint, understand how we would think like them, and then vice versa.

How would they think? They don't want others to have power over them. Well, they don't like the fact that the network is getting so big it's able to dwarf them or whatever. How would we want to have power or something? Within our organizations, we want to be able to flip a switch and make something happen. If there's resistance to that, that's a huge pain. We want to be able to make that happen—reorganize the company or something—so that it can be better and more functional.

There's another way of looking at it. Have you guys ever seen the political compass?

Yeah, yeah.

Yeah.

Yeah. So it's like top-left, authoritarian left; authoritarian right; libertarian left; libertarian right. And so, if you roughly, roughly, roughly sit in the lower-right corner—adjacents are often allied, but diagonals don’t get along—you know, as a libertarian right, I can sometimes understand the nationalist, and I can understand the libertarian left, but the authoritarian-left quadrant always seemed foreign to me until I ran a large tech platform. You know why?

And of course, you probably had this experience as well. But the libertarian-right framework is everybody has consent, and you pay them to do something, and it's a market-based process. That works for many things, but let's say you're running a giant tech platform with hundreds of millions or even billions of people and you want to change some parameter set.

You cannot put that up for auction or discussion with everybody on everything, right? Instead, you're going to just flip a switch and they're all going to be basically opted into this, and they're not going to get paid; they're just going to do it, right? Because otherwise, there'd literally be no way you could possibly have that level of coordination.

Every algorithm change, every update, every this, every that cannot be something that's discussed with everyone. You have 5 million parameter settings in any system. Even Chrome, as complicated as it is, sets a zillion of those settings as defaults, right?

So you have to pick defaults, and for the most part, the platform will actually know better than the people because it's got all these analytics and so on and so forth. You can put yourself into the Elizabeth Warren headspace if you think of yourself as a system administrator of a platform where you have lawful authority over it and you built it from scratch, and so on and so forth.

I think the difference then boils down to competition law, right? The thing is, these platforms are at least competing with other platforms, and if Apple or Microsoft or Google makes too bad a decision, then the people on those platforms have exit. They have choice. They can move between platforms, so there's ultimately a market constraint.

But the actual monopoly is at the D.C. level, where there wasn't a practical switching option between things. So they could just mess up the platform choices all the time. And us being the apps on that platform—if you think of the state as a platform and companies as the apps on that platform—we would not have that much choice. Let me pause there. Get your thoughts.

Well, I think that's a fantastic observation. I mean, one way to think about that is—and we probably won't even agree on some of this—but if you look at what the European Union has done with app stores, you actually see that dynamic playing out precisely.

Apple basically said, “Look, we want to build a platform the way we built PC platforms. We understand what it was like to build a Mac. We know how security violations happen. We know how privacy violations happen. We know how quality degrades over time due to software and third parties and apps. We know kernel mode versus user mode. We know all of this stuff.”

When we built the iPhone and the platform for the iPhone, including the App Store, we whiteboarded it out, and we deliberately said, “Well, we're going to constrain the API so that apps can't steal information from other apps. We're going to be secure, and so we're not going to run a bunch of stuff in system mode.”

There's not going to be third-party drivers, so there's no kernel mode—all these things. Then the European Union comes along, after the success of that, and says, “Good idea, but we actually want to return the phones to being PCs again.” So now here's our Digital Markets Act, which basically says phones have to be back like PCs.

Apple's like, “Do you understand? We literally set out to be more secure and to be private. You, the GDPR people—we actually wanted to solve this problem on the phone, and we wanted to be secure.” And they're like, “You're right. So we're going to put in a thing that says vendors should be allowed to be secure.” And you're like, “What does that mean?” They're like, “You have to go figure that out, but just know that, as the regulators, we're demanding that you be as open and free as the PC and secure.” You're like, “We did that once already. That is precisely what we went and did.” They locked you down.

And so you get in these crazy loops. It's actually not unlike the loop that the media had with the internet. The media said, “We believe in curation and editorial control, and we own the distribution.” The internet was like, “We have a way of doing distribution, and we have a different view on curation and control.” Then you get into these loops where the media decide, “We like the distribution that comes with the internet, but now we want to constrain the distribution. We like the distribution, but we want to editorialize the distribution.” Then the regulators come in.

That diagonal on the political compass is really just, “I'm actually walking in your shoes right now, and I realize what it is that you did, but I want that, and I don't want the tyranny of the OR. I actually just want security and openness.” They just—I think a big part of it, honestly, boils down to the fact that they're simply not numerate.

AI agents were helpful because they allowed me to model. I think all 3 of us are fairly verbal people, right? We can write, but we also have System 2 thinking, in Kahneman's phrase, or what have you, where you can just go heads down, you can program, you can do the math. The numbers actually have to add up. You have to do the spreadsheets. It has to be a numerical thing.

One of the things that I realized is that a good chunk of the people who are in the American state—not the Chinese state; that's a different thing, and we could talk about that—but the American state, a good chunk of them, are selected for having verbal and not numerical or mathematical ability. That's distinct from, let's say, the 1950s, when marginal tax rates were at 90% in America. They were at 100% in Soviet Russia: go to jail, do not pass Go, do not collect $200 or 200 rubles. They were at 90%, though, in FDR's America.

There was this book by William H. Whyte called The Organization Man, where it was an extremely centralized environment. You couldn't really found a company or anything like that. It was very hard for Shockley and Fairchild and so on to do what they did. But at that time, the Elons of the world would have worked at NASA or run NASA, and the Patrick Collisons would have probably run the Federal Trade Commission and so on. Whatever was written down legally, they would just call each other and make it work, right? You'd have a bunch of CEO-level people, founder-level people—because they couldn't found companies—channeled into the government to make it work.

Similarly, in the Soviet Union, when those guys couldn't do entrepreneurship, they put all their energies into pure math and science. That's why there are some amazing Soviet mathematicians and physicists and so on, if you're familiar with that, because that was an area where that kind of technical mindset could at least do something.

