The Stablecoin Future, Milei's Memecoin, DOGE for the DoD, Grok 3, Why Stripe Stays Private

I am your host, Jason Calacanis. With me again are a couple of my besties: David Friedberg, you know him as our Sultan of Science. How are you doing?

I’m keeping busy, thank you. Keeping busy.

On Valentine’s Day, Chamath and I had a little trio; we were on the MK Ultra podcast, and it hit number 4. The All-In podcast, of course, was number 1. Reflections on our Megan Kelly, our triumphant Megan Kelly Valentine’s spectacular.

I was fine. It was good.

Okay, wow. Thanks. You’re such a great performer here, giving me so much to work with, Chamath, as always. It was a great pod. Shout-out to our friend and friend of the pod, Megyn Kelly.

We’ve also got an incredible duo. For the first time, we’ve invited a duo to join us in David Sacks’s seat, who is busy saving the country. The Collison brothers are with us.

You guys want to hear a great lost-fortune story? John has one for you. The last time we met was 18 years ago, when we were working on our prior startup, Auctomatic, with Hadi and Koo Taggar. You were 17, 18, 19—one of those ages. It was one of these San Francisco setups where it was a 2-bedroom apartment, with a few of us living there. I think maybe 6 people were working out of there.

A normal number.

Exactly. A normal number is to load up a 2-bedroom apartment with 6 people. Chamath, you came and visited. This is what’s so brutal about this: I could have invested $1, a single dollar, and I would have made $1 billion.

I remember meeting these guys. I was with Alan Morgan, who was my boss at the time. I was a junior principal at Mayfield, shout-out. I think we tried to invest in the business. I don’t know if you remember, Patrick and John, but I think we tried to invest in it, or it didn’t happen. Then you ended up shutting it down, but right away you spun back up and started Stripe.

I just watched from the sidelines the whole way. First of all, it’s an amazing place for Silicon Valley, where you can see these people just keep pushing the boundaries up and up and up. Number 2, the thing that is such a learning for me is: Why didn’t I just pick up the phone and call them at any point in the last 17 years? What am I thinking? It’s so brutal.

Let your winners ride. Instead, we open-sourced it to the fans, and they’ve just gone crazy with it. Oh, my God. So brutal.

First, you probably don’t remember this, but I remember that meeting. We offered you something to drink. We did not have a broad selection. I think we had water or milk in the fridge, and you asked for a glass of water.

I went over to the sink and realized that we hadn’t really been on top of the washing up, so I had to gingerly wash a glass for you to get your glass of water. I can’t remember if you touched it over the course of that meeting.

John, tell us what that meeting was like. To take you back to the moment, here’s a picture. No, stop. Here’s a picture of pre-9-figure Chamath. Oh, my God. This is when he shopped at Macy’s. Does that jog any memories, John? That guy walked in with his khakis and that light pink Brooks Brothers shirt. What did you think?

I don’t know. You can go back and find historical photos of anyone and use them to make fun of them. If that’s the worst historical photo you have, that’s pretty lightweight stuff.

Exactly. Do you want to tell everyone in the audience what Stripe is?

Stripe processes payments. It’s a 10-plus-year-old startup that, basically, if you’re a startup company and you want to do transactions, you use Stripe. For example, the All-In startup uses Stripe to pay for the tickets, and then we give these guys, for some reason, a half million dollars every year. No discount, they don’t sponsor the event, and they’re making a fortune. They’ve got 10,000 employees, and the company changed the world. We’ve never been offered to sponsor the event. I didn’t know this was an option, but he hit us up for a half mil last year. Maybe this year we can hit you up and negotiate it live.

That’s broadly accurate. I would just fact-check that it’s nowadays not just startups, even though startups run on Stripe. The world’s largest enterprises—Hertz, Amazon, Ford, all these kinds of companies—use it.

When we started out with Stripe, we thought it would only be for startups. We thought those were the people who needed a problem solved, and we thought payments were broken for them. As time went on, we found out it was kind of broken for everyone.

Is it public how much volume you process a year? Do you talk about that?

It’s more than $1 trillion a year.

$1 trillion a year is processed through your network?

Yes. Global GDP is around $100 trillion a year, so it works out to around 1% of global GDP. You could say that GDP is final goods and Stripe processes more than only final goods, so it’s not exactly the right or fair comparison. But Stripe mostly is used to sell final goods, so I think it’s reasonable.

The other thing I’d say is that people reasonably think of Stripe as a payments company, because that’s certainly what we started out doing and it’s still the largest line of our business. But what we realized a couple of years in is that the structural, secular thing happening is that every kind of money movement is going from being manually orchestrated to being orchestrated by software. There’s some program somewhere making the thing happen.

Because of that, and because of what we hear from customers and the pull there, we’re now helping with lending, card issuance, treasury and money storage, cross-border money movement, and stablecoins.

We’ve got to talk about stablecoins. Why did you do a stablecoin?

Stablecoins are finally happening, and they’re really useful. We followed crypto for a long time. The Bitcoin white paper dropped in 2008, the year before we started working on Stripe, so it’s been funny watching Stripe and crypto grow up together.

We tried to make Bitcoin happen as a payment method on Stripe, but it just wasn’t that good as a payment method. It’s good as a store of value, as kind of a gold substitute, but transactions are slow and expensive. You never know exactly how much you’re going to get because it isn’t denominated in dollars.

Stablecoins are now really good. If you look at something on an Ethereum layer 2 or Solana, the technology is good enough. We bought a company called Bridge late last year, which is building the Stripe of stablecoins.

People like SpaceX are using them for treasury management. People are using them to offer U.S. dollar services to people all around the world. Stablecoins are, I think, the first really big payments use case, and I think it’s finally coming because the technology is good enough.

Is there a moment where you say that Visa and Mastercard can get challenged? Is there a set of boundary conditions that you’ve written down where, when you check a few of these boxes, you know it’s time for those companies to get dismantled?

The behavior we’re seeing right now is that stablecoins are most interesting and seeing the most adoption where there’s some cross-border component. You need to manage corporate treasury around the world, you want to send remittances to people in other countries, or people in other countries want to hold dollar balances.

What we’ve always seen is that, in the U.S., things work pretty well. In Europe, things work pretty well. We even see this pre-crypto, where the way people pay for stuff has been radically changing. UPI in India and Pix in Brazil are designed by central banks and are really good, government-run Venmo solutions. Those have all happened in emerging markets broadly, and not in the U.S. and Europe.

We certainly keep our eyes peeled for that changing at some point, but right now a lot of the interesting stuff we see is happening internationally.

With respect to Visa and Mastercard, an important thing to keep in mind is that most of the interchange fees charged to merchants—and you mentioned what we charge the All-In podcast—the vast majority of that flows right back to the issuing banks in the form of interchange. Almost all of that flows right back to consumers in the form of the lending that the cards themselves represent, but also in card rewards.

Card programs are not actually big profit pools for most of the major banks. Any substitute for Visa and Mastercard is a question of whether consumer rewards are going to go down, whether consumer protections are going to go down, whether we’ll be extending less consumer credit, and whether other points in that space are viable. It’s a set of trade-offs. It’s not as simple as this enormous rent extraction happening.

Patrick’s totally right. The interesting use of stablecoins is cross-border and outside the U.S. The big use case taking off right now is consumers in other countries seeking to hold dollars.

Here in the U.S., we obviously benefit from being able to do that. The vast majority of people in the world have a worse currency, in the sense that it’s less stable and more inflationary. Storing savings is much less favorable.

If you look at the naira, for example, there are a lot of people in Nigeria, and the currency there has devalued by a factor of 3 or 4 over the last couple of years. The use case of consumers being able to store dollars is really exploding.

We think about this as an analogy to the eurodollar system. In the 1970s and ’80s, the eurodollar system was a way for companies to store dollars and have something more stable and reliable. But it was only accessible if you had a very high minimum transaction size. I think it was around $1 million.

With stablecoins, you can now be a consumer in Ecuador and have a $1 U.S. balance. That was not a product accessible to you before. It’s a really big deal for people in those countries, and in some sense also for the U.S., because the dollar’s status as the world’s reserve currency is becoming much more deeply established.

That is the huge win for allowing stablecoins and making them legal, giving them rails. Putting aside Tether and all the bans and the fugazi stuff they’ve been doing, and all the lawsuits they’ve lost and the bans in different countries, having USDC and other ones in the United States means we can regulate them and they have to buy Treasuries. Dollar supremacy continues, and that’s fantastic.

Right now, All-In could accept payment in stablecoins with Stripe, correct? We just check a button and we get stablecoins?

Yes.

If we had $1 million sitting in our Stripe account and had to pay a venue or other vendors, and we were sitting there in your coins—what are they called? Bridge coins?

Bridge is the company. It’s the platform. Bridge primarily is a set of software APIs.

But you’ll obviously have a Stripe stablecoin at some point?

Bridge already has a small stablecoin, but we don’t need to get into the details. Bridge is primarily a set of software APIs.

The point is, if you turn on stablecoin acceptance with Stripe today, that’ll use USDC. Could we then pay people from our Stripe account? Could you lower our fees if they were also doing stablecoins? Does that exist today, or is that coming next year?

