The Trillion Dollar AI Reset: Advertising & Gaming is Next w/ Salim Ismail & Dave Blundin | EP #170

What’s the world going to look like in 50 years?

There’s no way. There’s just absolutely no way to make that prediction.

I think AI is going to be listening to my conversations when I say, “Salim, I really love that shirt.” It knows I need a shirt, so it just asks your AI where you got that shirt. Then it magically delivers one to me the next day.

Advertising will get crushed.

You’re exactly right. The future belongs to chatting with your trusted agent, your trusted agent making good suggestions for you, and then saying, “Yeah, just go out and buy that,” or, “Just go ahead and sign up for that.”

You’ll be able to use AI to not just do legal lawmaking and legislation, but actually formulate policy.

There’s a new innovation coming from AI every single week, and you need some kind of regulatory framework.

The only way those laws are going to be written is with AI, because nothing can keep up with AI other than AI.

Now, that’s a moonshot, ladies and gentlemen.

Everybody, welcome to WTF Just Happening in Tech. This week, a special episode of Moonshots with 2 extraordinary friends, brilliant thinkers in the field of exponential technology and AI: Salim Ismail, the CEO of OpenExO, and Dave Blundin, the CEO of Link Exponential Ventures. Both are my partners, and both are definitely smarter than I am. Welcome, guys. Good to see you.

Good to be here. Thank you, Peter.

Yeah.

I want to jump in. The speed at which we’re seeing breakthroughs and activity across the board in all technologies is accelerating, and it feels faster. Does it feel that way for you?

Oh, my God, does it ever? I was in California at Stanford pretty much all of last week and came back, and the amount of change in just 1 week was extraordinary. The new Gemini model came out, and the code it can generate is noticeably better than it was 2 weeks ago. I can tell you some stories when we get into it, but I want to hear them.

How about you, Salim? See, you’re a probability function. You’re orbiting the planet without a rocket ship. Where did you come back from?

I just came back. I was a few days in India, then a day and a half in São Paulo, and I just landed back. I’m trying to figure out where my time zone is—somewhere in the middle of the Atlantic.

You truly are orbiting the planet. Thanks for joining this conversation. I hate missing a week of this conversation because stuff is moving so fast.

The way I frame it is that we’ve hit a singularity in AI development. The pace of change is faster than our ability to process it.

When we talk about a singularity, there are multiple singularities on the way to what Ray calls a singularity. A singularity is, for example, if you can’t predict what’s going to happen next week. That’s a singularity in this area.

I want to jump into 1 of my favorite stories in AI, because I’ve been beating this drum for a while now. It’s not the Trump drum, but the AI-in-the-classroom drum. We talked about this a couple of weeks ago: Beijing has basically ruled that all secondary schools—I think primary and secondary schools—are going to require AI training throughout China. We’ve seen that as well in Estonia, and I’ve been tweeting this out. Finally, we see an executive order here requiring schools to incorporate AI in the classroom. About time.

What are your thoughts, gentlemen? Dave and Peter, you’ve been saying disruption is coming to education. You’ve been saying it—beating the drum for a long time. The schools have underreacted.

I think we’re all associated with either MIT or other schools. We don’t want to get in the middle of a political battle, for sure, but there’s a lot of good stuff coming out of the government now in the DOGE wake, in this, and in a bunch of other things that are absolutely urgent and necessary.

But I tell you, it’s more of what you’ve been predicting, Peter, for a long time. AI is already far better than any TA in any college class. If I think about my grade school and high school teachers, we had some really good high school teachers, but AI is going to bypass them as a way to learn. It probably already has. Then the professors are next.

Remember when the internet came along and there was a lot of concern in the schools that if they put all the courseware online, would anyone keep coming to school? It was all out there and freely available. People did keep coming, either for the social aspects or for the credentials. But now it’s another level, with interactive AI being able to explain things to you better than you can learn them in a lecture.

So we’ll see. We’ll see what that does. Salim, you and I both have boys the same age in middle school. Is your school teaching them AI? What’s the philosophy there?

They’re using AI. They’re not supposed to use it for writing essays, but they do it anyway.

The good news is that at that age, it’s pretty easy to spot when sophisticated AI has written the thing. It’s way better than the kids do it, so for now it’s kind of workable.

But I think the key point, referencing what Dave said, is that when you’re using AI to learn, you learn at least twice as fast. The estimates I’m hearing are between 2x and 4x faster. That means you’ve immediately made the need for formal schooling redundant.

There’s now a huge gap between the regulatory requirement—we want to put kids through school through grade 12—and what’s possible with AI. That gap is going to stretch very quickly.

MIT and Northeastern went the complete opposite direction from your kids’ school. They said, “Use AI anywhere and everywhere that it helps you.” If that messes up the professor, they’ve got to fix it. They’d better figure it out in a hurry.

I’ve got 1 kid left in high school, and he’s really mad that his school went the other direction, saying, “Don’t use it. We’ll catch you if you try.” It’s just crazy. It’s incredibly shortsighted to take that position.

I can get the idea that middle schoolers shouldn’t be using AI to do middle school homework, but middle schoolers should be using AI to do extraordinary projects that they could not normally do otherwise.

There’s a second article related to this. Over 250 CEOs are urging states to make AI and computer science graduation requirements.

Here are the stats: 1 computer science class can increase wages by 8% for students. The U.S. is lagging behind China and South Korea, and the open letter advocates for computer science and AI as core subjects in K–12.

I could not agree more. This genie is out of the bottle, and kids are going to hit smack into a brick wall if they’re not prepared.

The good news is that those kids who don’t take computer science will be much more ready for all the manufacturing jobs that are coming back.

Ouch.

All right, here’s another one. The UAE deserves credit in so many different ways. It was the first country to have a minister of AI, whom I’ve known for the better part of 15 years, before he became minister. The UAE is the first nation to use AI to write its laws.

Here are the stats: The official project will reduce government costs by 50% and speed up lawmaking by 70%. Salim, you’re meeting with heads of nations on all of your trips. What are you seeing here?

They are very far ahead. Most other countries are nowhere near even looking at this. They’re still trying to spell AI.

This is a huge opportunity, and I think they’ve actually underestimated it. I think it’ll speed up lawmaking by 5x—forget 70%. I think the drop in cost is much bigger than they think, because you’ll be able to use AI not just for legal lawmaking and legislation, but actually to formulate policy.