Anyway, because they're selected for this, one example is Lina Khan, who told Michael Solana that he was a billionaire, right? Or how Bernie Sanders is like, “Millionaires and billionaires,” which is like saying meters and kilometers, right? Because there's actually a 1,000x difference. Or when Binyamin Appelbaum, a member of the NYT editorial board, said that Bloomberg could give his fortune, divide it, and give $1 million to everybody, right?

So that's 3 examples where I really start to think they don't—they think “billion” means “big number.” A primitive tribe will have numbers for 1, 2, and many. They just don't understand 1e9: the difference between a billion and a million, or the difference between, for example, someone who has $1 billion liquid, someone who has $1 billion net worth, a $1 billion fund, and a $1 billion valuation. This leads to the fact that they literally can't—it’s not like they can't do machine learning and they can't do gradient descent, and so they can't divide.

To have some sympathy for them, if I were to say, “What's the difference between a picofarad and a microfarad?” Unless you've done something with hardware, you wouldn't know. What's a lot of capacitance and a little capacitance? There's a scale there for capacitance or inductance or something like that that, unless you've actually done electrical engineering, you wouldn't have an intuition for.

But they just have no intuition for scales of money beyond their personal experience, like $1,000 being beyond something that's in their bank account or checking account. They have no idea what's above that because they haven't run organizations or made investments. A billion might as well be a trillion, might as well be a quadrillion.

Why don't we bubble it up a little bit and talk a little bit more about this M&A stuff? I'm completely—

I'm completely fascinated by M&A.

Let's just take it in general, not about Figma, but this crazy revisionist thing that goes on with M&A. For me, one of the big things is that you have to start from the premise that, when a giant corporation does M&A, it's literally like a speculative investment.

It's a power law, but for M&A. Exactly. I love that you said that.

Right. That has a power-law return. There are only 2 truisms about corporate M&A. One is that it's provably a net destroyer of value. No matter how many studies get done at HBS or MIT Sloan, I literally went—when I was teaching at HBS, I spent hours in the library and pulled all these papers with math and calculus in them that were against M&A.

I found M&A at Microsoft very difficult to pull off because Bill was mandating constant synergy and synchronicity across products, and M&A was a huge perturbation to that whole system.

With the exception of something like PowerPoint, which was a huge one.

I'll get to PowerPoint, because that's near and dear to me. But all the business literature on M&A says it's a destroyer of value. So you'd think that the regulators would be out there against M&A, not because of the success it has but because of the failure. They should be saying, “Hey, companies, you shouldn't buy companies because you just destroy them.” But you never hear them doing that. And no company sets out to destroy a company.

And so I actually did the bibliography. I actually brought my visual aid this time, which we'll put up on the screen. This is what The New York Times—your favorite NYT—said when Google acquired YouTube. The headline on the front page of The New York Times was—

“Dot-com boom is echoed in the deal for YouTube,” followed by about 500 words about copyright infringement, how they overpaid, and how it was just 5 guys and kitten videos. It had pull quotes from all these people explaining what a disaster it was going to be.

All right. Well, I've got one. I love that. Can you put this one on screen? Instagram was something where, at the time Facebook bought it, Jon Stewart made fun of it because it had no revenue.

Little square pictures, too—retro square pictures with filters.

Yeah, exactly. Jon was like, “The only thing that would be worth a billion dollars would be something that would instantly get me a gram of coke,” or something like that.

The thing is, Instagram today has been retconned as this evil, obvious move for the evil monopolist. At the time, Zuck was supposedly an idiot for doing it because, first, it had raised money at a $500 million valuation the day before Zuck offered a billion, so he doubled the valuation. Second, that was about 25% of Facebook's $4 billion in cash on hand. Third, it was weeks before the Facebook IPO. Fourth, the board wasn't consulted. Fifth, Instagram had no revenue.

The balls to do that when you're going into an IPO, when you need to reassure the market that this boy-genius CEO needs adult supervision—blah, blah, blah. Why would Facebook pay $1 billion for a company with no revenue, with Facebook's public offering only a few weeks away? Spontaneous and improvisational business moves become more curious. Hacker News said, “This is not going to be one of the best tech acquisitions of the next decade. Instagram is a photo service in a sea of other photo services.”

Can someone please tell me how Instagram’s actual content is worth anything? It seems like mostly a huge waste of cash. And then, at a market cap of $950 million, the New York Times is worth, quote, less than Instagram.

True. True.

That one I’ll give them. That’s true, I think.

Yeah, but I think you also have to actually read the reasons why people thought it was valuable, because the reasons end up being these very pedestrian, sort of non-math, unable-to-see-exponential-growth, not-strategic arguments. They always fall back on something like, “In fact, Instagram had no revenue.” YouTube had no revenue, but it was also this morass of copyright violations: How will Google ever figure that out? No one ever wrote about the potential.

The potential is the venture bet that the company is making. The interesting thing is that big companies make the wrong potential bet 90% of the time. They generally think they’ll do something like an HP Autonomy acquisition, which was a very famous off-the-rails acquisition in the enterprise software world. HP thought, “This is sort of an nth-tier player in enterprise search and information retrieval, but we’ll buy them and put our magical sales force and platform strength behind it, and that will fix everything.”

That’s like 90% of the M&A that happens: a big company assumes that whatever it’s strong at, it can just wave that dust over this failing business and make it great.

Yeah, that’s right.

And I think that’s what it’s always like: This venture aspect is what’s missing in the retcon that they should have stopped it, or that they should go back and stop it, because it became a success against all the conventional wisdom at the time. It completely blows my mind that that’s the framework people would use for evaluating M&A. Nobody’s going to go back and retroactively consider whether Roomba really would have been better off being bought by Amazon.

Everybody wants a piece of the reward. No one wants a piece of the risk, right? So when the state goes and blocks these acquisitions, they assume no downside risk. They’re not taking, for example, “Hey, Adobe had to pay Figma $1 billion for the breakup—the breakup fee, or whatever—because the deal didn’t go through.”

