OpenAI为何支付65亿美元打造新iPhone，以及Google如何刚刚终结好莱坞｜Dave Blundin与Salim Ismail｜EP #174

OpenAI to buy AI device startup from Apple's Jony Ive for $6.5 billion.

It's funny, in this entire AI battle, I've never been a Sam Altman fanboy until now. The first thing he does is launch a product that competes directly with Perplexity, tries to take over search, and kind of declares war against Google and search. The next thing he does is go and get Jony Ive and say, "Okay, now I'm going to start building devices."

Veo 3: grand-slam home run.

The future of media is going to completely flip, where it's on demand. This is what I want to see created on the fly for me.

I'm curious as to how people are going to use it at a personal level. But will I really use it to generate my own entertainment?

Hollywood is decimated. How does it continue to exist in 2 or 3 years from now?

Now that's a moonshot, ladies and gentlemen.

Hey, everybody. Welcome to Moonshots. I'm here with my moonshot mates, Dave Blunden and Salim Ismail, and this is an episode of WTF Just Happened in Tech. It's been a big week, guys. A big week of announcements, in particular with Google I/O and our friends over at Anthropic. Dave, where are you today?

Hey.

Uh, I'm in beautiful Wakefield, 10 miles north of MIT. Actually, this is a company I founded, uh, right after DataSage. Manages $2 trillion of... There, there's $50 trillion in the US stock market, $2 trillion of it trades and flows through this building. So it seems sleepy, but there's a lot of digital movement through this building.

It's good to know someone does that. And Salim, I know that you're in the bedroom about 5 meters that way. Salim and I are actually up in Tiburon. We were at an event last night that a friend of ours threw. Everybody's asleep because they got in at 2:00 in the morning.

We're in this building on the water here. It's gorgeous. It was a conversation on quantum and consciousness. It was pretty amazing. But we got up at the crack of dawn to do this podcast because this is an incredible week. Salim, are you doing okay? Are you awake?

A lot happened.

I'm awake, and I think it's an incredibly busy week—one of the busiest weeks we've ever had in terms of tech stuff.

Mm-hmm.

It is. All right, let's dive in. Let's talk about AI. That's the winner this week in terms of tech news, and I think it's going to be the winner every week going forward. Dave, would you clue us in on this news? "OpenAI to buy AI device startup from Apple's Jony Ive for $6.5 billion." That made headlines.

It's funny, in this entire AI battle, I've never been a Sam Altman fanboy until now. This is the move of all moves.

A huge coup.

Yeah, huge coup and a very aggressive move. It was probably in Sam's mind the whole time; he was just waiting for the $300 billion valuation so he could afford it. But he was talking about this about a year ago. At the time, he backed off, didn't want to wake up the competition, and said, "You know, the iPhone is probably the most perfect device ever made by mankind. So trying to compete directly with the iPhone and the Android phone is kind of crazy."

He was backing off while he was still planning, and now this is the move. The fundamental thought here is that AI is such a game changer that all the big tech companies need to have a direct-to-consumer device or interface. Meta buys WhatsApp. Meta buys Instagram. You just have to control the consumer front end.

Google builds Chrome. Google launches Android. Android doesn't even make money. Why does Google need Android? We need to control the data and the front interface that the consumer uses every day. Of all the big AI companies, Sam is the guy who most intuitively seems to get that.

The first thing he does is launch a product that competes directly with Perplexity, tries to take over search, and kind of declares war against Google and search. The next thing he does is go and get Jony Ive and say, "Okay, now I'm going to start building devices that the consumer interacts with directly."

A lot of the other foundation model companies are just trying to build foundation models and let the economy bubble around them. Sam is actually trying to control the high and low, get in front of the consumer, and control the entire AI backbone.

I think the big insight here is that you can build a consumer device that's AI-first, which means you're always talking to it. You're interacting with it like a person, and that's disruptive enough that a lot of the iPhone magic isn't really relevant. We can potentially leapfrog and bypass that.

Then, to spend $6.5 billion on a no-revenue idea, but get the guy and the team—that's very Steve Jobs-like. We can riff on that, but I think it's just brilliant.

The question ultimately is, what's the next-level device going to be? Is it going to be glasses? Is it going to be a device that's just listening all the time? There are a few different companies coming out with this. Salim, you were going to say?

Yeah, this gets us straight to that Jarvis thing: an agent that, from your perspective, is always looking at the world, listening, and making sense of the world. I think this is a huge step. As you said, Dave, very ballsy, and I love it.

Mm-hmm.

A very similar move: Google spent $3 billion on Character.AI specifically to get Noam Shazeer back. It's another case where you're willing to write a 9-figure check for fundamentally a single human being, but you know that person's going to bring a team with them, and they're going to bring that kind of Steve Jobs magic with them.

These price tags are completely unprecedented, but the upside on a win is also orders of magnitude bigger than ever before, so it's actually very rational. I can't wait to see what comes up next.