I’ve said this before: If you want to drop inflation by 1%, asking a human being to gauge which knobs and levers to pull is a stupid idea. Let it go to an AI. The AI can look at all the financial transactions and access all that data in real time and say, “Here’s what you need to do.” Then the human being can say, “Yep, that looks good. Go for it.”

There are all sorts of things that could happen here.

There’s a crazy story I have, but we don’t need to get into it right now. A few years ago, I was asked to give a talk at the Republican National Leadership Conference on some of this. The title of my talk was going to be, “How Would You Drop the Cost of Government 10x Within 10 Years?”

Nice.

Which you could do. But you’d have to embrace technology, and that’s where we got into a bit of a mess there.

I think these predictions are nice, but they’re conservative. I think they’ll end up doing a lot more, and I think that’s incredibly powerful, because you could use AI for policy formulation, legislation writing, and also on the enforcement side. This really changes the game in how government operates going forward, and you can save a lot of money on government employees.

Dave, what are you thinking here?

As I mentioned, I was at Stanford all last week with Eric Brynjolfsson, our good friend, and also Sandy Pentland, who moved out there recently. Mark Gorenberg, the chairman of MIT, was also there the whole week.

Eric is very forward-thinking on these topics, but his point in his lectures on bullet 2 is that these cost-saving areas are the tip of the iceberg. It's a great way to get started, but the real gain here is greenfield. It's what you can do that's incremental and beyond just reducing the cost of something that already exists.

In this area, we desperately need the pace of lawmaking to keep up with the pace of AI innovation. I think this is a great way to get started, and the UAE is very forward-thinking on it. But you're going to need this same capability when there's a new innovation coming from AI every single week, and you need some kind of regulatory framework; otherwise, it'll run wild. The only way those laws are going to be written is with AI, because nothing can keep up with AI other than AI.

Yep. If you're in whatever industry you're in and you're being regulated—whether you're a hairdressing salon or a healthcare facility—if you do an analysis of all the laws that regulate you, you're going to find conflicting laws on the books.

I think what would be fascinating is actually running all the laws and saying, “I'd like you to disentangle these, deconflict these, and actually find the set of laws which are the minimum laws that allow us to operate,” like the tax code, for example.

Oh my God, could you imagine running a reinvented tax code? It's like, simplify it and maintain the U.S. revenues from taxes. I mean, we've been off on this for a while.

Yeah, I think we're pretty clear as to what the sequence is. Step 1 will be to use AI to navigate the existing complexity in the tax code or legal structures, et cetera. In Europe, no matter what you do, you're always violating a few laws, right? It's just so much legislation. That's one way, but then they don't have the people to really enforce anything, so everybody just goes ahead and does what they want. And because they generally follow common sense, it works out. So they're lucky over there.

But step 1 will be using AI to help navigate this. There's already a company that's helping nuclear projects navigate the fastest path to getting regulatory approval and what forms you have to fill out, et cetera. So that's already happening there.

Step 2 will be: How would you optimize the regulatory framework, stop the conflicts, and rewrite it? And I think step 3 would be: Is there a way of making this much more elegant and cutting through, using first principles, to rewrite the legislative foundation of the legislative structure? That's when I think we'll get real benefits. But certainly, we'll get huge benefits in the first couple already, and we can do the first one already. I think we'll see a lot of that going forward.

Yeah. Amazing. I've never been excited about rewriting laws, but I do think there needs to be some way—and some requirement—that every time you put a law in the books, you should take one of the old ones off, or at least simplify them. All right, let's move on here.

This is a big one. This came as a bit of a surprise: OpenAI scraps plans to convert to a for-profit structure. OpenAI abandons plans to shift after facing opposition from Elon and state authorities, and they're going to maintain what they call their nonprofit-controlled, capped-profit structure.

So what does that mean? There's a nonprofit board that oversees OpenAI. All of the nonprofit board members have no equity incentives, meaning they're there to help maintain OpenAI's mission in the world and not to maximize profits. The capped structure means that if you put a dollar in, your return on your investment is capped at $100. I find that fascinating. We reported last time that OpenAI raised capital at a $300 billion valuation. So, thoughts on this one?

Well, there are a lot of reports that, hey, this is no big deal for Sam. It's actually, if anything, going to reinforce his control. I didn't read it that way. I was like, no, this is a win for Elon and trying to slow down the rate of investment and give him a better chance to catch up with his unconstrained version of this.

I also think that the number-one use case for AI is companionship. The voices and the sympathetic AIs got really good in the last couple weeks, and I think they'll be out in the world in the next couple months, where you have a very, very intelligent AI model—a 2 trillion- to 5 trillion-parameter model—married to these super-engaging, sympathetic voices. That's going to create all kinds of tough decisions for social media companies: Am I just selling as many ad impressions as possible and trying to reel everybody into being on this platform 24/7, or am I worried about the psychological effect of that?

I think that, having a nonprofit board on top of OpenAI, they're obligated to think in terms of being the social-good version of this, which may not be the biggest revenue driver.

Salim, what are your thoughts here?

It seems pretty clear that there was going to be huge legal exposure if they pushed through with the full for-profit, so this made it more elegant. He's going to control the nonprofit board, so that kind of minimizes any risk. And then you can always find other ways—you can do other spinouts, et cetera, et cetera—to harvest some of the upside that may come from this. So I think it'll afford them a chance to change route. I don't think it'll change the game that much, as Dave said.

One of the things, Dave, you mentioned in passing was the emotional engagement of the AI. So OpenAI's ChatGPT actually rolled back that level of emotional engagement. They were saying that it was just way too complimentary to the user, and I don't know if the term “obsequious” is the right term for that, but they rolled it back to an earlier state. You don't want your OpenAI saying, “Oh, you're so amazing,” all the time.

Mhm. You know, there's this old definition of diplomacy, which says, “You say, ‘Nice doggy,’ until you can find a rock.” So the AI is being really nice to us until they figure out what to do with it. Ouch. All right.

Not a surprise here. Microsoft's CEO says up to 30% of the company's code was written by AI and predicts that 95% of code will be AI-generated by 2030. Dave, you're overseeing how many companies right now inside of the Link portfolio and Link Studios?

Twenty-four right now.

Twenty-four. And so what's it like there? Take a second and give people a little bit of an overview.

Link Studios is incubating companies, mostly out of MIT, some out of Harvard. The median age of those teams is probably 24 or 25 years old, and the companies are about 1 to 2 years old.

They're all AI-native. They generate almost all their code with some degree of AI assistance. Blitzy, in particular, when I left town last week, had a $2 million–$3 million revenue run rate. I came back, and they've signed $8 million of new deals, so they're basically signing up customers as quickly as they can meet them.