The state goes in and meddles, and then these people have the temerity to take a victory lap. It’s really just stolen valor. It’s stolen valor, right? Lina Khan: stolen valor, right? And genuinely, Dylan Field and the Figma team should be superheroes for this.

But just to talk about that for a second: with M&A, you made a bunch of good points, Steven, and I want to add to that. First, absolutely, there’s a power law for M&A, just like there’s a power law for startups, and the best M&A you do can completely transform your company. A lot of M&As essentially fail, and sometimes it’s a little bit unpredictable.

Second, I think, in general, a company usually needs, in my view—and you may disagree—to be about 100x the size of the small company in order to acquire it. The reason is that if it’s even only 10x the size, I can only think of 1 deal where it was about 10x the size and it worked. That was Illumina’s acquisition of Solexa, which was a must-win deal in the genomics sequencing space, or genome-sequencing space. There was a really important technology there that became the basis for everything Illumina did for the next decade, and the entire executive team was bought in on it.

Ten percent of your cap table—10% of your equity—is a huge amount. It’s basically more than you’re going to spend for the whole year, maybe for multiple years at a time. So a 10% bite is massive. It really has to be 1%, and that’s still a whole integration effort. That’s number 2.

The reason I say that is that lots of founders at various stages will say, “I’m acquiring another startup,” and I’m like, startup-to-startup deals never work in general because neither of them has money. Maybe one of them can shut down and join the other one. That sometimes works, once in a while, but in general they don’t work.

That’s why M doesn’t work but A works. A merger usually doesn’t work, but an acquisition works.

Well, AOL and Time Warner and stuff—what a huge disaster that was.

Yeah, exactly. Again, once in a while, it’s something like Steve Jobs and Pixar, and all the ones that work are sui generis, where they really acquired some amazing founder as part of that who then leads the company or something.

The third thing about M&A, as you again said, is that the smart big company values the acquisition on the basis of the big company’s distribution. It’s product times that distribution; that’s something. However, the dumb big company just thinks it can roll up anything and sell it, and that just doesn’t work.

I think one of the huge responses, by the way, on this Figma thing—and then let’s get to Windsurf. I want to talk about that as well, and then also the GENIUS Act—the response on the Figma thing fell into 1 of 3 categories.

The first was the Elizabeth Warren school, which is just the anti-crypto army, the anti-tech army. They hate tech guys, and I actually like that because that’s just pure tribal animus. Okay, meet me on the 50-yard line. You bring your guys, we bring our guys, and let’s win the battlefield of ideas. I actually prefer that because that’s explicit conflict. It’s tribe versus tribe. You’ve got the war paint on. Okay.

Then you’ve got the well-meaning, maybe—but I often can’t tell if they’re trying to kill us or they’re actually arsonists or whatever—which is, “Oh, we’re going to have more startups if we allow them to become big and not be eaten by these other companies,” or what have you.

I struggle for the analogy, but it’s like you can’t hire somebody until you interview 20 people, because then they’ll be the best of 20 people, and we’re going to let you hire the best of 20 people. What it does is, first of all, you shouldn’t be interfering with that choice.

Second, if you cut off the flow of M&A, obviously most companies aren’t either good enough or able to make it all the way to IPO. Oculus, for example, about 10 years ago, was burning a lot of cash. It probably couldn’t have made it to IPO. A lot of companies are like that: They’re burning cash, but they’re valuable to a big, deep-pocketed acquirer, and there are maybe 5, 10, or 20 guys who could buy them, whatever the number is, but they really can’t operate as a standalone company. They’ve gotten to proof of concept, and that’s enough, right?

Airlines are often like this, where there are a ton of fixed costs that go into it, and mergers make sense because they add routes and things like that. The JetBlue–Spirit one, right? So when they block those deals, they’re actually destroying value.

Moreover, one of the biggest issues—and this is related to regulation in general—is that they think of it as, “Oh, this is punishing the big tech companies.” Even though it’s an annoyance to them in the short run, in the medium to long run it makes them stronger. If the big companies can’t buy, first they’ll figure out other things, as we’ll get to with these complex deal structures. But second, that means less money for startups.

When a big company makes a big acquisition, that’s a big surrender, because it means a big company couldn’t have built it themselves. Google had Google Video, but it had to buy YouTube for $1.6 billion, which would certainly cause, I’m sure, some churning internally among the Google Video guys. Most of the time at big companies, there’s some faction inside that’s like, “We could build it ourselves,” or, “We’re paying too much,” or something like that.

So it’s often a surrender for a big company to do this. It’s not what they wanted to do. They didn’t want to pay $1 billion or whatever for this, and then that surrender money goes and excites everybody. They’re like, “Let’s make a million Instagrams.” When you see a big $1 billion acquisition for Instagram, then you get Snapchat, and you get TikTok, and you get all these other competitors.

For Facebook, it’s like throwing fertilizer on something that springs up 1,000 competitors that all start attacking you. So the actual way of regulating big companies is with 1,000 startup piranhas, not by this regulation.

That’s a fantastic observation, because you always remember that in a big company, whenever something new that’s adjacent pops up, the immediate reaction is, “Okay, we’re selling this giant blob of software. We’re selling this giant blob of software, and this thing is adjacent to it, so our blob needs to have that thing.”

That’s what immediately gets the antitrust regulators: “Oh, but that’s expansion by leverage, or by tying, or something like that.”

Tying? Oh, my God. What does tying mean? Tying means you’re tying peanut butter and jelly together in a sandwich. Tying is business strategy 101. Now, 10 years later, the truth can be told.

I'll tell you,

Right?

Microsoft did not go.

But—and so you have all these meetings at a big company, which is, well, should we make it or should we buy it? It's just make versus buy. That's the conversation that you're going to have. And, of course, this is the funny part about—

Big companies find it hard to make, though. But go ahead.