One of the questions, of course, is that we're going to have probably a whole slew of AI devices listening to your conversations all the time. I think it's going to get socially accepted eventually. But in the beginning, as you're having a conversation with your friend, you notice that they're wearing a pendant, or something on their head, or something on their wrist that's listening. The question becomes: Are you okay with every conversation you're having being recorded? Because that's where we're heading.

That's where we're heading. I think that podcast we had just a couple of days ago with Anish is directly related to this, too. Anish was pointing out that there are huge swaths of demographics, including older people and people in different countries, that have completely missed the iPhone and Android phone revolution.

The AI interface—the voice interface that's super-empathetic—opens it up. It's not like you have to stop everyone from using their iPhones tomorrow. What you need to do is open up whole new territories of users that were largely overlooked before. I think this is likely to penetrate very quickly that way.

It'll be a lightweight, easy-to-use, probably dirt-cheap device that you're just talking to, with an agent at the core doing everything you ask it to do.

The lurking is going to be interesting, where you don't know what's happening. Remember when Google came out with Google Glass and they were wearing them to bars? People gave them the term "glassholes."

They're getting punched and stuff. Yeah, getting beat up. It's going to be fascinating. We're about to see social norms for human-to-human interactions with AI there begin to get defined very quickly.

Look, it's started already. Every time I have a Zoom meeting now with a few people, all their AI shows up before they do. We have to go, "Wait, we're trying to have a confidential meeting here." Then we have to turn them all off.

The shift is happening. You're right, we could riff on that for a second. But on the regulatory front, there's basically no regulation. The regulators are completely asleep at the wheel.

If you look at TCPA, you can't tap phone lines, and there are very good reasons for that. But all of that is now moving to digital, where there are no laws whatsoever.

When Zoom calls started having AI spies on them, they announced themselves. Now they just listen, and you don't even know they're there. It's the same with consumer phone calls. A lot of our companies have either sales or customer support over the phone, and that's moving to AI voices at an incredible rate, using every GPU on the planet, by the way.

It works incredibly well. It's far better than a human call operator. But most of them don't tell you they're AI, and there's no regulation that says they have to tell you.

And so, a lot of the time when I'm listening to the calls, the consumer is aware they're talking to AI about 10% to 20% of the time; the rest of the time, they're not.

But that's amazing. Kudos to the designers there. That's extraordinary, and it's just going to get better and better.

All right, let's move on. We have a lot to talk about here. The bottom line is that Google stole the week, at least for a little bit. Dave, do you want to outline what we're seeing here? This is a slide titled “Gemini Leads Across All Models,” and we're talking about Gemini versus GPT-3045, DeepSeek, Grok, everything.

Yeah, this is “The Empire Strikes Back” right here. OpenAI has stolen the audience, the user base, and then AI primacy. xAI claimed it back in February, very briefly. Then Google came roaring back—or sorry, OpenAI came roaring back with new models, and that woke up Google.

Google is saying, “Okay, we have had this in the lab. We didn't know exactly how to announce it or when to announce it. But now that the competitive pressure's on, we're going completely to the wall here. This is the best of everything we've got, and we're going to roll it all out this week. And here you go.”

And it is mind-blowing. They're claiming the number-one spot in every single category.

Just read the categories across, which we'll get to in a second.

Yeah. The one that caught everyone's imagination was overall, obviously, image generation, hard prompts—difficult questions, I guess that's pretty obvious—coding and math. They claimed number one on that for all of a few hours.

They still own math.

Creative writing, the same as people doing their English papers. Long queries. They've actually had dominance in super-long context windows for quite a while; they just didn't tell the world about it.

And this is exactly what you were saying a couple of weeks ago, Peter: big companies have a real struggle with the innovator's dilemma. Even though this has been in the lab, Demis Hassabis and the team have been working on it for a long time, they've had it, but they were so worried about how to roll it out and whether there'd be consumer backlash. Now, under competitive pressure, it's unleashed.

I mean, this is Google over and over again. Google had Transformers first. It did all the fundamental work. It basically had the equivalent of GPT, but decided not to do what everybody said: “Don't put AI out on the open web. Don't make it accessible to everybody. Don't let it self-referentially program itself.”

And of course, once it was released into the ether, Gemini—or Google—had to come back and demonstrate itself. Listen, I love Google as a company. I'm impressed by them, and they've got an extraordinary base to build on. I'm sorry, Salim, you were going to say?

No. A few days ago, I generated an image of a Gutenberg printing press, and I said, “Give me a photograph of it.” I did that across several models, and the Gemini one was way better than all the others. It was incredible.

I think we're going to find over time that the interest is still with OpenAI. There's a lot of attention there, but I think over time, the sheer quality will come through.

I think if you love a good drama, this is the space race times a thousand. The players keep leapfrogging each other. You'll see later that Google's kind of ace in the hole is that it has hardware control—hardware design control—so you go from algorithm all the way down to chip design with the TPU v7s. Nobody else has that.