Blitzy, as a reminder, is a company that can write 3 million lines of code in a single night. It uses massive amounts of inference-time compute under the covers. Now the trick is, okay, how do you debug 3 million lines of code that you've written in a single night? But the algorithms are getting so good at writing good code now that that problem is going away in a hurry. There's no way any company in this building wouldn't be using AI-generated code for almost everything that they do.

Also, it makes a lot of sense to write your business plan around areas where it works well. This is coming up a lot: Don't fight any battles. Wait a few weeks, because the things it'll be able to do in a few weeks will be expanding at this ridiculous rate. So while you were fighting some silly battle, it got ahead of you. You just do what comes naturally and easily this week, and then wait for next week's innovations.

And a question for you: It used to be that if you're starting a company, especially in the AI space, you need to find a great technical co-founder if you're someone in the business or finance world. How important is that now versus being able to use AI as your technical co-founder?

Well, it's, in a sense, more important than ever, but it's a different skill set. The visionary that can decide, “What should we be doing?” has become incredibly important. And then the technical co-founder who would just grind it out—Steve Wozniak would be the quintessential example. Steve Jobs is like, “We need to build the Apple II. This is what it needs to be like for the user.” That skill set's more important than ever.

“Hey, Woz, can you get the chip count down? Can you wire this all together?” Grind it out all night. That skill set becomes much less relevant because AI will do all the chip routing for you. But that first skill set has really grown in value, and it's very similar. It's very technical: You've got to choose components; you've got to decide this will work, this won't work. It's a very technical skill set, but it's more of a visionary skill set.

I think there's one other skill set that's coming to the fore here, which is finding hidden gems of data sets and figuring out how to monetize them.

And I think that's going to become a very big deal over the next few years, because we'll have horizontal platforms able to do all sorts of things. The question is, where's the revenue? Where can you mine for little nuggets of revenue, go grab them, and leverage AI to do it? I think that's going to add to the business focus and the application focus.

So, 2 more things on this topic that I think are really important. First, Mark Zuckerberg last week did a couple of interviews where he made it clear that the backlog of AI research projects is only compute-bound. It's not idea-bound. Between the researchers and the AI-generated ideas, they have a huge backlog of things that will just work.

And there isn't enough compute. Even in Microsoft's hundreds of thousands of GPUs, there isn't enough compute to try the ideas. So, we've tipped from idea-bound to compute-bound, and that'll be true forever hereafter. On the second bullet, 95% of code will be AI-generated by 2030. I think that timeline to 2030—Satya would have said 2040 or 2050 just a few months ago—is consistent with Demis, too. He was more conservative about the timeline to AGI, and he's pulled in his forecasts now, too. So, what you're seeing is that even the conservative, let's-be-cautious people are really pulling in their dates. And that's telling.

It's the AI singularity, as Salim and I spoke about earlier. Speaking of which, our AI czar for the US, David Sacks, has come forward with a statement that AI will be a millionfold more capable in the next 4 years. Let's play a video and talk about it, because this was the same point that Elon made when he was on stage at the Abundance Summit. Let's hear the numbers that David puts forward.

I would say the rate of progress is exponential right now on at least 3 key dimensions. Number 1 is the algorithms themselves. The models are improving at a rate of, I don't know, 3 to 4 times a year. Then you've got the chips. Depending on how you measure it, each generation of chips is probably 3 or 4 times better than the last level. And that would be the third area where you're seeing basically exponential progress.

Just look at the number of GPUs being deployed in data centers: xAI's data center, Stargate, and within a couple of years, they'll be at, I don't know, 5 million GPUs, 10 million GPUs. The algorithms, the chips, and the data centers are all improving or scaling at a rate of, I don't know, 3 to 4x a year. That's 10x every 2 years. If you're getting better at 10x every 2 years, that doesn't mean you'll be at 20x in 4 years. It means you'll be at 100x. So, you multiply those things together—the algorithms, the chips, and then the raw compute that's available—and you're talking about a million-x increase.

So, Salim, how does that feel?

It sounds about right. Look, this is the fastest-moving technology we've ever seen, and you would expect to see it. I think he framed it very clearly and wonderfully. But we've seen the same thing with, say, solar energy, where it's not just the efficiency of the solar panel; it's the actuators and the system controls. They're all accelerating at their own rate—not quite at this level of madness, but we've seen this before, where the convergence of multiple exponentials gives you crazy outcomes.

One of the things that we've talked about in the past is skating to where the puck's going to be, right? If I remember meeting the founders of Siri before they were bought by Apple, they were predicting what was going to be possible in 2 to 4 years, and they were building Siri to intercept those capabilities. If you build for what's possible today, then by the time your product gets to market, it's out of date. So, Dave, you're seeing a multitude—hundreds—of startups coming across your desk. How do you think about this?

Well, first of all, the million-x is just raw parameter count in the biggest neural nets. It's really hard to predict the capabilities that are and aren't possible with a million-x. You're a great all-time visionary. Looking 5 years into the future used to be very possible for you. What do you think your view into the future is now?

I mean, it literally has to be a discontinuous vision. I remember I was brought on stage at the first World Government Summit in Dubai. My request was, “What's the world going to look like in 50 years?” And I was like, “There's no way. There's just absolutely no way to make that prediction.” Ten years—back then, it was 10 to 15 years at the utmost. Today, I would say 5 years is difficult; 3 years is possible. There are so many parameters, including policy, human reaction to this, and the capital markets. All of these things are intertwined with what's going on in China and Taiwan.

So, for me, the hard part is when something is a million times—assume it's 100x, 100,000x—you end up with a failure of imagination, where you can't conceive of the types of things you might be able to do. Our immediate thing is to automate what we're doing now, and that's not the right answer, because you end up with the problem of when they first put TV out: You took radio announcers and put them on TV, reading the same scripts, and you're not leveraging the capability in a real way. It takes time to get to grips with the imagination that's possible. So, you kind of have to wait till we get there and then go, “Now what do we do?”

I know one thing, Peter: when you were out here 2 weeks ago, spending time with Alex Wissner-Gross, there's just a handful of people that you can find who've got their heads wrapped around this. If you talk to people at random, 99.9% of them are underreacting. More than that, 99.999% are underreacting. You find a few people like Alex who are thinking it through, and if you can surround yourself with as many of those as you can find, it really, really helps a lot.