But this is the funny thing because, if the regulators get involved, then they get all the memos and all the emails. It turns out that, inside the company, half the people said, “We could make it,” and half the people said, “We could buy it,” and all the people said, “We have to have this thing.” They said it with varying levels of hysteria. It was, “We have to have this. This is going to put us out of business,” or, “This would be really nice, and we have some customers on the periphery asking for it.”

So which one of those enters discovery for the regulators: the hysterical person who's probably the person on point, losing a deal in sales, or the engineer who is mesmerized by the exciting implementation of something? That's exactly the psychology. But then they still go back and have the make-versus-buy. They always make 1 of 2 choices. They almost never really just try to make it, but if they have to because the one that they want—there's only 1, they can't buy it or whatever—it's very, very hard to succeed on the make-versus-buy when you choose make.

So then you go buy, and there's a fork in the road. About two-thirds of the time, the big company says, “Wow, the leader is really expensive, but we have our magic distribution beans, so we're going to pick number 2 or number 3, which is way, way cheaper, and get it in a fire sale.” And, of course, that never, ever, ever works, like Microsoft—

When does that ever work? I'm actually trying to think.

It doesn't. You could—you have Google buying Motorola.

Oh, yeah. DoubleClick. Google DoubleClick. And then what is it? aQuantive.

Microsoft bought aQuantive for $5 billion. You had Sprint merging with Nextel, which was like 2 number 4s, if that was a possibility. You have Microsoft and Nokia. I made a giant, long list. You had everybody in the phone business that needed—or everybody in the chip business that needed modems. Then they went and everybody bought these number 2 or 3 modem makers.

NVIDIA almost got overtaken by a private-equity raider because it bought a modem company, and that was a signal to the market that it was completely confused about gaming graphics. Why would you compete with Qualcomm from a gaming graphics company? This just goes on and on and on with that kind of thing.

But then you still get to the point where you want to buy something, which is still a power-law return. Again, one of the things that doesn't get taken into account is that venture investments or acquisitions by a company can actually be really transformative to the big company, which is something that the regulators don't really see because they see the world as a static, fixed pie.

So they think, once a company is like IBM and owns mainframes, or is Microsoft and owns Windows, well, that's just what it should do. It should then just make Windows forever and just be the Windows company. This is where tech people are very, very different because they just assume tech has this finite sell-by date and that the tech is just not going to be all that useful down the road.

So you have to reinvent yourself. It's not like people in 1985 thought Apple was going to be a phone company. That whole mindset just sort of escapes people. Here's an example of an acquisition that was hugely transformative for Microsoft that nobody knows about today.

In the throes of the rise of the internet, in 1996 or so, we bought a company called Front Page, which was basically a word processor for the web. It was a way to design a whole website and also do something that nobody else did: you could edit on a PC, push a button, and those things would end up on the internet. Like that.

No, it was—I mean, at the time, it was actually a pretty good—

It was super cool. But what it did to Microsoft was it galvanized Microsoft to say, “What's really important on the internet is editing.” Nobody was really solving the way to edit this very finicky thing called HTML and publish it to an Apache web server.

I was in Office; we did the deal, but then we had a fight with the Internet Explorer team, who thought they should do the deal. They didn't want to do the deal until they saw us wanting to do the deal, which is the whole way things work at a big company. We ended up having to solve the bidding war within our company first, and then we had to get on the phone with Marc and do the bidding war against Netscape for this company in Boston.

The thing that it did was galvanize Microsoft to say, “What's really important on the internet is editing.” Nobody was really solving the way to edit this very finicky thing called HTML and publish it to an Apache web server. We finally figured out that editing on the internet wasn't going to be like Word; it was going to be a different kind of tool that involved scripts and programming, and that's what led to a series of things.

The people we brought in were experts in the internet and editing on the internet, which we just didn't have. Although the product never materialized as a big Microsoft thing, the people infused that DNA into the company that enabled Microsoft to figure out how to do editing in a browser, which turned out to be incredibly important.

That kind of thing is transformative, but it also transformed the whole industry. Where would we be today had we not figured out these dynamic websites and the way you could edit the web in the browser? That stuff wouldn't have happened because we were just not innovating there.

I feel like that whole thing is missing even from the Figma deal. Again, the specifics of Figma aren't really super important, but it was a whole new, innovative category of how to do tooling, which then gets to AI and tooling. I think that's a good way to get us to Windsurf and everybody else.

Okay. Yes. In general, one of the things that's been happening is that people can only remember maybe a name. It's like a fleeting kind of thing. They're like, “Well, Lina Khan is gone, so therefore nothing has changed.” U.S. versus Google is a giant antitrust case that's still going. FTC versus Meta is still going.

All the antitrust stuff is also not just in the U.S. It's all these other countries that ganged up on this Figma thing. Every lawyer at all of these companies has the number 1 priority: do not get us into some antitrust situation. Because of that, a lot of the big companies have been forced to, quote, innovate on deal structures and do things that are new.

Scale, Character, Inflection, Adept, Covariant, and Windsurf are all very similar, right? They're not all the same, but we have a typical acquisition. You have an acquirer—let's call it Google—and it buys a company. When it buys a company, there's a process, which most people watching the show will know, but if they don't, there's something called the capitalization table, which says who owns what shares.

There's something associated with it called the liquidation waterfall, which says who gets what money and when. For example, if there are debt providers, how they get paid, who gets paid in the middle, who gets paid at the top, the preference stack, who's at the bottom, common holders, and so on and so forth. The capitalization table and liquidation waterfall give a very well-defined process for who gets paid when you just buy the whole company—eat the whole thing.

Then you've got something like an acqui-hire. An acqui-hire is something where the acquired company doesn't really get any money. It's not usually done through the liquidation waterfall. Instead, the company just shuts down, but there's a press release that says it was acquired, and then the team goes and gets jobs at the new company.

Maybe there's some cash that's given to the investors, but essentially, it's way better to at least get an acqui-hire than to have a total go-to-zero moment because you get the status, if not the money. That's one way of thinking about it.