You could pay people in stablecoins, but to the point about where you’ll see adoption first, paying people via bank transfer in the U.S. isn’t great. It’s kind of slow, but it’s fine. It’s not the biggest problem today.

Exactly. Whereas the people who are using Bridge—Scale AI, for example—have to pay contractors all around the world. When you want to get money to people in the Philippines, that starts to get really annoying and expensive.

From our point of view, the real hair-on-fire problem is the international stuff. Domestic payments will come later.

I think you’re answering narrowly with respect to stablecoins, and everything you just said is right. But my intuition is that it’s really inefficient and annoying to engage in B2B transactions, get invoices paid, and deal with the whole system.

If you look at most companies, they’re losing 1%, 2%, or 3% of revenue to accounts payable and accounts receivable. Some of that might be because of the transaction rails themselves, but a lot of it is because of bureaucratic, inefficient processes. You have humans sending invoices, humans reconciling them, and people trying to line up transfers in a bank-account statement and figure out what corresponds to what.

Stablecoins would be part of the solution, but there’s more to it. Separately, we’re trying to solve that with a product called Stripe Billing, which we announced last week has passed $500 million in ARR.

We could send an invoice to somebody, which is exactly what FreshBooks and the other products in the market do.

All the back office is there. Is there a version of a network effect inside Stripe for its customers? If I allowed you to integrate directly into my general ledger somehow, and you gave me some kind of phantom bank account, why isn’t it just a ledger entry if I’m making a payment from me to somebody else who’s also on Stripe?

The things we really want to solve are all the calculation, identity verification, and risk. Those are the things that are actually expensive. If you look at this flow, that’s where companies lose their money today.

Having said that, you’re right. The fraction of money movement on Stripe where the 2 counterparties are both part of the Stripe network is obviously growing. That will be another way we can reduce fees over time.

Again, I actually think the biggest part of that will be because we reduce fraud. Both counterparties are known. I talked to a payroll company recently, and they were describing how big a deal it is that people sign up and defraud companies. They can lose millions of dollars in a single attack.

Having some kind of trusted node rather than just routing an account number would be a really big deal for them.

You have a very good pulse. As a subset of the economy, you probably reflect a large part of the global economy. Have you ever considered regularly publishing some sort of economic sentiment?

One of the big things we’ve talked about is how many backward revisions there are to everything from nonfarm payrolls to GDP. They’ve become so unreliable that it’s very difficult for people transacting in the market to know what to do. Have you ever thought about that? I’m sure you have a much more accurate sense of where the economy is than many other people.

We have, and I feel a bit rueful that you’re asking that question because, on some level, we should have done it.

What makes it tricky is 2 things. First, Stripe is not a full cross-section of the economy. We’re more biased toward online businesses and innovative companies, so you have to somehow net that out. During COVID, the online economy was doing great, while the offline economy was a different story. The interpretation can be tricky.

Second, the Stripe business is growing so quickly and changing so fast that it’s not necessarily representative of the economy. Even if Stripe is way up year over year, you have to be hesitant about drawing conclusions from that.

Having said that, in principle, you could draw some conclusions. One thing we looked at over the last couple of years was inflation data, and the team constructed a pretty reliable leading indicator for inflation. We would like to share that openly, because I think it’s a public good for there to be better and more reliable economic data.

If you were to build the financial system from scratch today, we’ve got SWIFT, banks that store assets, credit cards and credit-card networks, and transaction-service providers that sit on top of this. What’s the right solution if we were to build a financial system for the world from scratch today?

Can you see a world where we bridge away from the credit-card networks and move out of some of these legacy systems, or are they so deeply ingrained that we’ll continue to build complicated solutions into and around the legacy financial infrastructure?

I’ll give my view, and then I’m curious what Patrick thinks. First, there’s just general technology scalability. The finance industry has its version of the [unclear] for sure, where everything should be highly scalable in real time.

In a way, stablecoins are solving something you don’t technically need full decentralization to do, but the ability to make real-time payments any hour of the day or night is a useful property. Some private systems have also built that.

A big one for us is trust, and the fact that the fraud problem hasn’t really been solved in online payments. A big reason people come to Stripe is that we are a reputation network across the internet economy.

When someone comes and buys something from a Stripe user, 93% of the time we have seen that card before. The merchant can know something and know that they can trust the end user.

It’s gotten to the stage where, if someone comes along and buys with a credit card, and they’re signing up with an email address or phone number that we haven’t seen before, that is ipso facto suspicious. They may be trying to use a stolen credit card or something like that.

A big part of what Stripe ends up doing is acting as a reputation network to keep fraud out of the system.

In fairness, Jason told me I could use that credit card anytime I wanted. I don’t think he remembered, but I think you need to turn my account back on.

Friedberg, I just got news from our CEO: Mastercard canceled its sponsorship of All-In Summit. This is costing us a fortune.

Stablecoins are going to be a big part of the solution. I don’t think they’re going to supplant all the consumer-facing networks. We’ll see consumer-facing networks built upon and substantially leveraging these systems, but stablecoins will probably be the common rail.

Most businesses lose more money to fraud than they do to the pure transaction costs themselves.

You’re hearing us talk a lot about fraud because it’s a huge economic cost for these businesses today. There are indirect costs, too, where you make the consumer experience more hostile because you have to protect against possible fraud. Why do you have to type in all this information or lock out your bank account?

We can see in the data that these problems are getting worse and harder because of machine learning, AI, and globalization. Various fraud metrics across the industry and ecosystem are way up over the last couple of years. Stripe’s fraud rate is actually down by 80%, but it’s becoming a really acute issue.

All right, we’ll get into staying private longer and when you’re going to pull the IPO trigger later in the show. But we have to get through this docket. We have so many great topics to talk about.

Jamie Dimon went on a rant about remote work and Zoom in a town hall. Here’s a snippet:

“A lot of you were on Zoom, and you were doing the following: looking at your mail, sending texts to each other, and, when asked the other person a question, not paying attention and not reading the stuff. If you don’t think that slows down efficiency and creativity and creates rudeness and stuff, it does.

“When I found out that people were doing that, you don’t do it at my goddamn meetings. You go to a meeting with me, you’ve got my attention and my focus. I don’t bring my goddamn phone. I’m not sending texts to people. It simply doesn’t work.

“The young generation is being damaged by this. They may or may not be on your particular staff, but they are being left behind socially—ideas, meeting people. My guess is most of you live in communities a hell of a lot less diverse than this. That’s not how you run a great company. We didn’t build this great company by doing the same silly things that everybody else does.”

Collison brothers, tell us about how you run Stripe. Are you remote? Does this resonate with you?

I love listening to Jamie Dimon rants. It’s like business ASMR.

Business ASMR itself seems like it would be a great podcast.

I was about to say, “I’m subscribing.” That’s an instant $10-a-month subscription.

People just said a lot of things during the pandemic. Do you remember? Handshakes were going to be over, business travel was going to be over, and every company was going to be fully remote.

Stripe broadly is in a pretty similar spot to where it was beforehand. Most people go into an office. Most people are part of our San Francisco, New York, Dublin, or Singapore offices, and then we have a bunch of people who work remotely.

I think Jamie is right on some points. Working remotely has also had a bunch of benefits, because there’s a much larger talent pool available to companies like Stripe.

You see the two-body problem, where maybe one partner is assigned to a hospital in Idaho. They don’t get to choose which hospital they’re assigned to, and the other person gets to work a high-paying technology job. Remote work solves that.

I think one theory for declining dynamism in the U.S. and declining TFP is that allocative efficiency declined as women entered the workforce. Now you have the two-body problem, where both people have to make coordinated switches. Remote work solves it.

You’re running a company now. You’re the CEO of Ohalo. Does this resonate with you, especially his point about younger people, being rude, being focused, being in the media? Maybe there are too many meetings where people are partially paying attention. Maybe there should be half as many meetings, and people should be paying attention. What do you think?

There’s always room for optimization there. We deal with this, too: too many meetings and too many people.

What was most striking for me about Jamie Dimon’s rant, and the resonance it seems to be having particularly in Silicon Valley and among people in leadership positions or on boards, is that this is another example of a different tenor for business leaders right now.

Leaders are starting to step up, speak their minds, speak more directly, and lead from the front rather than lead from the back. The last couple of years—and the transition away from what many people characterize as wokeism and coddled employee workforces—were a period when employees made the decisions, and leaders said, “Okay,” to their employees’ whims and needs.

Look at what has happened with Zuck. He said, “You’re with me or you’re against me. Here’s a buyout option.” Elon was obviously an exemplar of this at Twitter. We’ve now seen this with Coinbase and Brian’s letter, and I think it has become more of a standard in the post-COVID era.

Leaders can lead from the front, speak directly, and say, “This is the way things are going to be. My job is not to coddle my employees. My job is to lead my employees so that our organization, our team, wins and we achieve our mission.” The objective is not to create a family workplace where everyone is happy all the time. It’s to help the organization succeed.