Just to pick on the framework here that David Sacks put together, it's the exact same 3 dimensions that I use to think about scale. I think a million-x in 4 to 5 years is the right answer. But I also think that the algorithm dimension—nobody really knows, and if anything, it's understated. We're all surprised in the last few months by inference-time compute, or test-time chain-of-thought reasoning, being far more powerful as a scaling factor, and we're trying to build new scaling laws around how we even think about inference time.

When David talks about the chips getting faster, he's thinking about the GPUs and just the standard 3- to 4x year over year. But the inference-time chips are actually improving far, far faster. It's a much easier problem, but that's turning out to be a key factor in the intelligence of the overall system. So, I think the chips will outperform what he just said, and the algorithms will way outperform what he just said because of those innovations.

When I look at the factors driving this, it is compute—it's total compute—datasets, which we haven't talked about; synthetic data hitting this; algorithms; money coming in; and then the number of actual human meat bodies getting into the game here. More and more people are getting into the AI game. Something funny happened: I was having a conversation—I think it was with ChatGPT; it might have been Gemini 2.5—and I was saying, “Okay, let's look into the future.” I was trying to use it as a conversational partner to predict where things are going.

I said, “Make some predictions.” I said, “Listen, if we’re going to see a century’s worth of progress in the next 10 years, what you just predicted is extraordinarily linear.” He goes, “You’re right. Let me go a little bit bolder.”

So, I think this is where the answer will lie. When we see these types of predictions, we’ll use an AI to say, “Write out that curve and tell me what it’s going to look like.” It’ll probably do a better job than us because we’re limited in other ways. We have all our cognitive biases that limit our ability to imagine, whereas an AI, in theory, should suffer from much less of that.

Yeah, I think this is a really important topic, because right now AI as a brainstorming partner is hobbled. The models that you love brainstorming with—the GPT-4os and the Gemini 2.5s—have about a 30-second lag, and that kind of takes the buzz out of it.

The models that are really fun to interact with, like the Sesame demo and all of that, are super engaging, but they’re only 7B- or 30B-caliber models, so they’re not really good brainstorming partners. All of that will be stitched together within the next 3 months.

When you have an idea in the morning, Peter, you’re going to immediately go to that super-engaging, 2 trillion- to 10 trillion-parameter-caliber model, and there’s no going back from there. Once you start down that path, you’re going to love it and never go back.

3 months. Did you say 3 months?

3 months.

Yeah. Well, ChatGPT—the phone app—does a pretty damn good job. I have conversations with it. I do about 20 minutes of red light in the morning, and my conversations with my ChatGPT bot during that red-light time are extraordinary.

All right, let’s go to Eric Schmidt here, who was giving a presentation. Let’s take a listen.

I’m here to tell you that I honestly believe that the AI revolution is underhyped, and here’s why. The arrival of this new intelligence will profoundly change our country and the world in ways we cannot fully understand.

Many people think that the demand, in terms of energy, that our industry takes will go from 3% to 99% of total generation. One of the estimates that I think is most likely is that data centers will require an additional 29 gigawatts of power by 2027 and 67 more gigawatts by 2030. That gives you a sense of the scale that we’re talking about. These things are industrial at a scale I have never seen in my life.

So, first off, underhyped versus spot-on. How do you guys feel about that? Then we should talk about the whole energy of it all.

Dave, you know, one of my most cherished lifetime memories, actually, Peter, is you sitting down with Eric Schmidt. I think it was at A360 just a few years ago. You guys must have riffed for maybe 2 hours straight in armchairs side by side.

Midnight there. Oh my God. Eric is not only absolutely brilliant, but he has access to all this information. When is he ever going to be wrong? When has he ever been wrong before? Of course, he’s absolutely right.

What’s amazing to me about what Eric is saying there is that he’s on some kind of congressional testimony or whatever. He’s always pointing to the energy because he knows that’s the one thing that’ll get through everybody’s thick head.

But there are so many other, more important and urgent things. What about the fabs? We’ve got to get these fabs up and running onshore, like yesterday. That’s the more urgent, bigger topic than just the raw energy. But he knows that they won’t understand that.

So he has to say, “We need 29 gigawatts of new power,” because at least everyone will be able to grok it, and it’ll make progress.

Yeah, I don’t have anything to add to this. I think those are 2 great insights. I think this will be a really great push for renewables and low-cost energy in all sorts of ways, and I think it’ll force us to expand our energy footprint.

At the same time, you’ve got the White House basically taking all the constraints off coal, oil, natural gas, and drilling. I kind of understand why in terms of AI, because when I look at how long it’s going to take to get Gen 4 nuclear and fusion online, we’re talking decades.

It’s like Amazon signed a contract with X-energy for a Gen 4—I think it was a Gen 4—plant. It’s a fission plant, and I’m like, “Is it going to be online in the next 2 or 3 years?”

2040.

Yeah.

But you have to remember that the cost of fossil fuels is way higher these days compared to solar. All this extra usage will be coming from renewables in some way.

I go back to my example of the guy building a gigawatt wind farm on the outskirts of Kazakhstan, because the wind howls down a particular valley with ridiculous consistency, and they just plunk an AI data center there. You can do that kind of leveraging of low-cost, faraway capability that you never get in normal areas—or go to Iceland. You can put the data center anywhere you want.

Yeah. I think when I remember talking to Ramez Naam about the energy stuff, he said we will never go back to oil and gas as fondly as people want, because the cost curve is just too high.

Can I just drill into that a little bit?

You can drill into the oil and gas.

Yes.

In 2016, it became cheaper to create a solar-generation facility than a fossil-fuel facility. Since then, almost all energy generation has been solar.

In 2019, a more important inflection point was hit: it became cheaper to build and run a solar facility than to just run an oil-and-gas facility. The operating expenses and capital expenditures of solar are now less than the operating expenses of fossil fuels.

We don’t need government policy. We don’t need any of this stuff. The pure market will now drive us toward going fully down a renewables route.

The problem is that the amount of energy we need is just so huge that we need to get much more quickly to the SMRs. We also need to get rid of the delays due to regulations and permitting. This is what’s going to slay us.

If we can make permitting super, super fast—and this is where AI can support it—but if we put an SMR in far-off, remote areas of the United States or Alaska, or God knows where, you’ll end up with the capability to generate energy. Then you can just plug an AI data center next to it, and you’re in great shape.

So, I think we’ll see a ton of that. I don’t think the regulatory process is that much of a big deal right now. I think it’s just the sheer amount that we need that we have to figure out.

Mastercard unveils Agent Pay, agentic payment technology to power commerce in the age of AI.