Now we get to the third thing, which is what these deal structures have. They have an acqui-hire component, but they also have what I'm calling the acqui-fire. In these 6 deals—in Scale, Character, Inflection, Adept, Covariant, and Windsurf—the big company basically bought the top AI researchers and engineers out of the smaller company and paid a huge sum for that.

But it wasn't actually buying the company. The company was left as a shell, or as an actually existing entity. In the case of Windsurf, they had 40 people go to Google, and about 200 people were left behind. There was a huge chunk of money left in the bank account of the left-behind entity.

It was usually set up—and it was in the case of Windsurf—in such a way that the money left in the company was what they would have received through the liquidation waterfall. The point is that, in an acqui-hire, you get the status but not the money. In an acqui-fire, you get the money but not the status. So that leads to the whole Windsurf drama.

In the other 5 acquisitions, you got—you guys seen The Dark Knight? Bane's like, “We need 1 of us to be in the wreckage, brother.” Whoever was in the left-behind company was somebody well-behaved enough to basically be like, “Okay, salute. I'm going to go down with the vehicle, divvy the money out, deal with it silently,” and so on and so forth.

It was like there's a line of succession for the presidency: president, vice president, I think the Speaker of the House, blah blah blah. You get to number 37, and it's the Secretary of the Interior or something like that.

It's Kiefer Sutherland, the Housing...

Urban Sutherland. Right. Right. Exactly. Right. So Kiefer Sutherland is the designated successor if the entire leadership structure is decapitated—or, in this case, acquired. If they're all raptured, right, that is the person who's left behind and is now the president.

Now, one of the things I think we need to do in our contracts is have the concept of a keyman provision.

Yeah. Yeah.

We need to have a non-keyman provision, which is: This is the designated executive who, in the event of an acqui-fire-like thing—and we can decide how to describe it—stays behind. It's not an acquisition, because if it were an acquisition, then you have all the FTC blah blah blah stuff, right? But in the event of an acqui-fire, this non-keyman stays behind. He gets maybe a little more money, or a lot more money, whatever we negotiate, because he's not getting the status of being acquired. But he executes an orderly shutdown of the company, doesn't have any drama, and just divvies out the money.

The issue is that this deal structure was new enough that the other 5 times it went fairly well. But in this context, what had happened—and just to give you some details that I'm aware of—first, most of the Windsurf employees had just been hired within the last few months because they were all sales guys. Second, Google, when acquiring the company, didn't want to acquire these sales guys because Google has its own sales team. Google just wanted the engineers, right?

Third, Google put $100 million-plus in the bank account of Windsurf, where the intent was to dividend it out. But the guys who were left behind didn't understand what was happening because they're sales guys. They're just thinking about money or whatever.

And the problem was, it was so constrained in terms of what could be said about what was going on, since it's not an acquisition, guys, right? Since they couldn't say anything about what was actually happening, the people who were doing the deal couldn't communicate clearly about what was happening. So it just looked like, “Oh my God, the founders left and they left everybody in the lurch. Oh, they broke the social contract.” And that's not actually what happened at all. What happened was the FTC and others had made acquisition so difficult that they had to do this other structure, and it resulted in the people left behind not getting the hint about this.

That's 1 interpretation. The other interpretation is the people left behind got money but not status. So after all, if you put yourselves in their position, normally in an acquisition, Google might have acquired a 250-person company and they might have kept 40 people, and the other 200 people they'd say, “Hey, we're not acquiring you.” But those people would have had a face-saving thing, and they would have had a line on their CV saying, “My company was bought by Google. I decided to do something else afterward.”

And you know what? That's actually a common thing because both parties have to agree. Both the big company and the small guy have to agree: “Hey, I want to still work at Google rather than do another startup.” And it's very common. Everybody has a broad, warm halo. Your exact exit number isn't published online. Whether you've got an offer letter isn't published online. So everybody who is acquired has a junction point where they can choose to go to the big company or not. And they have the status and the money, right?

So the issue with the FTC interference in that is it broke the status part of the transaction, where the guys left behind didn't get status. But they did have money. So what did they do? Rationally, for them, they negotiated a second acquisition with Cognition, where they got the status of being acquired.

Now, the issue is Cognition's like 60 people, and they're acquiring the 200 people of Windsurf. That gets back to our earlier point: Usually, a company can't buy something unless it's 10 times greater than the thing it's buying. So it'll be very challenging for Cognition, I think.

By the way, I have nothing against Cognition, nothing against Windsurf, nothing against any of the people here. I wish everybody the best. Cognition is an awesome company. Windsurf is awesome. Varun is awesome. I have nothing bad to say about anybody. I'm just describing the incentives, right?

So Cognition will find it challenging, I think, to integrate those 200 people. And it'll also be challenging for them to lay anybody off because they said, “Oh, we brought everybody on.” I think on the Windsurf side—basically, Varun's side—he's muzzled, so he can't say anything. And in general, my view is, usually, the guy who's getting pummeled on social media and can't speak is not as bad a guy as he's made out to be. He literally can't defend himself, right?

But to defend him, this deal is essentially the same as the other 5 deals. The difference is the people left behind didn't want to play the Kiefer Sutherland role, or what have you, just because—for whatever reason—which is their prerogative. The way we solve it in the future is a non-keyman clause, and there's somebody who's maybe paid more to shut down the company, turn off the lights, because that does suck. I grant that it sucks, and I understand why their egos were wounded and so on and so forth.

But ultimately, the person to blame—one of the other things that happens here is, in something like this, the last person with a face is the one who's blamed, right? Because Varun has a face, but Google doesn't, and the FTC doesn't. And the general anti-tech, antitrust, U.S. versus FTC versus Meta kind of stuff doesn't. So the last guy with a face is blamed, but the faceless stuff isn't. It's almost like Bastiat's “seen and unseen,” right? Blame the FTC, blame Lina Khan.