But the other guys have other competitive weapons we can talk about later in the pod, too, so don't expect that this is the end. In fact, I've seen “game over” said now about 20 times every time someone leapfrogs someone else. It's definitely not game over; the game has just begun. So if you like a good drama, this is just the next act, and we'll see that in a second.

And if any of you ever wondered whether this field is accelerating, it is massively accelerating, and it's the competitive pressures between these companies. It's no longer between countries; it's between a set of individual companies that are just trying to outdo each other.

The challenge is that ChatGPT—that is, OpenAI—captured the mindset of the world through a first-mover advantage. Kids throughout elementary school, high school, and college, and the average mom and dad out there, are all using ChatGPT, and it's reflected in the revenues that OpenAI has. Gemini really needs to catch up because I don't care how good you are; if your product is not being used, you're in trouble.

This is an interesting thing to watch because throughout history, even though we talk about first-mover advantage, it's never been the first mover that made the big bucks, right? Google wasn't the first search engine. Facebook wasn't the first social network.

OpenAI has a lead, but it remains to be seen. It might actually be a completely new competitor that completely changes the game for everybody.

Oh, it will be. It will. There's no question about it.

So, this chart that we're showing, if you're watching this on YouTube, is Google as number one across the board. But how many hours later, Dave, did we get this chart and Anthropic introduce Claude 4?

An amazing little 24 hours there. Sundar gets onstage and talks about something we'll look at in a minute, where you can use basically all AI all the time and not use regular Google Search. We'll see it in a second.

Google stock goes way down. Then, the next day, they come out with Veo 3, with a whole bunch of mind-blowing AI and number-one benchmarks. Stock goes up 7% just in reaction to that. So everybody's like, “What is going on here? There's so much chaos.”

Right in the middle of it, Claude 4 comes out. Dario is like, “Yeah, look, Sundar just claimed number one in coding. Not true. Well, it was true yesterday. Now it's not true. Today, we have Claude 4, and here are the benchmarks.”

And these are accurate, by the way. It really is the case that Google was ahead in coding for about 24 hours. Now, if you look at this chart, it's a pretty significant gap, actually. Claude 4 has just launched to the top again.

Those scores on the far left are incredibly meaningful to the people who understand what the implications are. For companies like Blitzy in our portfolio that do long-form coding—3 million lines of code in a single night—Brian Elliott is all over YouTube talking about it.

It works fundamentally because of the benchmarks on the far left. The AI can now actually build entire products while you're sleeping, and you come back and use them the next day, and they're functional. That app works when these benchmarks get around 80%. It doesn't work when these benchmarks are around 50% or 60%. There are just too many bugs in the code the next day.

So even though it doesn't look like much on this chart, it's an incredible tipping point in terms of functionality when you get above that 80% level on SWE-bench. Now we need to reinvent the benchmarks, by the way. This is SWE-bench, which is the most fundamental benchmark for how good the code written by AI is.

There's a new one coming out called SWE Re-bench, which is just making it harder and upgrading it. But we need new benchmarks now because the AI has basically broken the benchmarks.

When Ray talks about the singularity—that point after which you can't predict what's coming next—that's what you're seeing here, right? You're seeing literally a drop-the-mic moment over and over again, with everybody leapfrogging everybody else, and it's awe-inspiring.

All right, here's yet another part of the equation. We've talked about AI improving based on a multitude of different factors: better hardware and more capital coming in. This one is really improvements in algorithms—algorithmic improvements. The prediction here is we're going to see a 10X to 100X improvement in algorithms. Salim or Dave, why don't you comment on this?

This is something we've been tracking in, say, solar energy forever, right? The materials become cheaper, and lithium-ion batteries have dropped 90% in their price-performance over the last decade.

The same thing. People forget about this. They kind of think, “Look at electricity needs on the current models,” and then they project out, and everybody goes, “Oh my God. Oh my God,” not realizing that the models are becoming much more efficient—very radically, very quickly. As a real-world example, the amount of fuel needed for a plane to cross the Atlantic has dropped by 2/3 in the last few decades—

Really?

—just because the engines are so much more efficient. We know which routes to fly and where the air is less resistant, et cetera, et cetera. So I think we're going to see the same thing happening in AI, where the models are becoming very much more efficient, and that's going to change the game. Not enough to drive the upside, but it will definitely help hugely with the overall energy needs over time.

Yeah. I checked this post and researched it for accuracy, and there are 2 things to riff on. One is Jevons Paradox, which we all love to talk about. Actually, our good friend Erik Brynjolfsson is probably the top guy on the planet on that topic, and we should have him on as a guest to talk about this. And then the other is that this is way, way, way understated. So, starting—

Yeah.