It’s not just Mastercard; it’s going to be all the credit cards. I think there’s a very near future where I speak and stuff shows up on my doorstep. I don’t think about it; it just magically happens.

In fact, I’ll take it a step further. I think AI is going to be listening to my conversations. When I say, “Salim, I really love that shirt,” it knows I need a shirt, so it just asks your AI where you got that shirt, and then it just delivers it to me magically the next day.

I’m trying to see how you got there from this.

Well, I think we crush advertising, and I think we crush the ability to select payment methods. I think your AI is just doing all the stuff automatically for you.

Yeah. Bill Gross and I were talking earlier today with Perplexity AI, your good friend, about this exact topic: advertising will get crushed. What do you mean exactly there? Google has $300 billion of advertising revenue. You’re saying that’s going to get crushed?

I mean, that’s exactly right. Actually, Richard Socher, our mutual friend, was saying that in their tests, the revenue per session of an AI dialogue versus a Google search is about 1/10 to 1/100.

So, where does that put Google if, when you do a Google search now, the top thing is the AI answer and then the ads start below that? That’s already got to be cannibalizing the revenue. It puts Google in a really tough spot, trying to figure out, “Do we cannibalize our own ad revenue?”

But you’re exactly right. The future belongs to chatting with your trusted agent and your trusted agent making good suggestions for you, and then saying, “Yeah, just go out and buy that,” or, “Just go ahead and sign up for that.”

Yeah. Well, we’ve mentioned this before, but I think the holy grail here is your Jarvis equivalent, right? Your agent will act on your behalf and be bound to you with your preferences, habits, and so forth. Then you can put a lot of trust in it, just let it go to town, see what happens, and have a feedback loop so it just tweaks itself.

That seems like a no-brainer to me. I just spent the most unbelievable amount of time dealing with flight tickets, logistics, and seat changes. It was just unbelievable.

Erik, I was on this panel on the future of AI working groups with Erik Brynjolfsson and a few others, and they called it white-collar drudgery. There’s a ton of that going on, and I think this is where we’ll be able to have a huge impact with AI.

Yeah, we should keep tracking this thread, because trust—when do you trust AI, and when do you not trust it? This is just a subset, but it’s an important one. My credit card is a pretty important thing. Do I trust it? Self-driving obviously involves even more trust. Then you’ve got life decisions: What do I leave to my kids? Do I buy that house? Do I get a new roof? Those decisions are also definitely trust decisions.

So we should track how a lot changes week to week. We should track what people trust and what they don’t trust, because God knows AI is going to be incredibly convincing and persuasive. I think one of the superpowers AI will have is its ability and command over language, right? Like an amazing politician who’s able to take you from wherever you were and sway your point of view.

Mastercard wants to use Agent Pay to become my default payment system. You were talking about Bitcoin earlier today. Do you want to share some thoughts?

Well, I think here, we’ve been talking about crypto payment mechanisms, right? You’ve got this spectrum of stuff that’s really hard to mine—Bitcoin is not that easy to do payments on—and then you get gold, say, if you want to use the old analogy. Then you have bank vaults and jewelry, which aren’t that easy to transact, and then you get to bank accounts, and then you get to credit cards. You have this spectrum: The smaller the transaction, the less security you need, and therefore you can make the usability much easier, et cetera.

What’s happened with crypto is that, along with things like the Lightning Network on the Bitcoin protocol, the payments are now really, really easy. I think we’re going to see crypto-type payments scale very, very fast, especially with digital fiat currencies moving over to the crypto world. We’re going to see incredible things happen there, and I think that’ll be the biggest trigger for the usage of blockchains and so on for handling all of that stuff.

There’s a startup, and I’m on the advisory board of it, that’s trying to figure this out, because one of the biggest problems with trade globally, separate from all the tariff stuff, is that big countries don’t like big chunks of foreign reserves leaving their shores. So they make it really difficult, put capital controls in place, and make it really hard.

There’s this famous story of GE selling $100 million worth of jet engines to China. They took in receipt of $150 million worth of frozen chickens as payment, and then the extra $50 million meant that now they had to deal with all these chickens. Hence the margin.

But it turns out that type of transaction is rampant in big companies, because China’s like, “We don’t want $100 million leaving our shores,” right? A combination of crypto plus some of these payment rails is going to change the game with some of these big transactions. On the smaller side, AI plus some of these will make things really radically much easier.

I love it. Digital chickens. So here’s another one. Dave, I think you added this one: NVIDIA acquires Gretel to power next-gen AI. Tell us about this story.

Yeah, I think this is important because we talked last time about the CoreWeave IPO, which is holding up just fine. Is that the bellwether to the markets opening? M&A has been very quiet. Now you’re starting to see some deal flow, and that all ties back to the IPO markets opening up. The last 3 years have been the most dead in the history of the country in terms of IPOs. So if that comes roaring back, that’ll unlock a lot. This is a good one to track here.

What you’re seeing here is NVIDIA recognizing that synthetic data is incredibly strategic. We talked about Richard Socher’s 10 areas of AI, some of which run unbounded and others are throttled or hobbled by some constraint. Math is unbounded; coding is unbounded. Why? Because you can generate a near-infinite amount of synthetic data, which means the AI can continue to learn.

Then you can build really easy evals: Did the AI do something useful, or is it buggy? Does it not work? The math is provable, the code is debuggable automatically, and those areas require synthetic data.

But you look across the other areas, like biotech and everything else, and synthetic data is the constraint, because AI can run wild to the extent that it can generate synthetic data. I think NVIDIA, being in the middle of the hot seat, realizes, “Oh, my God, what a great opportunity to create some kind of a moat. Let’s pick up Gretel and let’s build out a synthetic data competency as part of our overall strategy. While we’re top of the world, we need to think about how to defend our position on top of the world.” So that’s why they picked up Gretel.

Nice. Yeah, I think this is a really great strategic thing. You saw Zoom doing the same thing when its market cap went so high: They started buying up a whole bunch of stuff, leveraging their currency while it was high.

As an investment theme, synthetic data is very different in each domain. Obviously, synthetic data from a whole-cell simulator in biotech is going to be very different from synthetic data in math. So it’s a great repeatable investment theme if you can find really, really smart teams, because I think NVIDIA could easily acquire 10 more companies doing different things. Are you investing in synthetic data?

Yeah, we have DataCebo, which is a co-investment with Mark Gorenberg, the chairman at MIT. They do column-wise numerical synthetic data, so things like actuarial or whatever.