And there's 1 thing: Someone asked, “Well, why isn't the current administration reversing this?” The answer is, the current administration, for totally different reasons, I think they have a legitimate bone to pick with big tech because of the censorship and so on and so forth. But as a consequence of that, many of the cases that were started have been continued, right? So it's not like this thing just completely went away. The new administration is friendly to little tech, mostly, but unfriendly to big tech and continuing those cases. And then this is the big tech–little tech interaction effect that's going on there.

All right, that's a lot I just said. Let me pause. There's more I can say.

Well, those are very tough stories to hear, and 2 things really jump out at me. One is just purely about the shaping of the landscape and what's going on. I think these are extremely important—let's just call them deals—and the reason they're extremely important deals is because, with near certainty, or in general, but for me personally, we're undergoing a platform shift now with AI.

We don't know. I can't say who the winner is. I don't want to say it's not really important, but there is a shift in where the nexus of the broad tech ecosystem's energy is going to be, from mobile and cloud to AI. Now, whether or not that's a complete break, or the same players move to that transition, I don't know. But what that means is, first and foremost, the most exciting things that are going to be going on in the near term are in the tooling to enable the platform.

And that's especially true because the way that the AI platform shift is happening is there's just a lot of players and there's a lot of people. And it reminds me a great deal of the consolidation of the PC operating-system world, which was—there were dozens of PC operating systems in 1980. And when IBM came out with the PC and Microsoft came out with DOS, part of that consolidation was due to the implementation of the BASIC programming language, which had already gained strength across many of the platforms.

But this consolidation happened because, on boot-up, there was BASIC, and then this proliferation of tooling appeared on DOS because Microsoft invested irrationally in tooling. IBM invested irrationally in tooling, far more than there were any independent toolmakers.

And I think what's happening in AI right now, when you look at all of the energy around coding, is that this is really building the tooling for the AI era. And so it's going to be an irrational investment, because tooling itself is never a really huge business—you have to have it. And so it's sort of this: Well, if you have to have it, then there are going to be many alternatives. There are going to be some low-price ones, some high-price ones, but the people that want to have the predominant platform will invest irrationally in tooling. And so that's why you're getting these deals that don't look rational, because there's just a bunch of tooling.

The second thing is it's really important to put this in perspective if you're one of those people who think that this is kind of gross or hacking the rules in some way. Antitrust law itself was designed—if you go back to the Sherman Act, it was this very vague, barely 3 pages of legislation, and it was really designed to attack 1 specific thing. And the word “trust” in that context just meant contract. So what was happening was between the railroads and manufacturing and resources and stuff.

The way that interstate commerce happened, a company in one geography would sign a contract with a company, a provider, or a vertical partner in another geography. The interstate commerce laws had not yet really been established, and so it was sort of this free-for-all of exclusive contracts by geography, by resource type, and by train tracks. It was locking out whole parts of the country from the availability of those things. So this antitrust became: break up these vertically integrated or horizontally constrained entities.

When the Clayton Antitrust Act came along, it said, “Oh, you know, the real problem is pricing and tying.” And so then all the laws became about how much you can charge, whether you can have exclusive deals, and so on. At each step—the first time, Delaware came along and started being really favorable to companies that were doing business in multiple states. And so you ended up with this sort of—and I don’t want to get criticized by legal historians or business historians or whatever. I’m not playing fast and loose; I’m trying to be abstract about what took place over 30 years.

Then when pricing came along, businesses just started to develop all of these different ways of dealing with pricing. One of the most common things people know is that if you buy a lot of something, you get a better price. But if you actually read the Clayton Act, that doesn’t appear to be legal. And so it took a whole bunch of court cases to establish this very basic premise: the more you buy, the better price you get because I like good customers, or if you commit to not buying my competitor’s products, we’ll give you a good price.

And the Clayton Act was like, “You cannot do that.” And you’re like, “But that seems to be a fairly reasonable constraint. You’re not going to buy it from me and play that off my competitor, and I’ll be nice to you.”

At each juncture in the evolution of regulatory oversight, the next step could be viewed as a hack to the systems that had been put in place, which generates this animosity with regulators.

You could go back—banking is a classic one—where checking accounts didn’t have interest. Someone clever with software invented the notion that you have a checking account and a savings account. Your savings account has all your money in it, and the minute you write a check, we move money from your savings account to your checking account. It stops earning interest, we pay the check, and you’re covered. That was called a NOW account. That innovation allowed you to have interest on your checking account, which turned out to be a really big thing.

When MCI came out with, “We want you to use our deregulated long-distance service, but we want you to get a really good price only when you call 10 friends and family. We’ll give you a really good price,” everybody around the country signed up for MCI when phones were deregulated and had to make a list of all their friends. Of course, that turned out to be a massively great marketing tool, because then they would take the list you gave them, give you a discount for calling those 10 people, and then hit those 10 people up to be part of Friends and Family.

That was just a software innovation that completely worked around the idea that the price of long-distance should be the same for everyone everywhere. Then AT&T did it with free minutes, up to unlimited long-distance calling.

So what’s happening now is just: The regulations have been fixed for a long time, we want to invest irrationally in platforms, and you’re making this part of it very difficult. So we’re going to go figure out an innovative way. We have to be careful because, of course, they are going to circle back and make this difficult in some way. That’s the cycle you get in with regulatory oversight.

I was always the guy who used to stand up and fight about this. It feels like the guy in basketball who decided that when there are 7 seconds left, you should intentionally foul someone. I always thought that was the most unsportsmanlike thing because I’m like, they didn’t invent fouls to be executed on purpose. They did it so you wouldn’t poke the other guy’s eyes out.

But it became part of the strategy. And that is the ultimate American capitalism: exploiting the rules that way.

Well, the issue is also that it’s Silicon Valley.

Yeah. So I think what happens is the following: You start out in a totally honorable—I think fairly honorable—capitalistic way, and then when the government attacks you enough, sometimes the companies that survive get a taste for the One Ring.

They’re like, “Okay, well, you know what? We just built this huge lobbying team to defend ourselves. What if we go on offense?” They’re kind of corrupted by it in a certain way.