—with the way it's understated, this is worth reading, but it was built by extrapolating from the bottom up. I know the actual innovations under the covers that are driving this, and they're much bigger than this, and they're much faster. Software innovation that's done by AI can be deployed in real time. You're not waiting for any hardware or anything to come online. You just deploy and go. So this will happen much, much faster than what's shown on this chart, now that the AI can improve itself.

And the other thing is, it's extrapolating to about 100X, anywhere from 30X to 240X, but I know that in the quantization of the neural nets alone, there's about a 20X to 40X improvement in that dimension. Then in the chain—

Just in that.

—of that reasoning... Just in that.

Okay.

These are all multiplicative, by the way.

Yeah.

So you have at least 3 dimensions that I know are 20X to 100X, and they're multiplicative with each other, so it's going to be more like 1,000X to 10,000X. There's a brilliant researcher at MIT, Shane Longpre, that I was talking to yesterday. He desperately wants to build a company around him, but he has to finish his PhD first, and it's frustrating as hell. He's just got a few more months to go.

He's doing a lot of great work with this. Right now, when you use an AI to do something for you—build a video or be an agent for you—the parameters in there also speak Swahili and know about quantum computing. All this other knowledge is actually using up the GPUs and the compute, even though it's not relevant to that question. So you can actually take much smaller subsets of that great brain to give you just as good an answer, and that's on the order of a 100X factor by itself. He's wrapping up his research on that right now, and then we'll end up trying to productize it. So, no doubt in my mind that this is way understated, both in time and in impact. It's pretty exciting.

Bottom line, when Eric Schmidt says AI is massively under-hyped, it's the realization that we have all of these knobs we can turn, and they're going to give us improvements—each one of them 100X improvements, times 5 or 6 different knobs. It's insane.

Let's talk about Jevons Paradox for just a second there, too.

Yeah, please. Go ahead.

Because I think no one had ever said Jevons Paradox, I think, for 100 years, until now. It's going to come up every week, so we might as well all understand it.

Well, define it for us.

All right. Here's the deal. In normal economics, Erik Brynjolfsson is the master of this. In normal economics, you have supply and demand, you know, IS-LM. I don't know if you ever took 14.02 at MIT. It's like—

I did.

So if you said, “Hey, I found a way to grow bananas. They're twice as cheap now. They cost half as much,” people don't consume twice as many bananas because they're half as expensive. They consume maybe 5% more bananas. So that's normal economics, right? Price comes down, and that fraction of the economy is smaller because the price is lower, oil being the classic example of this.

Then you get into tech and something really weird happens. If I make a GPU 10 times faster for the same price, my video game got better. Do I use 10 times less video game? No. No, I actually use more video games than ever before because it's more fun than it was before.

So even though it got 10 times cheaper, my overall consumption of that, actually in dollars, goes up, not down. That's completely paradoxical for an economist. You're seeing that most acutely in AI, more than ever before.

If it is more capable because you made the model, in this case, 30X to 240X more efficient, doesn't that bring the need for GPUs and data centers down? It's not. It's going to go the other direction. The need for data centers, chips, and electricity actually goes up very, very steeply. A lot of people are mispredicting the investment in energy, in data center supply, and in chips. Chips will be sold out for at least the next 5 years, and maybe infinitely in the future, for exactly this reason.

Because the capabilities—like one of those Veo 3 videos, which we'll look at in a minute—

Yeah.

—and you'll see immediately, as soon as you do it, you want more. Every time the capability gets better, you want more.

All right, let's dive into Google I/O. What we've done here is grab a 10-minute video. It's a recap put out by Google, and it's a Google I/O recap in 10 minutes. We're going to play this at a little bit of accelerated speed, and I'm going to have my moonshot mates here weigh in and comment as we go along.

It's very rich here. It's important that people understand what Google just announced. Google remains the dominant player here, and there's a lot. Okay, so we'll go to Sundar, who opens it up, and we'll break at every occasion for some fun conversation.

Hello, everyone. Good morning. Welcome to Google I/O. We want to get our best models into your hands and our products as soon as possible. I'm particularly excited about the rapid model progress, and today, Gemini 2.5 Pro sweeps the LM marina leaderboard in all categories. It's getting a lot of love across the top coding platforms, thanks to all of you.

Gemini 2.5 Pro is our most intelligent model ever, and now that it incorporates LearnLM, our family of models built with educational experts, 2.5 Pro is also the leading model for learning.

Right. One of the things we've talked about a lot is using this to re-educate the planet, and I think we're going to see a massive disruption in education across every field. One of the challenges we're going to have is whether the educational field is going to accept this. I don't know how you guys feel about this, but teachers' unions are going to have to get in the game here. We're going to start to see: do I put my kid in school because they're using old-style, or do I use AI to educate my kids? Thoughts?

Well, one high-level thought on that is Demis Hassabis is going to push this agenda. He's taking the high ground. It's really interesting to watch Google now decide, “Look, our Nobel Prize-winning brilliant guy—we're going to put him on stage and make him a centerpiece.” And he rises to the occasion incredibly well.