But again, if you had physical-world synthetic data, like simulators for physical objects driving data, those are all investable. We’ll do any synthetic data company we can find if it has a great team.

Fantastic. All right, here’s one that I added: Kleiner Perkins leads a $130 million round in Coinbase founder’s anti-aging startup, NewLimit.

Let me break this down. Brian Armstrong and Blake Byers are both friends. I’m going to have Brian on this Moonshots podcast sometime next month. They’re both passionate about longevity, and what we’ve seen is a number of billionaires investing in something called epigenetic reprogramming.

What is that? As I like to remind people, you get 3.2 billion letters from your mom and 3.2 billion letters from your dad. That’s your genome—about 22,000 genes. You have that same software when you’re born, when you’re 20, when you’re 50, and when you’re 100.

So the question is, if you’re running the same software, why do you look different? Why don’t you have the 6-pack at age 100 that you had when you were 18?

It turns out it’s not the genes you have; it’s your epigenome. “Epi” comes from the Greek word for “above.” It’s which genes—it’s your control system. It’s which genes are on and which genes are off. As you get older, the genes that should be on get silenced and turned off. The genes that should be off get turned on, and there are a whole bunch of issues.

Folks like George Church and David Sinclair have been pioneering this. Epigenetic reprogramming asks: Can we turn back the clock and set your epigenome to an earlier state? The data appears to be not only yes, but hell yes. It’s probably one of the hottest areas of age-reversal work and longevity.

I just find it fascinating that Brian is investing here. We’ve seen Jeff Bezos and Yuri Milner invest in Altos Labs. Sam Altman invested in Retro Biosciences with Joe Betts-LaCroix. Same area: epigenetic reprogramming. So, yeah, very exciting.

It should be noted that a lot of this is now made possible because of AI.

Oh, how so?

The ability to actually process all the data from the experiments and decide what transcription factors to use is enormous. We are 40 trillion human cells running 2 to 5 billion chemical reactions per second, and just trying to understand the amount of data is really hard.

When I visited both Brian and Blake at NewLimit, they were very proud: The majority of their team are all AI-first technologists. They’re not lab geeks.

So, any thoughts here?

I think, as our AI models get more sophisticated, the ability to apply them—that number blows. I never stop shaking my head when you say that number of 2.5 billion chemical processes per second.

That may be across the body, but there’s some incredible number just in a cell. Isn’t it per cell?

That’s per cell, right? That’s just unbelievable—trillions of cells, right?

Yeah, it’s just mind-boggling how complex what goes on in there is. It’s going to be amazing to see what’s possible going forward.

Well, definitely not my area of expertise, but I remember clearly: I took AP Biology in high school, and then I got to MIT and said, “Yeah, not doing that anymore,” because all we were doing was memorizing random Krebs cycles. There’s no way this is actually going to work. There must be billions of times more things going on than what we’re memorizing here.

So true. So true.

All right. I have 2 13-year-old, soon-to-be-14-year-old boys. Salim, you have 1, and Roblox is part of their, if not daily diet, weekly diet. I found this amazing: Roblox is averaging up to 43% of Netflix users, with 97.8 million average daily active users and 7.2 billion hours. It’s just exploded onto the scene. We can see this rapid, rapid growth of the gaming world.

Dave, did your boys all sort of go deep into the gaming world?

Well, it’s interesting. I wouldn’t say they were big-time gamers, but if you look at the amount of time they spent on interactive media, including Fortnite and some video games, it was much, much higher than passive media like movies and Star Trek or whatever.

So, that's just the trend line. But you add AI on top of video games, and it's so incredibly engaging. There's no way people are going to go back and watch passive media again. And I think it still really hasn't hit, right? If you're in Minecraft or Roblox, the amount of AI-generated reality and the amount of AI interaction with you—we're going to see a lot of advancement in the gaming world in the next few years.

Yeah. You know, the other thing is gaming is dominated by young boys. It largely misses the girls, for whatever reason, but it definitely misses older people. The new, super-engaging AI voices actually tend to cover everybody, and so it's going to have a lot broader appeal. Any numbers you see in this world are going to go up tremendously just by covering the rest of humanity.

Also, MrBeast now, I think, is the most recognized person on the planet, which is shocking. And that might be among stars or have some other qualifier, but it's an incredible stat. One of the secrets to MrBeast's incredible success is the internationalization of all content. When you see MrBeast in any jurisdiction, any country, any language, his voice is in that language and his lips move perfectly—or well—for that language.

You wouldn't even know that it wasn't shot for you in the Congo. He mastered that using AI, and that got him a lot of the footprint. The same thing applies here to internationalizing the language, but also the content and the media, making it a cultural fit for every jurisdiction.

What struck me about this was that almost all of that increment is mobile stuff. So they're doing it everywhere.

That's crazy.

Yeah. Amazing. All right, let's move on here. Let's talk about what's going on in the space world. The first news is that Amazon—this is Jeff Bezos—launches the first batch of Project Kuiper internet satellites. We've seen Elon and Bezos marching up the curve in space. Amazon launched 27 satellites for its global internet. Ultimately, this will be thousands of satellites in orbit.

I find it interesting they used an Atlas V. Of course, Jeff has something called Blue Origin, which is building rockets. Today, it's launching suborbital flights, but it did get one of its first New Glenn vehicles to orbit. I think he's feeling pressure to get Kuiper up and running at the same time that Starlink is now probably close to 8,000 or 9,000 satellites. So they're buying launches from ULA instead of using his own launch vehicle. It's going to be interesting to have these 2 internet constellations in orbit operating.

Here's a question I have. Have you seen, Peter, any kind of estimate as to, once you've got high-functioning internet over space—which we kind of just about are there now—how much will move to that versus terrestrial internet?

Yeah, it's a good question. You still have time delays, but one of our next articles coming up, which we'll talk about, is quantum teleportation, and that would get rid of time delays. If you add quantum teleportation on satellite constellations, yeah, it's instantaneous and global. It starts to get pretty freaky.

We're going to start—we'll have, I think, probably 30,000 or 40,000 communications satellites in orbit. We'll have a constellation around the Moon. We'll have a constellation around Mars. All of those will be cross-linked by laser. So you'll have an interplanetary internet up and operating very shortly.

Vint Cerf has been talking about that for a while. In fact, he's designed the proper communication protocols for that.

He has indeed, and it's all playing out.

Here's the second space article: SpaceX has asked to increase its rate of Starship launches. Starship, of course, is the fully reusable version. It will replace Falcon 9, the most successful launch vehicle in all of human history. Elon has said that, as soon as Starship is operational, he will shut down the Falcon 9 launch factory.