I’ve heard from people individually and seen this tenor shift underway. I think Jamie Dimon is another exemplar of it, and it seems to have some resonance.

Chamath, I want you to respond specifically to this next clip. Let’s play the second clip, about organizational bloat.

“Every area should be looking to be 10% more efficient. If I was ready to part with 100 people, I guarantee you, if I wanted to, I could run it with 90 and be more efficient. I guarantee you I could do it in my sleep.

“The notion that these bureaucracies need more people and that they can’t get it done—no, because you’re funding requests that don’t need to be done. Your people are going to meetings they don’t need to go to.

“Someone told me, to approve some wealth-management thing, that they had to go to 14 committees. I am dying to get the name of the 14 committees. I feel like firing 14 chairmen of committees. I can’t stand it anymore.”

Chamath, the bloated bureaucracy at big companies. Your thoughts?

There’s an adage that says something akin to, “50% of advertising is useless; we just don’t know which 50%.” I think it’s probably true for most corporate structures in general.

A lot of organizational bloat has evolved because of the way people have responded to technology. If you look back 50 years ago at that famous picture of Microsoft’s early team, they didn’t rely on software in the same way. There wasn’t Salesforce, Workday, and all this infrastructure.

Instead, they probably organized around what they were good at and tried to do things efficiently. In the absence of technology, many companies found a way to be very efficient.

That started to change when you had rigid demarcations of where one job ended and another job started. Part of why that happened is that all this software convinced people it would create efficiency. In return, the chief marketing officer’s job is this, and the chief sales officer’s job is that. This is how the roles are defined, and this is how people do it.

I think things have become bureaucratic and bloated because there’s a propensity to run toward software because you think it’s a solution. At best, it’s a symptomatic aid. It doesn’t address the root cause. In fact, it promotes bureaucracy and bloat.

If you look at JPMorgan’s P&L, it spends $6 billion a year on IT. I suspect that if you streamlined that, you’d have half as many people because they’d be doing the job in a wholly different way.

The counterfactual is that companies like Facebook, Google, Tesla, SpaceX, and I’m sure Stripe design a lot of things internally and custom-build them for their organizations. You see that in revenue per employee and other efficiency metrics.

I think Jamie is a victim of this push to productivity. He would look like a Luddite if he didn’t adopt technology, but by adopting off-the-shelf software, he introduces organizational bloat because those systems are demarcated very rigidly. You have the marketing team using HubSpot, and then you have the sales team using something else.

The other thing I want to say on the first topic is that, other than engineers—who are naive but can be extremely productive from day one—there are very few job types where naivety is an asset.

Most people early in their careers are in a J-curve where they are negatively contributing and slowing everybody down. The goal is to invest in these people so that they come out of the J-curve.

There are probably other jobs that are like engineering, but many are not. It’s important to get the mentoring you get by being in an office. In the absence of that, these young people, as Jamie said, are totally lost.

John, Toby from Shopify did this zero-based-budgeting concept for meetings. He purged all meetings at the beginning of the year. He just deleted everybody’s meetings from the top down. I’m curious how you think about bloat, meetings, and committees. Do you worry about that at Stripe?

We know Toby very well. I always feel like we should take some of his ideas. We haven’t done the meeting-deletion one. You might say the meetings get recreated, but he measured it and they didn’t.

It sounds like he wrote a script to delete all the meetings from the Google Calendar instance. I enjoy Toby’s perspective that many organizational problems are software problems and that you need to write a script to solve them.

There’s a kind of purity to that, but you can also be over-intellectualizing your problems. I do agree with Chamath on the remote-work point. It’s dangerous when CEOs think about this stuff, because there are unfair anecdotes that feel unfair and get people really riled up: quiet quitters, the anti-work subreddit, and all the talk of people working 2 jobs.

That generates a lot of energy with corporate leaders, but you don’t want to design your policies around the bottom 5% of the company. That would be a horrible mistake. You want to design your policies around the top talent.

We have some outrageously productive remote people. They’re often in a cabin in Idaho somewhere, coding up a storm.

The thing we saw, interestingly, is that we measured this before COVID because we were doing a lot of remote hiring and wanted to see how much we should lean into it. It isn’t good for early-career people. We could measure it in our productivity data before the whole discussion about remote work happened during COVID.

It’s bad from a work point of view, and it’s also bad from a personal point of view. They go mad because they’re 23 years old and in solitary confinement.

And, by the way, breaking news here: Jamie Dimon now knows which 1,739 employees to lay off first.

There’s a Coworker.org petition to get Jamie to retract his statement. The petition has been created.

If I know Jamie, I know he’ll be retracting that statement right away.

Absolutely. He’ll bend to the pressure of those 1,700 mids.

The median employee at Stripe is awesome. The median employee at Stripe is not the median person in the population at large.

I was using the term “mids.” Mids are people who are just average, not above average. How do you deal with low performance?

You need to have an aggressive performance-management culture and stay on top of it. It isn’t good for anyone to keep those people around, because nobody likes feeling that they aren’t succeeding.

If their careers aren’t advancing, they aren’t getting positive feedback from their manager or peers, and they aren’t shipping things, that’s just not a good equilibrium for anyone. We try to stay on top of that and track it closely.

The thing to say about this discussion broadly is that people readily fall into a normative, moralizing perspective. People should be in the office; people shouldn’t be in the office. There’s a lot of “should” here.

It’s helpful to be empirical and objective and look at what the data says. It’s also important to recognize that there’s a lot of heterogeneity. People have different preferences and different abilities to work effectively when they’re by themselves. Some do, and some don’t.

Organizations are doing different kinds of work. Nvidia, last I checked, is doing pretty well, and Jensen is on the record saying he doesn’t care where you work. Coinbase and Shopify are remote-first companies. I was recently chatting with the folks at Jane Street, and they really believe that being colocated and able to share ideas on the trading floor is important.

I don’t think these pictures or worldviews are necessarily contradictory. They probably hire different kinds of people and are in different kinds of businesses. I’m skeptical of flat “shoulds” in this space.

There are many paths to heaven. Also, keep in mind that labor productivity in the U.S. is up around 20% in the last 10 years. The median person in the economy—or the average person—is producing 20% more, on an inflation-adjusted basis, than they were 10 years ago.

That’s going to keep ramping up with AI and all these amazing tools that are coming out.

We’ll leave that on the side for now because that would be an hour-long rabbit hole. We could jump down it, but we have to get back into DOGE.

I’ve heard a couple of criticisms of DOGE. One is that it’s one-sided: We’re only hearing about people on the left doing grifts and USAID. The other is, “You’re pointing at little tiny things like USAID. When are you going to get to defense spending and Social Security?”

Well, here we are. The Washington Post is reporting that, in between doing sets of 47 push-ups, Defense Secretary Pete Hegseth asked senior leadership at the Pentagon to develop a plan to cut 8% from the defense budget each of the next 5 years. That’s a compounding 8% a year.

We’re talking about close to $300 billion in savings over 5 years if they hit it, which isn’t a crazy target. It’s just crazy in our country, where we haven’t even been able to have the Defense Department pass a basic audit, if you’ve seen those reports.

Let’s pause there and talk about military spending. Chamath, I think military spending needs to sit downstream from technology. If it doesn’t, you’re misappropriating the money.

We’re inventing incredible capabilities in AI and autonomy. You need to take those things first and figure out how to productize them, because that builds the kind of modern war machine we need.

I tweeted about this. Nick, maybe you can find it. The CBO red-flagged a project where the Navy was about to appropriate $1.2 trillion to build frigates.

There’s a body of military planning that says this is a projection of power, so you need to spend this kind of money because people want to see the big boats and the big iron in the water. Maybe there’s something to that, but you can’t be spending $3 billion or $4 billion per boat and taking 8, 9, or 10 years to build these things. It isn’t sustainable.

Part of why they do that is because it isn’t coupled to what’s actually happening with innovation. There are core pockets of companies doing this. Saronic just announced a $600 million raise today. Saildrone announced hundreds of millions of dollars of contracts with the Navy, and Anduril is doing that with the Army.

Military spending needs to happen downstream from what’s happening in technology. Broadly speaking, we don’t have that. What we have instead are systems integrators with extremely deep connectivity that are able to contract well, but not necessarily invent well.

If you take defense down to first principles, there was an excellent tweet we were all texting about yesterday. It made the observation that Trump’s negotiations with Russia and China—where there’s all this hemming and hawing about whether those negotiations are complying with the wants and needs of dictators—may actually be a shift in the strategy governing the United States’ global relationships with other powers.

In particular, it could be a shift from the objective being U.S. primacy, with the U.S. as the sole great power on Earth, to recognizing that this is no longer the case.

In a multipolar world, we may no longer need to invest in wars, conflicts, and defense with supposed allies in order to build up our strength across the globe. I’m not saying that this is necessarily the right strategy, but the observation was that maybe the strategic imperative is now to have a multipolar stance in the world rather than a stance of primacy.

If we settle into a new world where China, Russia, and the United States are not necessarily equal powers but are shared powers across the globe, do we need to invest as much in global defense? Do we need to continue pouring dollars into building up arsenals, military bases, troops, stations, and positions all around the world?