But he can't compete with Sam Altman in the use cases that are going to propagate at light speed: virtual friends, creating your own internal copyright-violating versions of movies, all that stuff. He can't touch that because he's a Nobel Prize winner. He's going after saving hundreds of millions of lives with biotech innovations because he's the AlphaFold guy.

Yeah.

So he has to take this incredible high moral high ground. On stage, he's very focused on new models that work from fundamental physics principles, have fundamental scientific innovations, and teach the world everything they need to know. It's incredibly high-ground, important use cases that are going to take a long time to percolate out because of, like you said, teachers' unions, regulation, and so forth.

While Sam, on the other end of the world, is just going to run away with consumer, whether it's spyware or not, we're doing it. We're moving ahead of time. This is the quandary that Google's in. Sundar doesn't want to be that guy. Demis doesn't want to be that guy. This next woman who's going to talk doesn't want to be that person. It's just big-company versus small-company mentality everywhere.

All right, let me jump back in, and then, Salim, I'll come back to you next.

So let me pause at Google Beam here. Salim and Dave, we had a chance to see Google Beam at the Abundance Summit this year, and my 12- and 13-year-old boys were with me. They said it was the coolest thing they saw at the Tech Hub, right? The ability to actually be there. Cisco tried this years ago, and it was expensive, multimillion-dollar hardware. I don't know if the price point has been announced yet for Google Beam, but it's going to be an appliance that's in your home when you want to contact your mom or your grandma.

It'll be very cheap.

Yeah, for sure.

It'll be very cheap, and I think this is a very big deal because you can think about the language potential that comes with it. All the other capabilities will come with it. I think it'll up the game for companies like Zoom and force them into a new innovation cycle also.

Fascinating. Agreed.

Yeah. We need to talk about price point for a second here, too, because a lot of this stuff is just mind-blowing, and you need to experience it. Then Google says, “Well, to access anything I'm showing you here, you gotta get a Google Ultra or Gemini Ultra account for 250 bucks a month.” So, oof.

That's a ballsy move on their part. Ballsy move.

Ballsy. One of the things I love about this podcast is that we're trying to improve people's life choices and trajectories through life by giving them as much actionable information as possible. A lot of people who listen to this podcast are going to say, “I don't know. 250 bucks a month? That's like rent.”

Yeah.

You can't do it justice.

But Google's trained us—Google's trained us to get everything for free, right? And just give up our rights and such. I mean, OpenAI has proven a $20 price point is acceptable to most individuals. I'll give up a couple of lattes and have access to the world's intelligence. 250 bucks, that's—

I think it'll become mandatory for anybody in the professional modality. These tools will be so rich, and there'll be enough there for them to do quite well out of it, I think.

Mm-hmm.

But 6 months ago, a lot of people were saying, “Look, foundation models are a race to the bottom. They're a commodity. The price points will come down. This $300 billion valuation for OpenAI is insane, given that it'll commoditize.” Now it's gotta be obvious to everyone that all of that was wrong. It absolutely could not have been more wrong.

Because you're a foundation-model company, the capabilities that you can invent and launch, and the price point that they'll command, are astronomical. Right now, any of our listeners are basically going through what's inevitable: “I can't be competitive if I don't have my hands on this. I need to try it. I'm gonna drop behind society if I don't get on top of this.” Well, sorry, it's gonna cost you 250 bucks right away.

I noticed that it took my money instantly, but it took 24 hours for me to make my first Veo 3 video. I interpreted that as it's oversubscribed and overloaded. So it's obvious—

Well, let me just point out—

One thing, one thing, Dave, which is, the capabilities that existed before—when I say before, last month, last quarter—will be available for free, right? We're gonna still get a lot of fundamentals for free. But if you want the cutting edge, that price will be increasing over time.

Mm-hmm. Let's jump back in here.

We are introducing this real-time speech translation directly in Google Meet.

It's nice to finally talk to you. Es bueno hablarte finalmente. No puedo esperar alquilar tu casa.

So instant translation—we've been seeing this for a while now. I think it's gonna be fascinating to incorporate across all the video platforms. Right now, YouTube is translating our podcast in every language out there. That's fantastic. Zoom's gonna have to catch up to Google Meet. Any comments on translation, Salim?

I travel a lot, and this will be so great: to hold up my phone and do real-time translation conversations in different countries. I'm super excited about this. There's such enablement that comes from this. We've been kind of tickling at it. It's been available in a clunky form for a long time. I think this sounds like it gets to a level of seamlessness from a user-interface perspective that could change the game.

So, Salim, this directly affects our side bet, which we need to finalize, by the way. Peter needs to adjudicate this side bet that we agreed to a few weeks ago. I see the forces of concentration of wealth into just a few hands are overwhelming, and the forces of democratization are moving much more slowly. Your side bet was, “No, this is gonna benefit broad swaths of humanity.” We haven't quantified that, so we have to settle that, but I'm willing to put any amount of money you want behind it. Here, it changes the definition—

Well, you talked about trillions before, so you can put down trillions, and I'll put down a few bucks. We'll see how that goes.