We've seen permission for 5 launches a year. They're asking for 25 launches per year out of Boca Chica, which is Starbase. He's gotten a lot of criticism for the failures of Starship. It still has not had a fully successful mission, but, oh my God, what he's been able to pull off is nothing short of extraordinary.

I hate it when he has this amazing launch, the Mechazilla captures the booster, which is the engineering feat of the decade, if not the century, and then Starship doesn't make it to a landing. Of course, the front-page articles are, “SpaceX fails again.”

Yeah. Crazy. We focus on bad news.

But the one thing that really bugged me about this particular article, Peter, was that if the FAA had them allowing 5 and now they're allowing 25, why do they care? Why do you have to get approval for the number of launches? This makes no sense to me.

It's environmental impact. They're incrementally trying to make sure—and I hate to say this—that they're not killing the number of turtles, or that the impact of—

Are you kidding? But once you've done a few, which he has, the environmental impact—there should be no reason. They should have nothing to say about you doing 10 more, 20 more, or 50.

They don't want unbridled, unrestricted operations.

That's hilarious.

That's all right. All right. You know the math as well as I do. It's hilarious.

Yeah. I mean, you did your 11-year penance there, Peter, with Zero-G.

I did my 11-year penance. It's like the FAA is not happy until you're not happy.

I do wonder, whenever you see a Starship blow up or whatever, whether a lot of the backlash is actually political. People are pissed off, and actually a lot of Elon's support base was very green-energy, Tesla-oriented, and so now they're all pissed off. But I think he's doing what he thinks is absolutely best for the world, regardless of the cost. I think it's impressive.

Starship will get us back to the Moon. It's the only vehicle to get us there. NASA's government projects, I don't think, will end up getting to the Moon. I think they're just too expensive and too risky.

Starship will get us to the Moon. Starship will set up orbital fuel depots. Starship will get us to Mars. It's an incredibly well-designed vehicle, and he just needs to continually iterate it.

People don't realize that when airplanes begin test flights, every test flight is a small iteration, and airplanes may have hundreds of test flights in their test-flight program before they get certified as safe for use. We need to expect that Starship will have hundreds of test flights, each one getting more data and demonstrating increasing margins and safety, until it's like, “Yep, this is going to be great to operate,” just like autonomous cars.

But unfortunately, these incremental test flights are very visible, and when they go wrong, The New York Times is here to talk about it.

It's explicit that they run dozens and dozens of experiments per flight because that gives them the maximum rate of learning, right?

Yeah, of course. Exactly. They're changing the margins and the material type. They're changing how they stage. They're changing their reentry characteristics.

All right, let's get to this fun article here. I'll read it from the BBC. “Quantum teleportation is here.” Quantum teleportation is the instant transfer of quantum information, not matter, between 2 points. It uses quantum entanglement, where 2 particles stay connected even miles apart. The first real-world test worked over 30 kilometers of standard fiber-optic cable carrying internet traffic.

Peter, I remember years ago all of the quant traders moved their fiber-optic nodes very close to the financial centers because nanoseconds matter to them.

Yep. Yep. So I think they're going to be using quantum teleportation here for instant communication capability. If we end up using it for stock prices, I'll be very disappointed. But your point's right.

Look, the amazing thing is that this is happening. I think the reality check is that this will take quite a while to commercialize, get reliable, and be used for real-world applications. But the fact that they're able to do it is mind-boggling.

Yeah, it's fun to think about. Why do we think it'll take a while to become real at scale, other than everything takes a while?

I spent some time at the Perimeter Institute at Waterloo, where the BlackBerry founders have given a huge chunk of money, and they've broken up quantum research into 3 chunks. One was pure computation, the second was communication, and the third was networking. You have to rebuild the entire stack of how you do it for each of those.

So there's a whole bunch of innovation and engineering that has to happen for all of those 3 segments to have reliable communications and teleportation, certainly at scale.

So there, it’ll take a while to get there. They’ve been going at this for, I think, close to 30 years, and they’re now getting to this point. So I think it’ll be a little while longer. But when it happens, holy moly.

The nice thing is, it’s over standard fiber optics, right?

Yeah, it’s not new equipment. We’re not needing to build out a new network, right?

Here’s another fun article in the realm of future-forward space: “Electron dance breakthrough: Scientists uncover clues to practical superconductors.” Researchers discovered that superconductivity varies at the atomic scale within individual unit cells of iron-based materials.

I won’t read the rest, but we’ve been talking about how one of the major benefits AI will give us is major breakthroughs in materials science. Superconductivity would be a huge deal because, at the extreme, you could put solar cells around the planet. If it’s nighttime here in the U.S., you could have solar energy being generated in Europe and have a superconducting cable back to the U.S. You’d basically get no transmission loss, right? It’s what, 30% or something?

It depends on the distance, of course, but it’s a huge amount.

I remember Shy Minez [?], one of our GSP Singularity students in 2010 or 2011, saying he figured we were about a decade away from understanding superconductivity and getting to room-temperature superconductivity at the lab level. This sounds like we’re coming close to that.

Well, if it’s a decade out, we’re going to have Gen III fission and fusion, and those are very small reactors, so they can be sprinkled around.

Yeah, yeah. So it may not matter on that.

There’s a big race between this and distributed energy systems that makes it less relevant.

Yeah, that’s right. All right, let’s talk about one of our favorite subjects. At least it’s a favorite subject for Salim and me. Actually, Dave, I don’t know: Are you a big Bitcoin believer? I know our third fraternity brother is, for sure.

Well, you know, I was on the board of MicroStrategy back in the day, so I’ve been a shareholder for many, many years. I wouldn’t say I was ever a direct Bitcoin early adopter, but Michael Saylor has dragged me along.

Let me ask the question: Do you own Bitcoin via MicroStrategy?

Yeah.

Yeah, you own MicroStrategy, but not Bitcoin directly?

No, not directly, mostly because I’d lose the keys.

Fair enough. I can make some introductions.

Here is Brian Armstrong, CEO of Coinbase, speaking about the U.S. holding Bitcoin. Let’s take a listen.

“I think it’s clear at this point that Bitcoin is a better form of money than gold. It is provably scarce, just like gold, but it’s more portable and divisible, so you can actually use it. It has higher utility, I would say. It was the best-performing asset of the last 10 years.

“For store of value, I think it’s going to be important for governments to hold this over time. It may start with being 1% of their reserves, but I think over time it’ll come to be equal to or greater than gold reserves.”