Perhaps not. Perhaps the world gets divided peacefully, we open up global trade relationships, everyone benefits economically from advances in technology and improvements in productivity, and the world order is peaceful but multipolar.

Maybe that’s the new era we’re entering.

To Chamath’s point, there’s different technology now in play. We’ve seen it in the Russia-Ukraine context: A $10,000 drone can destroy a $10 million piece of equipment. China now has drone factories that can output millions of drones each month.

If China develops this new type of arsenal, with millions of autonomous flying systems that can attack troops and expensive equipment, do we really need aircraft carriers? Do we really need tanks?

I think that’s the whole Hegseth-led, Trump-led conversation underway in defense right now. Number 1: multipolarity. Number 2: Therefore, we don’t need as much defense spending. Number 3: The defense spending we do have should account for the new technology in play on the battlefield.

That really changes the character of how the Defense Department is structured and how funding is structured. That’s the way to look at it, rather than saying, “Let’s just cut defense spending for the sake of cutting it.” That might be what’s going on right now.

Obviously, what Anduril and others are doing is amazing, but we’re not defense experts. I’ll bring the credit-card merchant perspective to bear here.

We naturally look at the time series and the data around it. I’m struck by the fact that, as far as I can tell—and I may have some of the details wrong—the proposed cuts over the next couple of years are approximately the same magnitude as the reduction in the Defense Department budget that occurred between 2010 and today.

It’s not like this is some unprecedented transformation in the Department of Defense budget. We’ve done this before.

Second, as far as I can tell, one of the most universally shared bipartisan issues in Washington is the inefficiency and profligacy of defense procurement. James Fallows was writing a book about this in the late 1980s. You had Augustine’s Laws and an entire book about this.

Everyone seems to fervently believe that defense procurement is monstrously inefficient. It’s possible to make budgetary changes without fixing that, but the prospect of meaningful improvement seems really beneficial.

If I can give a quick book recommendation, this book, Boyd, by Robert Coram, is about John Boyd, the Air Force colonel who was part of the reformist movement.

I feel like everyone in Silicon Valley has that book on their shelf, and no one has actually read it.

It is a great book. It’s about Air Force procurement, essentially. The Air Force generals of the time wanted planes that were bad, and Boyd had a theory about better fighter jets. He had his fingerprints all over the F-16, the A-10, and the F-15.

It was a real battle to get the Air Force to produce better aircraft. The generals really wanted the bad aircraft that they had planned. It’s a fun read at this moment in time, when it feels like we have this similar transition from man to machine.

Sprinkling some OODA loops into your remarks always helps.

Sounds smart.

Chamath, you added a crypto update. Crypto Corner is back.

We had an exciting week of innovation in the crypto space last week. Argentine President Javier Milei, who is a hero to a lot of people on the right and to people who support government efficiency, promoted a memecoin. It was called $LIBRA.

He originally tweeted, “This private project will be dedicated to encouraging the growth of the Argentine economy,” with a link to LIBRA for his citizens to buy. Buy it they did, but he deleted that tweet when the whole thing came apart.

He said, “I was not aware of the details of the project, and after having become aware of it, I decided not to continue spreading it.” The market cap reached $4 billion and then crashed 95%, as these memecoins always do.

About 74,000 traders lost almost $300 million. Twenty-four wallets had losses of more than $1 million. Milei has been sued more than 100 times already, and this just happened last week. He’s being investigated by his own government, and an impeachment attempt is underway by the opposition.

Milei’s team told CNN that his endorsement of the coin was a mistake.

Really? That’s going out on a limb.

According to insiders, Milei never actually owned any LIBRA and was not associated with the coin. I think family members may have put him up to it. The details of why he promoted it remain unclear.

Chamath, your thoughts? It’s crazy. He was on such a positive upswing of momentum. It doesn’t make much sense why he got embroiled in all of this.

The problem is that the cover-up is always worse than the crime itself. The first message was very Clinton-esque: “I did not have sexual relations with that woman.” He was saying, “I did not endorse it. I just shared it.” That was his justification for how he could rationalize what he did.

The kid behind this thing, Hayden Davis, was on Coffeezilla. It was an incredible 1-hour interview. Did you see the Coffeezilla interview?

I saw some of the clips on X, and it was pretty brazen. He essentially said that he had Javier Milei in his pocket. There were text messages using some pretty colorful language to say the same thing.

There were also text messages that seemed to implicate Milei’s sister as having received some of the money. The whole thing makes absolutely no sense. Milei was doing so much good, and now he’s going to go through this whole cycle of trying to wash his hands of it. I don’t know why he did this.

There was another interesting tidbit. David Portnoy, friend of the pod, supposedly got involved. He’s a gambler and loves gambling. He apparently put millions of dollars into it, and this guy gave him his money back. This guy also has something like $100 million sitting in a bank account somewhere.

I don’t like memecoins. I don’t think they’re okay or productive. A bunch of people are going to put money in and lose money, and a few people are going to make a lot of money.

At the end of the day, it’s no different from people who sell trading cards or create and sell collectibles. It’s effectively a digital collectibles business.

Unfortunately, it’s amplified by 1,000 times because collectibles businesses have friction. They’re manual, you have to ship them, and so on. This creates a digital frenzy where you see the social-feedback loop happen quickly in real time.

That drives these things to a high value, which means people can lose a lot more than they otherwise could. These aren’t helping rebuild the financial system we talked about earlier. They aren’t creating productive value. They’re entertainment mechanisms, just like any other kind of gambling system might be.

People can choose to do that if they want, but personally, I’m not into it. I think it’s stupid, but whatever.

Patrick, do you think these are collectibles, or do you think the people buying them perceive them more like securities and more like Bitcoin? They trade with a ticker symbol, they’re traded on major platforms like Coinbase and Robinhood, and people share charts about them. You’re in the finance business. Memecoins: good or bad?

I’m basically with David. They seem to me to be analogous to gambling. I don’t know that we want to ban gambling. If you can do it responsibly and understand what you’re getting into, I guess that’s fine.

But judging by the tweets I see, there are a lot of ticker symbols, charts, and prognostications about future price trajectories that lead me to think people are placing more weight on the asset and security value of these than on some numinous intrinsic aesthetic value.

Maybe 2 things can be true here. People are gambling, and these are being presented as financial instruments. They’re trying to trick the suckers at the table. In this case, the suckers are the people who voted for Milei.

I learned recently that state lotteries are a relatively recent phenomenon. I think one state started doing it in the 1970s, and then a bunch of other states followed suit.

It’s odd when you step back. I pass a billboard on 101 for the state of California trying to get me to buy a lottery ticket. It’s a negative-EV bet, but it has become very normalized.

This is the first time I’ve seen the details of how this stuff happens. Hayden Davis laid it out. There are people called snipers who pump up the bids as soon as the coin is launched, and then they’re able to exit.

There’s an entire mechanism here that’s so shady. The specific thing within memecoins that’s probably most pernicious is the rugging dynamic. If you could have a memecoin without the pump and rug—just some memetic tracker of sentiment—maybe that would be okay.

But the particular way in which they seem to be employed is some discontinuous run-up, followed by the rug.

What do you think? I agree with you, Chamath. Milei had the greatest PR run of all time.

He became an inspiration to all of us here in America who were concerned about the deficit, out-of-control spending, ridiculous departments, and the committees Jamie Dimon was talking about. I don’t know if you remember, but Milei was the minister of culture and the minister of deregulation. This was the precursor to DOGE, where now we’re saying USAID deleted, Department of Education deleted, and the Defense Department minus 8%.

What I find terrible about this is what it means for leadership. What Milei did was rug-pull the people who put him in office. The people who voted for Milei are the ones who got hurt.

Leadership at its core is about putting the needs of your constituents ahead of your own interests. If you’re running Stripe, you have to think about all these shareholders and investors. Leadership is setting the example. You set the standard—the moral standard, the ethical standard, and the cultural standard.

Milei had set such a great standard that we all loved. The appearance of impropriety is impropriety in my mind. That’s the leadership standard that should apply here. Even being near this—whether it was his sister launching it or his brother launching it—is a problem.

Then he went on to taunt his own followers. He said, essentially, “I’m out on Milei” at that point. People make mistakes, and this was a stupid one to make, but the taunting of his own followers was even worse.

The reality is, if you go to the casino and lose money, what is the claim if you knew it had these characteristics? Leaders own their mistakes. They don’t attack the victims. You take ownership of it.

The way you should judge people, I think, is by what they do when they’re given a lot of power and what they do when they make mistakes. Milei is a failure on all of those fronts. It’s absolutely abhorrent.

That’s it. Thanks for coming to my TED Talk.

Do you need help getting off your moral grandstand now?

I do, actually. I’m over it. I’m sorry. I care about morals, ethics, and leadership. I think there’s a standard set by these people. That’s what I think about when I think about you.

With friends like these Collison brothers, can you imagine?

Please say their name. Pronounce the goddamn “I.”

I’m pronouncing it the Irish way.

We speed things up a little bit. We put them together. It’s a little bit different.

You wouldn’t know this from being from Sri Lanka, a great country.