All right. We'll call for that.

I want odds.

A Bitcoin exchange rate.

But it's interesting: when you listen as a parent, in the past you would say, “Oh, it's really important for my kids to learn French or Greek or Turkish or Japanese,” whatever your cultural roots are or whatever you think. “You need to learn Mandarin because China is rising in the world.” And then you need to learn coding, right? Coding was always like, “Push your kids to learn to code.”

So now we have this digital world allowing us to do coding through natural language, provide coding, and why would I ever learn a language if I can just be translated instantly? There's an interesting social reengineering going on here. Of course, when you learn to code, you learn how to think in a different way, and when you learn languages, you learn culture in a different way, which is gonna be obviated through here.

I'm fascinated by this.

Well, there's something that will, I think, persist here, and the reason is that when you learn a new language, your brain rewires fundamentally differently.

It does.

It's the same reason so many creative people play music or are into music: their brains rewire fundamentally around a different paradigm, and you triangulate now. You've got different neural circuits that you can bring to bear, right?

In India, when they have countrywide maths competitions, it's always won by a South Indian. Their ability to do mathematics is 10X greater than everybody else in the north of India, and they've worked out that it has to do with the structure of the language. The rhythmic patterns and the structure make their minds naturally available for that.

Hmm.

So I think the idea of learning piano or learning coding or learning certain deep skills or languages will persist just because it'll create so much more creativity in people, and more creativity will be the only thing that's available for humanity going forward.

Well, the reason I think this affects our side bet, Salim, is that there's latent talent all over the world, and it's usually locked up behind language barriers.

And now that the translation is real-time, it also keeps your voice intonation, so with your natural voice, you can actually make a real friend in a different country where you don’t have a common language. I think that’s new in the world. We don’t know where it’s going to go exactly, but all of our successful startups are dominated by this theme of best friends grinding it out, working 24/7—

Oh my God.

That’s brilliant, Dave. There’s so much talent in Pakistan, in India, and in parts of Southeast Asia that are locked up and don’t have access to our capital markets. We have Link Studios as part of Link Exponential Ventures. They’re on the campus of MIT, or adjacent to it.

Mm-hmm.

And we’re opening up Abundance Studios here in L.A. You can imagine tapping into that deep bench and having natural conversations like we have with our buddies down the street.

Ooh. So can we just say that I won the bet right now?

We can.

I think I’ll scream.

I’m jumping back into Google I/O. We’ve got a lot to cover here.

All right.

Okay.

Our research prototype, Project Mariner, is an agent that can interact with the web and get stuff done. We released it as an early research prototype in December, and we’ve made a lot of progress since. We are starting to bring agentic capabilities to Chrome, Search, and the Gemini app.

Let me show you what we are excited about. We call it Agent Mode. Say you want to find an apartment for you and 2 roommates in Austin. You’ve each got a budget of $1,200 a month. You want a washer-dryer, or at least a laundromat nearby. Using Agent Mode, the Gemini app goes to work behind the scenes.

You might be familiar with our AI-powered smart reply features. Now imagine if those responses could sound like you. That’s the idea behind personalized smart replies. With your permission, Gemini models can use relevant context across your Google apps in a way that is private, transparent, and fully under your control.

Let’s say my friend wrote to me looking for advice. He’s taking a road trip to Utah, and he remembers I did this trip before. Gemini can do almost all the work for me, looking up my notes in Drive and scanning past emails for reservations. Gemini matches my typical greetings from past emails, captures my tone, style, and favorite word choices, and then it automatically generates a reply.

Wow. Just wow. It’s like Agent Mode there and Google’s ability to personalize in a very deep and meaningful fashion. I’m blown away.

No, listen, the fact of the matter is their goal is to get everybody to switch over from iPhones to Android and from Outlook to all of Google Suite. It’s getting compelling. The cost of switching platforms is huge, and if Google’s able to provide enough eye candy and enough cognitive candy to get me to switch, that’s a big deal.

The use case they demo is just so funny to me because Sundar’s like, “Hey, suppose I want to go to Utah and I want to rent a bike.” Come on, man. We already did this week here at Vestmark: we took that same A2A, agent-to-agent, and MCP capability—

Yeah.

—and we moved it inside our firewall. It now opens inbound emails in Outlook with trade requests. Then the AI will read the trade requests, automatically trade the account, generate the trade file automatically, and route it out.

We’re already using that exact capability in a much higher-leverage use case, but we completely built the AI operator. The financial services industry spends about $100 billion a year on back-office operations related to these activities, and when we sample them, every single one of them is possible to do with AI now if you’re willing to connect it to sensitive information and financial accounts, and you find a way to put it behind your firewall.