In chemistry, if you think about reaction kinetics, sometimes you can have a bunch of time constants where you have this reaction, this reaction, and this reaction. They’re all going, and you have to do the math to figure out which one goes first. I think there are several things that are all hitting at the same time in this space that I’ll just give in quick succession.

The first is that these big companies are now getting a taste—they’re forced to. They wouldn’t actually want to consider this in the first place—of getting the milk for free rather than buying the cow, right? Decapitation rather than acquisition.

Right now they know that that’s a thing. It’s actually faster than an acquisition. Just leave the money in the company. It’s almost like a deal; you’re just buying something from somebody. It’s closer to a big purchase order than it is to an acquisition, with all the complexities involved in that. So now they’re like, “Oh, I can do that faster and with less overhead. Let me have 5 of those.”

That’s one thing that’s happening: You’re giving big companies a taste of this. We’ll have to figure out our deal terms to account for that as something that counts as an exit but doesn’t count as an exit. We’ll figure it out.

The second thing is that AI is making it so that you can do more with less, more with fewer people. This will become more common, where there’s an internal amplified intelligence rather than artificial intelligence. You’re going to have a stratification within every company and between companies where the top people will become more and more and more valuable because they can do so much more, so much more quickly.

The third thing is that one thing people say about AI that I actually don’t agree with—I didn’t agree with it then, and I don’t agree with it now—is that this is the worst it will ever be. They used to say that, right? But I remember with Napster: Napster was actually the best it was, and then all the copyright lawsuits and attacks on it made it worse and worse over time.

Google Books was amazing, and then all these copyright lawsuits gutted it enough so that you could get some little snippet preview, and then you couldn’t see the whole thing. So it’s quite possible—I would even say probable—that the combination of all the copyright lawsuits—

Those are desperate lawsuits, by the way, desperate attacks by all these journalists, authors, writers, and so on who hate AI. I understand why they hate it, but they just hate it, so they want to kill the thing. You know what they’ll say? They’ll say, “Are you an AI supporter?” with venom in their voice. It’s like, “Have you not heard that one?”

Yeah, go ahead. I’ll follow up. I promise I won’t let you just get away with that.

Okay. Because they say, “Are you a Trump supporter? Are you an AI supporter?”

There are a few companies where it’s similar, actually. When Discord tried to roll out crypto, people were like, “You’re doing crypto?” People got super mad. That is a budding thing: an anti-AI, anti-crypto, anti-tech movement, this setting fire to the windmills. It’s a real thing that we should not just watch out for. I think it’s going to become the future political axis between futurism and primitivism. That’s going to be the new left-right after the whole thing finishes rotating.

The issue is that those attacks from a copyright standpoint, the energy constraints—because data center buildouts are going to start hitting spare energy constraints—and the fact that Chinese models are open, actually pretty good, and being distributed quickly all hit at the same time. China is distributing them quickly. Did you see my post on AI overproduction a few months ago? That’s happening now. You’ve got Kimi, Qwen, and DeepSeek.

These are good models, and they’re open. I shouldn’t say fully open source, because they’re open weights but not open source. They haven’t released the full source code to build them and all the complexity that involves, and so on and so forth. But they are open weights.

The combination of those 3 things means it’s quite possible. The fourth is that I already saw something where there was some government restriction on using hosted DeepSeek. I could understand that hosted DeepSeek is going to China, but I wouldn’t be surprised to see something where it all combines such that U.S. AI companies are hit with copyright lawsuits, they’re blocked by a lack of energy, the Chinese open models are out there, and U.S. regulations prohibit people from using the Chinese open models.

So that lead in AI is actually lost, and it becomes harder to do AI in the U.S. It’s similar to what happened with crypto, where crypto had to decentralize outside the U.S. in the 2020–2024 range. I think that’s the intent, but I can see those storm clouds coming.

The one other thing I’d say is that because AI does middle-to-middle, not end-to-end—it doesn’t do everything, but it does a lot—there are a lot of bureaucratic jobs, jobs that lawyers do, doctors do, teachers do, professors do, artists do, and journalists do. This is going after the blue base, really going after them.

Doing all of this AI in San Francisco, publicly making millions or even billions of dollars, being demographically different with all these immigrants, and being very publicly rich and recognizable in the blue city in the blue state in the union is not, to me, a good long-term recipe for peace and prosperity. It results in accumulating too much capital too publicly, and then you start to see some very, very nasty things happening.

So, because of all that, I think I am bullish on decentralized AI.

I think this is a good way to close. I’m going to do the impossible thing with you, which is try to get the last word in and let Erik just say thank you very much.

Go, go, go, go, go.

We opened up a lot of topics, and I would encourage comments and dialogue out on X about where we should take the next part of this, because we should keep going. But I want to say, broadly and deeply, that I agree: the biggest issue we all face right now in the technology sector of the economy is the risk to the AI innovation trajectory in the U.S.

You could look at that from a technology perspective, a regulatory perspective, a business-practices perspective, an immigration perspective, or a research-funding perspective—any way you want to look at it. There are arrows aimed at it from various perspectives, preventing it, when the right answer is that we need to let the market work. The market for talent, the market for technology, the market for people—there’s a very strong market that can really work.

From Clayton Christensen’s perspective, what China is trying to do is commoditize our strength.

The release of a bunch of pure open-source, open-weight models coming from China is specifically designed to go after a rigid or complacent American view of AI, which is cloud-hosted by a few big players, closed-source, and so on. China is just doing—and I can look at this very emotionally and personally—this is Google releasing Google Docs for free.

Yeah. Yeah. Exactly.

And I’m running Microsoft Office, and Google is just like—

“We’re never going to make money from this.” Here we are, 2025, and they still don’t make any money from it. What Microsoft had to end up relying on is the worst part of the business, which is enterprise distribution lock-in as the core part of your business—not innovation, not moving forward—which bums me out.