You guys wouldn’t know why anyone watches this show, would you?

Yeah. Nobody says that. There’s no context for this.

Why do you make every show a train wreck and make us get it out of the banter?

The banter is why people come. So many TV shows are about how it’s nice to have friends. You look at Friends or How I Met Your Mother. My wife and I are rewatching The West Wing right now, and it’s basically a show about a group of buddies who are loyal to each other.

I think the underlying idea behind a lot of TV shows is that it’s nice to have friends. I think that’s the success of All-In.

Which season are you on?

We’re up to season 4 now. God, I’ve got to get to 5 or 7. Never got in on The West Wing. Of course, Sorkin left after season 4.

The great debate that America needs to have, and that I think is still the missing aspect of modern politics, is the great debate. Let’s talk about the topic at hand and discuss it on the merits of what’s right for the country, as opposed to everything being about attacking the other side.

If the other side brings an idea forward, we attack it and frame the idea as beneficial to them and hurtful to us. Nothing gets resolved because we don’t have objectivity around the major issues the country faces.

Many of these issues have valid points of view on both sides. We should be able to have the great debate and have conversations about DOGE, abortion, states’ rights, and spending, rather than use every moment as a way to attack the other side politically and make sure we have points and talking points for the next election cycle.

I miss that about The West Wing. It feels like a beautiful, pure way of thinking.

I wonder what it would be like to watch The West Wing and then House of Cards back to back. That’s something I should do. It would be a real juxtaposition of those 2 shows.

Isn’t The West Wing kind of the opposite of what you just said you want All-In to represent? I see The West Wing as being fully immersed in and representing one particular worldview.

We look back on the 1990s and the Clinton years as a period of great harmony in the country. The economy was doing well and things were good, but it wasn’t exactly a period of tremendous ideological debate, fervor, and schisms.

You’re thinking about the 1990s and the Clinton era?

Yes. Maybe I’m wrong. I wasn’t here in the 1990s, but from afar, it did not feel to me like a period of tremendous ideological debate.

Maybe they were the compromising party. Tell me another modern Democratic president who had a point of view on balancing the budget and creating a surplus that was aligned with the Reagan point of view at the time.

Clinton was a centrist, and he brought the parties closer together rather than further apart.

I agree with Chamath. The thing that makes The West Wing a great show is that it’s about the insider nature of the White House and the West Wing.

You see characters like Toby, who would never be a star in any other show, under any other circumstance, on any other network. Instead, he’s one of these central, quasi-good, quasi-nefarious bully characters. He was a precursor to the Rahm Emanuel archetype in the Obama White House.

I also find it funny how Dominic Cummings has talked about his experience of life in government. It’s so distracting when you’re trying to get anything done. You have a plan, you get up in the morning, and you’re going to go do something that matters for the country. Then you’re instantly, by 8 a.m.—

You know, sideswiped by some kind of silly controversy of the day. That’s basically many of the episodes of The West Wing, where they have some actual important thing that they want to get done, and then they just get hit by a silly controversy.

It seems to me like you’re also the product of the technological innovation that occurred during your presidency and during your term. If you think about Clinton, he got to ride the internet and this massive economic boom, and you look at Reagan and the PC boom. Sometimes the timing really matters.

I think, though—and again, I’m not any grand expert on the Clinton years—but I think it is interesting that one of the first acts of the Clinton presidency was the Deficit Reduction Act. Dave, to your point, when’s the last time that a Democratic president really, really cared about the deficit? I think federal spending fell by 5 points of GDP over the course of the Clinton presidency, which is really not a small amount. So obviously there were some kind of structural tailwinds from technology and the internet and all the—

Yeah, a bunch of that was defense. Nonetheless, he did it. In the last 2 administrations, you look at California: there were massive windows of surplus, and there were massive windows of a surging stock market over the last 8 years. We plundered and wasted them by adding $16 trillion to the debt during a good time. What’s going to happen during a bad time? Just absolutely brutal.

Let’s move on. Where do we want to go here? We’ve got Grok 3, we’ve got the China private sector, we’ve got a victory lap for Friedberg. I want to ask you guys questions about Arc Institute and the Evo model. We should do that.

Let’s do the Arc Institute. Friedberg, why don’t you ask the question? Patrick runs the Arc Institute, right?

Patrick Hsu

Yes, I’m one of the co-founders.

And there were scientists, and you guys are funders of it? Maybe you guys give us a lot of money into this?

Patrick Hsu

Yeah. Arc is a nonprofit that does basic biology research. It’s in Palo Alto, in partnership with Stanford. It’s about 230 people today. John and I are among the funders of it, but there are a bunch of other very generous donors.

Can you explain the idea of curiosity-driven research? That’s on the website.

Patrick Hsu

There are kind of 2 things behind this. The first is scientists. The vast majority of biology scientists today receive NIH grants doing basic research. The NIH grants are, 1, just hard to get and, 2, annoying to get. Scientists spend 40% of their time working on grant overhead and so forth. But worse, and even more perniciously, the grants are very restrictive in terms of the kind of science they can do.

We ran a survey of top scientists a couple of years ago, and 4 out of 5—79% of them—told us that if they could just spend money however they wanted, if they weren’t limited by what was prescribed by these NIH grants, 4 out of 5 told us they would change their research agenda a lot.

The analogy here is: imagine if there was only 1 VC firm and it was run by the government. How would that VC firm have strong opinions on what kind of companies people should build?

Exactly.

Patrick Hsu

The grant panels at the NIH are explicitly consensus-based. They have consensus-based scoring mechanisms, and they penalize you if you’re doing work outside of your field and so forth. We go to all this work to train these amazing scientists, and then we don’t let them pursue their best ideas. That’s problem 1.

The Arc investigators are funded to do whatever they want—curiosity-driven research. The second thing behind Arc is this idea that you can divide diseases into 3 categories. You have infectious diseases, and we broadly know how to generate cures for and treatments for infectious diseases. You have monogenic diseases, where there’s 1 genetic mutation or something. We don’t know how to cure those in most cases, but we can screen for them and so on.

Then you have what biologists call complex diseases, where there’s some kind of gene-environment interaction. That’s most cancers, most autoimmune diseases, most neurodegenerative diseases, Alzheimer’s, things like that.

We’ve never cured a complex disease. Many of these diseases are very tragic precisely because not only have we not cured them, we don’t even have treatments, as John says. In the case of Alzheimer’s, for example, we don’t even have treatments.

The question is: can we do something about this? What would a research agenda and program that can help shine some light on these complex diseases look like?

Our hypothesis—we’ll see how much it’s borne out—is that we’ve gotten a couple of new technologies over the last few years. We have single-cell sequencing, so we can sequence the DNA or the RNA in just 1 cell. We have fancy new functional genomics and CRISPR technologies, so you can make these fine edits and perturbations, again, even just in a single cell. Then, obviously, you have transformers and AI and machine learning and all this stuff.

This is a new read-think-write loop in biology that just didn’t exist a decade ago. The question is whether this is powerful enough now to solve some of these previously intractable diseases.

Yesterday, Arc released a new foundation model for biology. It’s the largest biology ML model ever. It’s actually, I think, the largest open-source AI model ever. This is Evo 2.

You’re talking about Evo 2?

Patrick Hsu

Evo 2. It’s not just open weights, like the DeepSeek model or Llama or something. It’s actually open source, and the training code is public. People can read the blog post or the paper.

The thing I find amazing about Evo, and that really surprised me, is that it’s trained on 9 trillion base-pair genomic tokens. ChatGPT LLMs are normally trained on human language. This is a language model, but it’s trained on DNA, the language of life.

There’s only 1 human genome in the training set. It’s mostly other species. Even though it’s only seen 1 human genome, it’s state-of-the-art at predicting the pathogenicity of human genome mutations. A famous mutation is the BRCA mutation for breast cancer. It’s state-of-the-art at predicting the pathogenicity—the harmfulness—of BRCA mutations.

Again, it only saw 1 human genome, and that human did not have these pathogenic mutations. It’s learning something deep across the tree of life. I find that pretty cool.

Is there a phenotypic data set that’s used in training? Typically, when you’re building models in genotype-by-phenotype models, you’re trying to look at the phenotype—the physical characteristics of the organism. What can it do? What does it look like? What are the features?

Then you look at the genome, and that tells you, “These are the specific genes or alterations or mutations that drove this particular phenotype.” That’s what the model tries to learn over time, with the objective being: can I ask it to define a genotype or a genome based on a phenotype, based on a physical set of characteristics I’m looking for, or vice versa?

Can you help us understand what it’s trained on and how that prediction in BRCA is possible?

Patrick Hsu

Great question. It’s totally unsupervised. You’re just showing it lots of genomes, and any kind of latent structure that it learns is based on trying to figure out how to organize that knowledge. We’re not showing it any labeled data or phenotypic outcome data or anything like that.

You can give it a genetic sequence and ask, relative to its understanding of the genetic universe, how likely is this particular sequence? You can do things like predict anomalousness or pathogenicity.