They just rolled that out here this week, and it’s such a huge unlock. But when you get on a stage like this, you’ve got to talk about something like, “Hey, I’m renting a bike.”

A bike.

And I think, okay.

Well, why does Sundar sound like an AI? I mean—

His name is Pichai. It is PichAI, so that’s why they chose him, I think, to be CEO.

Oh, no. That’s terrible. But Sundar is in the maelstrom, right? And he’s got—

Yeah.

Sergey and Larry looking over his shoulders. He’s got the innovator’s dilemma again. Their entire revenue base has been Google Search—not their entire revenue base, two-thirds of their revenues. They’ve got YouTube, which is incredible. They’ve got Waymo.

But switching things over—and we’ll see this in a minute—switching this over away from Search, they have to do this in order to continue to survive and grow. Any comment on the agent element, Salim, or the personalization?

Does it feel weird to have the AI searching all your emails and responding on your behalf? I find that incredible.

The responding on our behalf is a little creepy and a little unnerving, right? You don’t know what it’s going to say, et cetera, et cetera. I think it’s just one of these things we’ll get used to, and it’ll become routine fairly quickly. The initial emotional reaction will be there for people.

I think the potential is much, much bigger here on the business side than on the personal side. The B2B implications of this type of stuff are off the hook for customer service responses and all sorts of things. So I think we’ll see a lot more agentic AI stuff at the enterprise level than at the personal level.

Yeah. All right, why don’t I continue—

Yeah, there’s so much to talk about.

Go ahead.

I don’t want to go forever.

This is rich. We could spend days on this stuff. But I wonder if you say, “I want to hear it.”

On the consumer side, everybody worries about privacy, but then they roll out functionality that you absolutely have to have, and you have no choice, so you always check the box. How many times a day do you push the Accept button and not even know what you just accepted?

Privacy is long since dead. Come on.

Yeah.

People want privacy, but your Alexa is listening, your Siri is listening, everything is listening all the time. Privacy is a great thing to desire, and I don’t accept that it’s true. Do you?

Well, this is why the truly brilliant Demis Hassabis types and Mark Pincus, our friend, need to get involved in regulation, which is probably the last thing they thought when they were kids growing up: that they’d get involved in regulation.

At the rate these capabilities are coming out, if people like that don’t get involved in regulation—and I mention those guys in particular because they’re genuinely concerned, and they really want a functional—

Yeah.

—non-dystopian future, we have to use AI as a tool in creating the regulation on AI. There’s no other way it’s going to keep up.

It’s heartwarming that people like Demis are at the top of the stack, because they’re just incredibly humanist—

Good human beings, yeah.

—good-natured people.

Yeah.

So that’s encouraging. But on the government side—

On the government side, this is a Rubicon we crossed a long time ago. In the U.S., there is no right to privacy anymore. It’s been totally dissolved by technology.

The best framing I’ve seen for this is: consider that you’re living in a global airport. In an airport, you know you’re being surveilled, and your rights can be taken away at any time. Essentially, that’s translating now down to every aspect of life in every country.

Yeah. No, you’re dead right. But it’s so convenient to give your AI access to your conversations and your emails. It’s going to make the world automagical for you. Turning that off will be like turning off half of your cognitive capacity.

Remember, Peter, 10 years ago at Singularity, we played that video of Mark Zuckerberg. It was an Onion clip of Mark Zuckerberg, and they said, “Wow, that was the best CIA agent ever. He got the whole world to just tell us where they were. We don’t even need spies anymore because everybody just tells us where they are. So Mark Zuckerberg, here’s an award for greatest CIA agent in history.”

I think we’ve been going down this road for a long time, and I don’t think there’s any going back from it.

Yeah.

Well, I had meetings with 2 of the biggest insurance companies, top Fortune 50s, earlier this week, and I said, “Look, you guys, how are you going to act on this? You need to act on it tomorrow.” One of them wants to use AI to do their quarterly closing.

So you can't talk about it. It's a public company, so there's nothing more sensitive in the world than your quarterly closing data. They said, “Well, we have 2 choices. Microsoft says, ‘Just give it all to us. Trust us. We've already got all your emails in Outlook, so just throw it over to us and trust us.’” They're like, “Yeah, that sounds kind of scary, but maybe.”

The other choice is to take Llama 4, distill it, and bring it in-house, inside their firewall. That's startup heaven right there, and very, very viable as a solution. They just need a vendor to come in and wire it up for them. So I think a lot of the companies and countries that care about their future are going to take that latter path.

But the alternative is to concede every aspect of your life, your data, and everything to either OpenAI, Microsoft, Google, or xAI. It's just not viable for sovereign states, or for a lot of fintechs, to take that path. So it's an interesting, very dynamic time.

I am jumping back into the Google I/O video here.

Okay.

All right. Let's do it.

Gemini Flash is our most efficient workhorse model. The new Flash is better in nearly every dimension, improving across key benchmarks for reasoning, code, and long context. Flash will be generally available in early June, with Pro soon after.