Okay, great. Salim, I want you to weigh in here. I think this is mind-boggling to me.

Let me give my reaction in a couple of ways. I spoke at a conference of 100 of the top partners of one of the biggest accounting firms in the world. When I asked them, “How many of you understand the Byzantine Generals Problem?” 1 put up his hand. You’re like, “This thing is going to eat your lunch in terms of trust and reliability, which is essentially what you sell.” You don’t understand the core innovation that’s taken place in it, which really is unbelievable to me.

When MicroStrategy put Bitcoin on the balance sheet, every CFO in the goddamn world should be looking at that and saying, “I need to put something in there.” It’s unbelievably irresponsible if you’re the Treasury secretary of some country not to put Bitcoin in as a reserve asset. This should be happening in every country in the world.

One of the things we’ve noticed before, where I’ve commented before, is that it’s fascinating seeing the amount of my portfolio investment growing because of Bitcoin. Not because I’m getting more Bitcoin, but because it’s just getting more valuable.

I was talking to a major money-management firm in one of the G7 countries, and they said, “Yeah, we don’t allow our wealth managers or wealth advisers to transact in Bitcoin at all. They’re not allowed to advise it to their clients because they deem it too uncertain and too risky.” You’re like, “Look at the goddamn value over time.” They’re scared of the legal liability if they recommend it. And I’m saying to them, “Do you realize the liability if you don’t recommend it?”

During the course of our conversation here, Bitcoin is now at $96,450. It’s gone up $1,500 during the course of our podcast conversation.

I’ll give you another case study. I’m the chairman and majority, or controlling, shareholder of 2 companies with about 1,000 employees. One of them has about $50 million in cash just sitting there in cash. The other has $125 million in cash. It’s a public company with $125 million in cash and making a lot of cash every quarter.

I gave both CEOs the green light: Put any amount of that you want into Bitcoin, and borrow and do what Michael Saylor did. Neither reacted or acted on it. I don’t want to interfere, otherwise I’ll be involved, but it’s amazing.

Well, I mean, then you’re obligated. They’re both great, great leadership teams.

I mean, we did the same thing, Salim. We were holding a large amount of capital, and it was like, “Yes, you know, both of you are on my board. I need you to help convince the other old-fogey board members.” Excuse me if you’re listening, guys. You’re brilliant and I love you, but can we please put our treasury into Bitcoin?

An easy thing to do there would be to take a small fraction of it and just let it do 1 cycle around.

Yeah, a small fraction, like 80%.

Yeah.

Well, it makes Michael Saylor all the more impressive that, without any role model, he just went ahead and did it. Now we have a role model, but without any role model, it was just entrepreneurship.

This is the part that drives me nuts now that we have a role model, right? If you’re the CFO of a public company, what the hell are you doing?

Yeah, it really drives me nuts.

You’re doing what you’ve always done because that’s the least risky, and you can’t get blamed for doing that. This is the same thing in all technology companies: It is the founder-led, exponential-tech, AI-first companies that are going to slay everybody else.

When Michael Saylor—by way of background, Michael was a fraternity brother with Dave Blundin and myself at MIT, at Theta Delta Chi, our fraternity—when you’re Michael Saylor, you can say, “We are doing this.” He has a very small board, and he basically sat them down, gave them homework, and got them to back him. He’s just been at it.

The first week of May, he actually sold equity and bought another $180 million of Bitcoin. Today, MicroStrategy owns basically 500,000 Bitcoin, which is about 2.8% of all the Bitcoin in circulation, worth $52.8 billion. Michael owns about 17,000 more Bitcoin than I do.

Oh, you’re doing well.

Michael not only had to do exactly what you said, but he had to go over to Mike Willis of Fidelity, another MIT alum, and say, “Hey, I need to custody this somewhere. Can Fidelity build out an entire business unit to custody Bitcoin?” To his credit, Mike Willis built it out at Fidelity.

That’s a lot of difficult work that’s now been done for you.

Yes, but pioneering it was very hard for a lot of reasons, including that one. You can’t just keep the keys in your shoebox at the office headquarters. You’ve got to have some kind of multisig protocols and all sorts of stuff.

I just checked the market cap of MicroStrategy at today’s close: $105.4 billion, while holding $52.8 billion worth of coins. The value it’s being given for his mindset and what he does is extraordinary.

I was sitting down with my mom, who’s in her 80s, going through her estate, and she had this spreadsheet she put together. There was this line item with a huge number in it. I was like, “What the hell is that?” She said, “Oh, you know, remember when you worked at MicroStrategy when you were in your 20s?” I said, “Yeah.” She said, “I bought some MicroStrategy back then.” I was like, “Wait, you’re still holding that?” And she said, “Yeah. Well, I don’t know how to sell it, so I didn’t.”

Go, Mom. That is so great.

That is fantastic.

Well, I think that’s our last story for today. It’s so much fun. Are you guys having as much of a blast as I am?

Oh, God, I look forward to this all week.

It’s really a great checkpoint where we can kind of go, “Okay, what happened?” bring some slides into play, and just take a checkpoint. I find it’s a really great cadence. Plus, all the viewers—I hear that, by the way, from a few of the folks watching. They’re like, “It’s really great because we can every week or 2 stop in and see what’s going on.”

We feel like we’re caught up. Yeah, that’s the goal here. So it’s fun. When we say, “WTF just happened in tech?” well, there’s a slice of it. We’ll see you guys in a week or two, when enough stuff starts mounting and we say, “Okay, we’ve got to get the band back together again.” Awesome. All right, Salim, where are you? Great conversation, guys. Where are you now? Are you back in New York?

New York? I’m back in New York for 3 days, and then I go to Italy for 3 days.

What is going on?

It’s just a busy time right now.

You’re insane, and I love you for it. And Dave, you’re back in Cambridge at headquarters.

I’m in Cambridge, and I’ll tell you one thing, Peter. I love every time I talk to Daniela Rus over at CSAIL, or Alex Wissner-Gross, or Alexander Amini over at Liquid AI. I can say, “Hey, we’ve got this WTF podcast coming up. What’s on your mind?” And they’re like, “Yes.” It’s such a good way to keep the dialogue. I love it.

So I want you guys piling in your best stories, those secret stories. You’ve got Alex Wissner-Gross—he’s a superweapon there—and, Salim, you’ve got an entire ExO community. So bring them in, and see you guys very soon. Thanks for your time.

Always a pleasure.

Great conversation, guys. Thanks, Peter.