You can also use the embeddings of the upper layers. We’re getting technical here, but you can train another model on top of the model. Even if you show it only a couple of examples, it learns very quickly: “Here’s how the weights of Evo 2 correspond to this particular task.” Those models trained on top turn out to be really accurate.

Did you guys open-source the base model, or did you open-source the fine-tuned model, or both?

Patrick Hsu

We open-sourced the base model. There’s no proprietary reason that we didn’t open-source the fine-tunes. It’s really easy to produce them, and if anyone wanted one of them, we’d happily share it.

Where does it stand in the spectrum of different tools that folks would use to solve these life sciences problems? There are cell models being developed by some people, and there are protein models. Where does this fit in the landscape of foundation models in biology?

Patrick Hsu

It’s obviously very new, so it’s an open question how exactly people are going to find ways to use it and applications for it.

Part of what I think is cool is that proteins and RNA and phenotypic expression—all these things sit on top of the DNA. In some sense, the DNA encodes everything, because the whole organism comes from the DNA. The question is whether DNA is all you need.

With Evo 1, we saw some encouraging suggestions that you can build really good protein-structure-prediction models out of a DNA foundation model, even if you don’t train on a lot of protein-structure data.

It’s a really exciting time, and it’s an open question. I don’t know if you analogize Evo 2 to GPT-2 or GPT-3, but I think we’re going to see a similar Cambrian explosion of applications over the next couple of years.

The thing we’re really excited about at Arc is training cell-state models and trying to better understand how cells change states and what causes them to change states. We’re thinking a lot about that. The reason the weights are in Hugging Face is that hopefully we’ll be surprised by what people do with them.

Patrick, do you expect that over time, as Stripe continues to grow, you can just take some of your excess capital, and other people will do the same, and keep funding Arc? If there’s something that Arc creates or innovates on, and it can generate some amount of money, would that just flow back? Is it meant to be self-sustaining, or is it always going to be via patronage from successful people who just want to keep it going?

John and I are ourselves very committed to it, and we’re underwriting it in that regard. But we’re lucky that there’s a growing donor pool of other people supporting it.

It’s better for an institution if it isn’t beholden to the whims of 1 donor or 1 group of donors. I think that’s a much healthier structure for it.

There’s also a large group of people who are becoming interested in science and realizing that all is not well in basic research in the United States today. The way to see this is to talk to the scientists themselves. They’ll tell you how inhibited they are and the problems caused by the strictures and structures around them.

We don’t see Arc as the answer. Hopefully, it can be 1 point in the space. There are other people doing cool stuff. Brian Armstrong, of course, started NewLimit in the longevity space, and Yuri Milner and others started Altos Labs. There’s the Chan Zuckerberg Initiative. People are trying different things.

Arc is something we’re very happy to support. It’s possible that, over the long term, Arc could become self-sustaining, but that’s not going to happen tomorrow.

When they have this technology-transfer department at every major university, when scientists get grants and work on some innovation, it gets monetized. What happens here? Who owns the innovations, and how do you license them?

It would be amazing if it just wasn’t based on—I believe you guys have put over $1 billion into this. Is that true? You guys have put over $1 billion into this effort?

Not quite. The numbers are public. Arc spends around $100 million a year.

Oh, okay.

It started about 3 years ago, so hundreds of millions of dollars. This is a really significant thing. Again, I want to emphasize that there are other donors, so it’s not just us.

It’s a nonprofit. There have been spinouts, and there will continue to be. If one of those becomes Moderna or the next OIC or something, that could be really good for Arc. Arc might have an endowment and be able to become self-sustaining.

There’s no prospect for us to make money on it in the sense that it’s a nonprofit.

Actually, John, one thing there: I was talking to a friend of mine. You could flip this nonprofit for-profit. I’ve got a guy you could talk to, John.

On the whole modeling world, we talk a lot about the idea that you can use a computer to state the phenotype, or the physical characteristics, you want in a biological organism and have the software resolve the whole genome—all the DNA needed to make that physical organism real.

It can do it from its prediction ability of what genes and what combinations are needed. But we’re a couple of orders away from that, right?

Ultimately, we always talk about wanting to define the plant that can grow on the surface of Mars. It knows the soil type of Mars. It knows the air. It knows that it’s carbon-dioxide-based. It’s 1% of Earth’s atmosphere. It knows what the daylight structure looks like, and it needs to be wind-tolerant.

Then the software predicts an organism that might be able to do that. Obviously, there’s a lot of this predictive work going on in proteins. The higher order is cells—single-cell organisms, microbial organisms—and then, ultimately, multicellular organisms: plants and, finally, animals.

You could basically create organisms from scratch using software, because we have all the other tools to biologically put these pieces together today. But this is a pyramid. There’s a ton of phenotypic data that still needs to be fed in, ultimately, for us all to understand protein-protein interactions and a lot more.

I think that’s right. You can probably derive a certain amount from first principles, just by looking at the genomes. But I think the really powerful models are going to need to do exactly what you say and feed in a lot of ancillary phenotypic and other data—how they fare in different environments.

The sequencing data got ahead of the phenotyping data because there’s so much sequencing data coming in. You can do a beautiful job predicting correctness in a genome, but the sequencing data is really nicely digital, whereas with the phenotypic stuff it’s like, “What even is the data?”

Totally. Dave, while we’re in the science corner, I have a question for you. With your strawberries, you might know the answer to this.

A bunch of tree species around the world are under attack. In Ireland, we have this problem of ash dieback. Ash is Ireland’s national tree, and they use it to make hurleys for the national sport.

Since the mid-2010s, especially as the live-plant trade has ramped up, we’ve had this real problem where so many beautiful trees are under attack. There’s the bark beetle in California and the various conifers that we’re losing. We’ve got to solve black pod disease. The black-pod fungus is destroying cacao and coffee. TR4 is destroying bananas right now.

No, it’s a real issue. This is exactly what we aim to address at Ohalo.

In some cases, you can silence a gene that’s a suppressor of immune function in the organism, which can improve disease resistance.

How do you do delivery of that? Is it airborne sprays, or how do you treat the tree?

Ultimately, if you’re going to use a genomic method, you would transform the genome. You would edit the genome, regenerate a plant or regenerate a tree, and then propagate that tree.

But then we have to replant all the trees.

We’d have to replant the trees.

Can we do a little thing on ash in Ireland?

Absolutely. That’s some of the work we do. We announced a few weeks ago a partnership with the University of Florida to use our methods to introduce disease resistance for a major fungal pathogen that’s destroying the Florida strawberry crop.

That’s what we call a trait program at Ohalo, where we can identify a specific genomic trait and introduce it into that plant. But then you’re right: you do have to grow all the plants back and put them back in the ground.

That’s the second-best alternative to pure extinction. I ended up owning this country house in Ireland with virgin woodlands—woodlands that Ireland used to have when it was fully forested, before it was denuded with the arrival of agriculture. There are ancient woodlands on it from when Ireland was fully covered in trees.

I find the die-off of species very sad, so we’ve got to get on this.

I’m very optimistic. We know how to address these problems. We know how to regenerate the trees. We can do this quickly. We can resolve these problems.

You are right, though. You should be selling a scheme to the people in Tahoe. The Tahoe Basin has been decimated—although decimation is only 1 in 10. Half the trees in Tahoe have been hit by bark beetles.

Those are very interesting because, with insects, you can build very specific defense mechanisms. But we generally have to improve genetic diversity. There’s a natural resistance that comes from evolutionary diversity.

The reason we have a TR4 problem in bananas is that all the world’s bananas grown commercially come from 1 original banana clone called the Dwarf Cavendish. They took that 1 plant, cut clippings of it, put them in the ground, grew another plant, cut clippings of that, and kept multiplying it.

All the bananas we eat, and all the bananas planted across tens of millions of acres worldwide, come from 1 original clone. Because of that, this fungus has been exceptionally capable of evolving to better eat that banana plant.

Sixty cents of every dollar we spend on bananas today goes toward fungicide. We’re spraying these banana trees once or multiple times a week to kill this fungus, and we’re consuming that fungicide. It’s super expensive. If we had genetic diversity—if we had better genetics in the banana programs around the world—we’d be able to radically improve the situation.

No matter what the administration says, you think we need more diversity. Are you in favor of DEI, Friedberg? They cornered you, Friedberg.

You’ve got to make 1 promise to me. You’re not going to start working on raptors. I don’t want to see any of these raptors running around San Francisco.

I find it incredibly inspiring that there’s so much movement in these foundational models. Every day, it seems like there’s something new.

The biggest problem that I think the commercial community is going to deal with is how to take advantage of it, because your head spins. You don’t exactly know where to start.

The biological models are different in that I think it’s a much smaller population of people that will use them, and I think they do have to figure out how to take these models and complement the existing pipeline. The pipeline they have right now is pretty brittle. I think we all know that in life sciences.

My wife struggles with this a lot: how to complement a very traditional pipeline with this kind of stuff. I see it firsthand in how she tries to allocate capital toward these problems.

On the other side, I think these foundational models are really incredible. I was completely wrong on a couple of my earlier thoughts.