I mean, they make money from G Suite. The cost of G Suite is starting to get expensive. So it gets expensive, but relative to what you guys are making, it’s like—look, the profits from Office are still the profits of Microsoft, with Windows and stuff.

But the other angle is copyright. I’m going to come at it from a different angle, and we should maybe think about talking about this, because people are rightfully panicked about the U.S. position and point to one of those slings and arrows being copyright, which, of course, has no issue in China at all. They have no problem with copyright—ask the pharma industry. I lived in China. I worked on copyright. I know exactly what they’re doing.

But the truth is that copyright also created the technology industry in the world. It was Microsoft and Intel, with intellectual property and copyright, and Apple that enabled the industry. So we have to look at it a little more critically and not think just about the starving novelist in Brooklyn who is frustrated by being used as training data. There’s a lot more depth to the copyright issue.

And finally, I do think there’s a lot to wrap up on just the M&A side.

The recent wave of deals is going to get looked at with scrutiny, and the truth is that something will change in what’s permitted in deal structures in terms of oversight, because I think they’re too big to get ignored by regulators in a tech industry that they’re no longer just going to ignore.

But I also think there’s a lot of opportunity to have much more clarity in deal structure. Maybe it’s a great idea that Balaji raised, to have designated survivors as part of corporate governance. There are a lot of interesting things you can think of to make that kind of outcome something that’s thought about, because, of course, today people who take on money from very late-stage private-equity investors or corporate venture know the terms and conditions that you have to have in those deals to attract that money.

In the same way, if you know the kinds of things that might happen, you structure your cap table and your corporate governance to facilitate that or prevent it. Right now, it becomes part of the business practice, then it becomes formalized, and it’s less likely to be something that could just be stopped by an arbitrary ruling by an appellate court in the 8th district that doesn’t like a deal that happened to a local company.

I think that’s where we end up with the risk right now: it’ll just be arbitrary, and nothing is worse for anybody than arbitrary. But I feel like we had this very long arc of discussion that was super interesting in terms of M&A and where we’re heading. Look to where to pick it up.

If Lina Khan, a few years ago, when she was sort of empowered, was asking for advice or perspective, is your view, “Hey, let the markets work,” because M&A helps everybody, from big companies to small companies to the tech ecosystem to the consumer? Or how should we think about antitrust?

Well, I’ll go first, and then we should wrap up on Balaji for sure.

The truth is, M&A will almost certainly fail unless they want to come out on the regulatory side by defending against the potential for failure. They really can’t come out on the side of saying, “We predict that this one will be successful,” because that just isn’t a statistically supportable public-policy approach to the action. The markets are much better; otherwise, they’re basically instituting rent control on investing, which is definitely not going to be the right way.

Yeah. There was just lightning and thunder for a second. Can you hear that over there?

Yeah.

Oh, is it lightning? You said—

Lightning. Okay, okay, fine. Just lightning.

Not a tsunami, not an earthquake. We’re good, right?

So, yeah. So Erik, to your question, I would say we have to actually think more deeply in the following sense: if you model the public sector as a platform and the private sector as the apps on that platform, sometimes an app gets big enough that you just have to actually build a platform or become the platform. Google was search, and then it grew and grew and grew and actually had to build its own platform. Essentially, Google, with Chrome, kind of became its own thing, as Steve is aware, and built things.

And so what we have to do is stop being reactive to Lina Khan or Scott Wiener on the AI bill or things like that. We have to be proactive in the following way. A: for every space that we’re in, we figure out what the ideal set of laws is. B: we write model legislation for all 50 states and all 190 sovereign countries. AI can help with this, but obviously it’ll just give you a first draft. It’ll get you on base.

C: we build a sales team that goes down and knocks on the doors of those 50 states and 190 countries. And, of course, there are subdivisions of cities and counties and all kinds of stuff, both within and outside the U.S.

Next, we actually find politicians. Before you go and knock on the doors, you can rank that list by those who are the most pro-tech, the most amenable to tech. For example, Jared Polis in Colorado is friendly to accepting Bitcoin for payments there, or you have somebody who’s posted about AI and clearly they’re conversing with it.

Often you’ll find some state senator or governor. Obviously, Nayib Bukele, before he became who he is today, was a very pro-tech person in government in El Salvador. So we identify all the pro-tech politicians around the world, and in particular in this process, small states are the friends of little tech because they’re the ones who don’t take anything for granted. They want to build their economy and so on and so forth.

And so we go to them and say, “Here’s a draft of a bill, and then here are 10 CEOs or 10 founders or 10 investors, or whatever—50 representing X billion in AUM or Y billion in revenue, or some combined thing. And if you pass this legislation, then we will invest in your country,” because that is now unlocked. Now we can build at the speed of physics, not permits.

I should write an article on this: “Elon Salvador.” Okay, “Elon Salvador” is what it sounds like, which is the tie-up where, in an American time zone, Elon gets some space where he can build at the speed of physics, not permits. All 20th-century barriers go away.

You keep the common-sense stuff, like “bash, not kill,” assault, murder, blah blah, whatever. You don’t have to sunset every law, obviously. There are some laws that are just eternal laws, but lots of 20th-century regulations are just very stupid. As Larry Page said, after the internet, you have to go back and look at a lot of laws and see if they still make sense. Permanent laws—these laws—do they still make sense when you can build in different ways with robots or other things?

So the answer is, I don’t think it would be a micro answer, Erik. It wouldn’t just be about advising economists. It’s a macro answer: go between countries, essentially. You know what it is? Rather, here’s a flip.

Rather than say, “How do we let them decide whether we’re a monopoly or not?” assume the U.S. government was a monopoly—the federal government—and ask: How do we build competition to that? How do we build jurisdictional competition? How do we build choice? Because 96% of the world is non-American, and only 50% is non-blue even within the U.S. So you’ve got lots of jurisdictional choice. So how do we do antitrust on that?

Maybe we’ll wrap on that big idea. Apologies, Steven, but this has been a fantastic conversation. Thanks so much.

Thank you.