One thought I had for a long time was that all these base models seemed to be asymptoting, so I wasn’t convinced where all this capex would go in a productive way. Why are you buying all these NVIDIA GPUs? Then, if you looked at Colossus—the Elon Musk and xAI project that built the largest data center, with over 100,000 GPUs and going to 200,000 in 122 days—what he basically proved was that there are still valuable gains in pre-training. The larger the cluster, the more value there is.

He also benefits, I guess, from the X feed. But it was really interesting. Now I’m a little bullish on NVIDIA. I’m thinking, “Oh my God, if this is true, then all this capex may be justified. You could be buying a lot of stuff.”

I also want to riff on this Grok 3 thing for 1 second. I had 3 takeaways. My first takeaway was that I was sneakily surprised by the upside in pre-training and the value of having a larger cluster. I think that’s very pro-NVIDIA, and it’s also really good in general for foundational models.

My second takeaway is that I don’t know if you watched the livestream, but did you hear some of what these guys had to pull off to make this happen? One of the most incredible things was the way Elon narrated it: first, they had a physical problem. They had to search all around the country for 1 location where they could put 100,000 GPUs, and they found it in an old Electrolux factory in Memphis.

They had only about 15 megawatts, and they had to get a quarter of a gigawatt. They basically had to buy every useful generator that was available, and then they had to liquid-cool it. They bought one-third of all the portable liquid-cooling capacity in America and located it on-site.

Then they figured out that there was a power problem, so they took all these Tesla Powerpacks and had to do power smoothing. That required rewriting all of the Powerpack firmware.

You know how we talked about DeepSeek being this moment where we had lost sight in America of capital being the source of innovation? He proved a more generalized rule that you always have to have a constraint.

Let’s say there’s infinite capital in his case and infinite talent, because he can recruit basically anybody he wants. What did he do instead? He created an artificial constraint of time. He was able to say, “We’re going to get this done in a month.”

Nick from Artificial Analysis showed me the third graph. I want to put it up here because it shows the quality of Grok 3 relative to the amount of time they’ve spent on this problem. That’s what’s staggering to me.

If you project the rate of change—and this isn’t judging OpenAI or Anthropic or anything else; those guys have been doing it for years, and these guys have been doing it for a year—they did all of this MacGyver engineering and were able to pull it off.

That’s my second takeaway: innovation needs a constraint sometimes. Sometimes it’s capital, sometimes it’s talent, and sometimes it’s time. If you can be completely rigid on 1 of those dimensions, you can get a great team to create something.

My third takeaway is that this speaks to the notion of a keiretsu, which is the Japanese word for companies that work together while still remaining independent.

Conglomerates?

It’s more like interlinked companies. Koreans have chaebols; Japanese have keiretsu. This is the manifestation of an American keiretsu.

Elon is able to get engineers from Tesla. He’s not just buying the Powerpacks; he has them re-engineer the actual firmware in real time, on-site. There’s this positive ability to organize effort and human capital.

Could we all stand up a data center and buy $500 million worth of Powerpacks from Panasonic? Absolutely. It would take a few months, or 18 months. Then, when it turned out that we needed to rewrite the firmware, it would take another 18 months.

It’s really incredible what these guys are able to do together. Those were my takeaways. It was really inspiring.

Chamath, a book you might find really fun is called The Henry J. Kaiser Story: Builder in the American West. Kaiser is underappreciated these days. He was the Elon of his time.

He started as a road builder, of all things. He won the contract to build the Hoover Dam. He started a shipyard during World War II.

Yeah, exactly.

He made cars. He decided to make cars. He decided to make airplanes, ships, and TVs. The famous 4-day Liberty ship—remember the propaganda win during World War II, when they were able to lay down those ships? That was at the Kaiser shipyards.

Kaiser Permanente spun out of them as part of their medical system. He was just a complete phenomenon. He kept finding new industries. “Building cars? How hard can it be? Building airplanes? How hard can it be?”

That is the nature of entrepreneurship. The nature of entrepreneurship is doing something delusional and then just letting it happen. Most entrepreneurs do 1 delusional thing once and stay there. Elon and Henry Kaiser, back in the day, did it in the world of atoms—very hard things, on short timelines.

David Sacks

San Francisco now, at least in the physical domain, stands for a kind of stasis. It takes you 10 years to build anything. When Kaiser had the shipbuilding yards here, he went from 0 to 100,000 people in Richmond in 1 year. He basically built the city of Richmond, California.

How do you think these guys pull this off?

Personal sacrifice. Massive personal sacrifice.

I understand that, Chamath, but I’m talking about how they tactically pull it off. You have to be on-site at some point, organizing and directing the team, being able to isolate these problems and fix them. It seems impossible to do it once, let alone 6 times. I don’t understand how they do it.

I actually have some insight into this, just from knowing Elon. A lot of these things compound. Much of what he learned in materials science doing SpaceX, about making the engines and working with metal, shows up in his production at Tesla, specifically in the Cybertruck.

He has learned so much about factories. I don’t think there’s a person on the planet who knows more about factories now, having built a battery factory, a space factory, an engine factory, and a car factory, and now building Optimus on top of that.

These things compound. A lot of the engineers float between the companies. There are people who have worked at SpaceX who then go do a tour at Tesla, and so forth. A number of those people wound up coming into this project.

I’m going to read you a few quotes in this book and see if they remind you of anyone. “Kaiser’s managers challenged convention from the start. As builders, they were expert at coordinating workers and materials. Kaiser was almost contemptuous of traditional methods. His partners had long since despaired of getting him to follow customary procedures.

“In preparing his bids for each new job, Kaiser would try to conceive every possible technique that might justify making a bid low enough to win the job. Once construction was underway, he was forever trying to come up with ideas that would expedite the work. Perhaps more than any other builder, he believed that the faster a job gets done, the lower the costs can be.”

That’s incredible.

What happened with Colossus is that they had told Elon that it would take 18 to 24 months if he wanted to use other network operations centers to host it. He looked at the quotes and saw that they weren’t available. When he found quotes from them, he determined that there was no reason to do this if he couldn’t get it done in 100 days or something. He would be so far behind.

If you look at these 2 charts about Grok, to wrap this segment up and get on to our final 2 segments, these benchmarks and arenas have a lot of controversy around them. People keep leapfrogging each other, but they do give us our best shot at looking at progress.

This is the benchmark for Grok across a bunch of different tests—math, science, and coding. As you can see, Grok 3 has now eclipsed Gemini, Google’s LLM, DeepSeek from China, Claude, and GPT-4o.

Top of the LLM leaderboard.

The thing here, Friedberg, that I’d like to get your comment on is: if hardware is the constraint, does that mean that the person who understands hardware and buildout, as Chamath was pointing to, wins by default?

Jason, hold on. This is what’s counterintuitive. It wasn’t clear that—

No, it was not.

I would guess that in the last couple of iterations, OpenAI has moved to what comes after the base model in the allocation of resources and what they were creating. This is what’s so counterintuitive.

Elon was like, “No.” I don’t understand what he knew that everybody else didn’t know, but the size of that cluster made no sense unless it could produce a result like this—where he basically proved that there was still value in pre-training, where size actually led to better outcomes.

That’s super consequential. I’m in complete agreement with Chamath.

Just to wrap the segment up and put a bow on it, we see these LLMs, and they’ve made incredible progress, as we just heard from Evo 2 and Grok 3. We’re making these giant gains in space and in work. Specifically, in space, Dave, do you think this will get us any closer to Uranus?

So sad. So sad. It didn’t even land.

Okay, let’s do our final 2 segments. We’re going to talk about staying private longer and when you guys are going to go public. Then there’s an asteroid coming. What do we want to do first, boys? Do you want to talk about this asteroid coming? Is it the end of the world if it hits us? What’s going on?

NASA dropped the probability of it hitting Earth to 1.5%. Every day, when the sky gets dark, they can do a better job seeing this asteroid that everyone’s freaking out about.

We finally got a good night sky 2 nights ago. The telescopes were able to get a better trajectory reading on it, and that allows the models to estimate the probability of this asteroid hitting Earth in 2032, when it’s projected to cross our orbit.

Right now, the probability is estimated at 1.5% that it will hit Earth. Based on the size of this asteroid, there’s a range: it goes up to 320 feet in diameter and as small as 80 feet in diameter. That can have a pretty big effect on how much energy would be released if it actually hit Earth.

Even on the high end, if it were 300 feet, it would be the equivalent of a 20-megaton bomb, which is not insignificant. If it were that big, it would hit Earth. If it were smaller than that, it would probably detonate in midair and create a massive shock wave and firestorm.

The region it would decimate would probably be limited to a couple dozen miles, with up to 1,000 miles of effect. If you look at the total surface area of Earth, we’re talking about 10% to 15% of the Earth having enough people to be affected. It’s probably going to land in an ocean.

All right. It’s a 1.5% chance of hitting Earth, and then call it a 15% chance of hitting Earth and causing loss of life. That’s 10 basis points. Then a 1% chance of hitting a city—1 basis point for a city.

It’s a function of how big it is. If it’s actually as small as 80 feet, then it’s not going to be that significant, even if it does get close to a populated area. I’m not losing sleep over it.

Did you come across the Tunguska event in your research? I feel like this is a real boys-are-monitoring-the-situation moment.

No one knows this, but in 1908 an asteroid hit the Earth. It hit a relatively uninhabited part of Russia.

First off, the asteroid did not hit the Earth because it got so hot on reentry. There was an airburst, and it was 1,000 Hiroshimas in size—the explosion. It was the largest impact event in recorded history. Obviously, there was other stuff before recorded history.

It flattened 80 million trees. Weirdly, basically no one was killed because it was so uninhabited. This is quite comparable to the one NASA is talking about.

That’s right. It’s about the same size?

Exactly. The Tunguska asteroid was about 60 meters, or 200 feet, so if this asteroid is in that range and it enters Earth’s atmosphere, you have this kind of explosion in the air.

If it gets above roughly 250 feet, they think it doesn’t burn up fully in the air and actually strikes the Earth. But there you go. This is roughly what we think the size will be if it hits.

Is there a countermeasure? I don’t mean to get all sci-fi here, but if this thing was coming, let’s say in 5 years, is there a countermeasure possible?

It’s a very fast-moving object. It’s moving tens of thousands of kilometers an hour. It’s pretty small—roughly 160 feet.

You’ve got to figure out the exact trajectory, get it perfectly right, launch something off the Earth, and intercept this thing at the exact moment you need to in order to push it off course or detonate something nearby to redirect it.

Technically, it’s very complicated and very hard to pull off. But this is exactly why we have planetary-defense funding at NASA: to track these objects.

This is another example where I’d say AI can play an important role. I have a thesis that AI, more than anything, unlocks deeply complicated projects for humans that would otherwise be infeasible in the pre-AI era.

In the post-AI era, we’re going to say, “Here are all these projects that we do. We mine to the center of the Earth on a daily basis and get rare-earth minerals from 500 miles down. We go to space and colonize the moon.” All these crazy things become possible because AI unlocks large-scale projects that would require millions of people to do things in a coordinated way.

AI can be very smart in this way. I think AI could also play a role in these planetary-defense initiatives.

In the future, you can actually build a complete project model in software for how you would address this problem and then execute it with automation.

There’s a planetary-defense function at NASA. They track these objects, and they’re funded to do it. We hope NASA continues to get funding to do this work. It’s very important.

Guys, it just came through that NASA dropped the probability of an impact event to about one-third of 1%. It’s gotten even smaller, so we can all go to sleep comfortably tonight.

All right. Now everybody’s been waiting for this. Patrick, John, you founded the company in 2010. It’s 15 years later, and the entire LP industrial complex and venture capitalists everywhere—and I’m sure some employees—are wondering: when will Stripe go public, under what circumstances, and what’s the holdup here? Why aren’t you public already?

I think people sometimes hold us out to be dogmatic on this topic, whereas so many other people out there in the world are dogmatic. We’ve just tried to be pragmatic.

Keith was on the show, and he was saying that he believes companies should go public as quickly as possible. Maybe that’s the right thing for some companies, but in Stripe’s case, that hasn’t been the case.

The environment has changed quite a bit. It used to be that, to do any return of capital to shareholders, or if you needed any kind of large sums of money, you needed the public markets. That’s obviously not true today, where stable private markets exist.

We look at it and ask, “Is Stripe better off at the moment as a private or public company?” Up to this point, we’ve determined that private is better. That could change at some point, but there’s no dogma from our point of view.

The last thing I’ll say is that people generally make the argument that it’s critical for discipline to be public, and that public companies run in a more disciplined fashion. I think that’s hogwash.

If you need a 25-year-old Fidelity analyst asking you to double-click on your capex to run the company with discipline, something is horribly wrong at the company and you need new management. That argument has never really resonated with me.

Basically, what you guys are saying is that, from your perspective, you get a lot more return on the time you spend talking with the private investors you have, your team, and customers. It would just be deleterious to your outcomes if you had to talk to these other folks who are talking to you and 50 other companies, don’t really know much of anything beyond a surface level, and may actually distract you and force you to make decisions you don’t want to make.

We’re not even that negative.

Not that negative?

I was going to say, there’s no spiritual status associated with being public. Why be public? It is a cheaper source of deeper and more liquid capital. If you want cheaper and more liquid capital, then by all means, go with it.

But it’s not more moral. It’s helpful to get away from that kind of framing.

I also think it’s noteworthy that, if you look at financial services in particular—and we’re a company at the intersection of financial services and technology—being private for a long time is the norm.

Bloomberg is a private company. Fidelity is a private company. Vanguard is a private company. Jane Street is a private company. Citadel is a private company. Goldman waited 130 years to go public. JPMorgan waited 70 years to go public. Visa waited 50 years to go public.

Those are all different times in history, so you can draw different conclusions from them. But in financial services, there’s always a tendency to be procyclical. I think you need to be particularly careful as a public financial services company to avoid some of those temptations and tendencies.

Financial services generally—and if you look at companies like SpaceX—they’re able to provide yearly liquidity, which is probably better because it smooths out a lot of the volatility. Then people can get back to work.

Are you guys profitable, by the way?

We are profitable.

Fully loaded, GAAP net-income basis? Not community-adjusted EBITDA?

Not community-adjusted EBITDA.

Shout-out Adam Newman. Come on the pod anytime. You’ve got to wear shoes.

I do think, as it pertains to people joining the business and being compensated, everyone loves the idea of an IPO pop. But if you look at a bunch of the other fintech companies, Square is 70% off its 2021 peak, and PayPal is 80% off its 2021 peak.

If you’re an employee and you joined those companies in 2021, it’s not a great feeling. The good and the bad of being public is that you’re priced every single day by the markets, but that isn’t only a bad thing.

What’s the framework you use?

If I’m trying to predict our actions, the framework we use is basically 2 things. First, what matters is less the returns in a given year and more the duration. The question is: what enables the best compounding over a 10-year time horizon, and what’s best for shareholders as you take the longer-term perspective?

Second, what’s best for customers? What helps you build the best products?

You said it: at this juncture, with the business growing at this rate, we want to spend the marginal hour with a customer. This is our life’s work. We’re not going anywhere. We’ll be very happily running Stripe in 10 years’ time.

There’s so much going on in this space. We’ve spent a bunch of time talking about stablecoins and AI and everything like that. It’s hard enough to stay ahead in the world of business without all these distractions. It’s just a question of how you set yourself up to win and do right by everyone. The world is pretty competitive.

If you had to steelman Bill Gurley’s point of view, there are very few founders who are probably as steely-eyed as you guys. What I think a lot of board members in most other situations—not Stripe—deal with is what’s a good forcing function to keep these people on track, focused, and thinking in a multidecade kind of way.

They found that the public markets do that more than anything else. That’s probably the most compelling argument for people who might otherwise get distracted. But for guys like you, who can frankly just do it, it’s great.

All right. Impressive. It’s really impressive. Congratulations.

We appreciate you guys coming on the program. Come back anytime. You were awesome today.

Listen, let’s recap what we’ve learned. People have to put some pants on and get back to work. Constraints make for great art. Stripe’s going public in 2050. Chamath lost $5 billion by not investing. The Collisons read a lot of books, but I’m still kicking.

Live and kicking, bro. Still in the arena. I’ve got a lot of chips still to fire.

South American presidents shouldn’t have their own memecoins, and life finds a way.

We’re coming to South by Southwest, brought to you by the Collison brothers and Stripe. All-In is headed to South by Southwest on March 13th. Friedberg and I are going to sit down and do our interviews—two besties—on the future of media and building businesses in this new media ecosystem. We’re going to have a casual party, food, drinks, the whole thing. The event will be pretty intimate, a couple hundred seats. It’s by application only, with a small $30 registration fee, of which Stripe will take $19. Go to allin.com/events to apply. I’m not BSing about it.

Programming note: the besties are on a tear. We were on Megan Kelly last week, and next week our bestie Friedberg is representing us on Celebrity Jeopardy. We can’t say what happened. Get the clips ready—we’re going to do a recap of every single question. When does it air? Monday next week, I think—I don’t know—Wednesday at 9 PM. Perfect. Before the taping? Yeah, perfect, perfect. There he is between Ana Navarro. She's from The View, right? Well, she's pretty angry. I've seen clips of her.

I should have gotten some counsel ahead of signing up for Celebrity Jeopardy about the lack of upside in doing this. You will see why. We'll talk next week.

Oh no. Bye. Oh no, you lost. Not good. You lost to The View? You didn't lose to The View, did you?

Look, guys, I'm just telling you, I've got a 160 IQ. The View put together doesn't have a 160 IQ, let me just tell you.

Well, we'll talk about it afterwards. Don't tell me they got you on pop culture. You're pretty good on pop culture.

No comment. Okay, love you guys. I've got to go. Love you.

Bye-bye. See you next time. Bye, boys. Let your winners ride, and instead we open-source it to the fans, and they've just gone crazy with it. That's my dog taking your driveway.

Oh man, we should all just get a room and have one big, huge orgy because they're all just useless. It's like this sexual tension that they just need to release somehow. We need to get merch. I'm going in.