Gemini Diffusion is a state-of-the-art experimental text diffusion model that leverages parallel generation to achieve extremely low latency. The version of Gemini Diffusion we're releasing today generates 5 times faster than our fastest model so far.

This is our ultimate vision for the Gemini app: to transform it into a universal AI assistant. In Project Astra, we've upgraded voice output to be more natural with native audio. We've improved memory and added computer control.

We are going to show a video example of Project Astra. I mean, listen, Jarvis is here. My ability to talk to Astra and ask it to do things for me, research things for me, and show me where something is in the room—I mean, it's insane. We'll come back to this because the example video is so compelling.

A quick aside, you probably heard me speaking about Fountain Life before, and you're probably wishing, "Peter, would you please stop talking about Fountain Life?" And the answer is no, I won't. Because genuinely, we're living through a healthcare crisis. You may not know this, but 70% of heart attacks have no precedent, no pain, no shortness of breath, and half of those people with a heart attack never wake up. You don't feel cancer until stage III or stage IV, until it's too late. But we have all the technology required to detect and prevent these diseases early at scale. That's why a group of us, including Tony Robbins, Bill Kapp, and Bob Hariri, founded Fountain Life, a one-stop center to help people understand what's going on inside their bodies before it's too late, and to gain access to the therapeutics to give them decades of extra health span. Learn more about what's going on inside your body from Fountain Life. Go to fountainlife.com/peter and tell them Peter sent you. Okay, back to the episode.

The Jarvis, if you will, is finally here. Thank you to Google for providing it.

To more people than any other product in the world. In our biggest markets, like the US and India, AI Overviews are driving over 10% growth in the types of queries that show them. What's particularly exciting is that this growth increases over time. We are introducing an all-new AI Mode. It's a total reimagining of Search. With more advanced reasoning, you can ask AI Mode longer and more complex queries.

We're excited to start rolling out AI Mode for everyone in the US starting today. Over time, we'll graduate many of AI Mode's cutting-edge features and capabilities directly into the core Search experience. That starts today, as we bring the same models that power AI Mode to power AI Overviews. So you can bring your hardest questions right to the search box.

Okay. Big move on Google's part. Very, very ballsy. A bet-the-company move. It would be great to overlay the stock price with this exact timeline because this is exactly when the stock started plummeting. It clearly cannibalizes the core business. It's incredibly impressive that they're willing to do that.

Then right after that, they rolled out all the incredible AI capabilities. They got it all back and more the next day. So it actually paid off as a strategy. By the end of the week, it worked as a strategy.

Everybody knows OpenAI is running away with the actual AI user base, and Google is miles behind. But now they're going to claim that anyone doing a Google search that turns on this mode is an AI user, so they're actually ahead of OpenAI. That's pretty smart and aggressive, too. It's interesting.

I hear my family members and friends say, “Oh, yeah, I ask Chat.” They have a personal relationship with ChatGPT. They've given it a name. You need to personalize these AIs if they become part of your everyday life. I guess I had a conversation with Gemini as to where it's going to go. Salim, what are you thinking about this?

I think kudos to them for doing this. It would be really easy to hunker behind the old search box and just leave it the way it was while they experiment. But to make something this aggressive, where it's built right into the core product that they have, is incredibly courageous. I agree they don't have much of a choice, but many other companies would not have done this.

So let's see where this goes. Or they may have waited. I think if you step back from this, too, the reason America works, fundamentally, is because startups innovate and push the agenda. But this capability would have been buried inside Google for years—maybe forever—without the competitive pressure.

So it's a great little case study in what we need to preserve in the American economy: exactly this dynamic, where the big guys only move if we keep funding the small guys. If the IPO economy and the venture economy ever fell apart, this whole country would just grind to a halt. This is a great case study, isn't it? I love it.

Well said.

All right. Let's jump back in. Let me make a point related to that. If Google was a European company, this never would have happened. All right, let me find where we were.

It's like having my very own sports analyst right in Search. Search figured out that the best way to present this information is a graph, and it created it. Complex analysis and data visualization are coming this summer for sports and financial questions. Using my camera, Search can see what I see and give me helpful information as I go back and forth in real time. We're bringing Project Mariner's agentic capabilities into AI Mode. Search can take work off my plate while still under my control. Search helps me skip a bunch of steps, linking me right to finish checking out. Tickets secured.

With AI Mode, we are bringing a new level of intelligence to help you shop with Google. Search dynamically generates a browsable mosaic of images and some shoppable products personalized just for me. To create a try-on experience that works at scale, we need a deep understanding of the human body and how clothing looks on it. To do this, we build a custom image-generation model specifically trained for fashion.

All right. Here we go, Google. Google's getting into Amazon's business—instant shopping. It's going to be an interesting overlay there. Dave? This feels like they're just throwing everything against the wall. Just throw it all out there.

I'll tell you the dirty little secret behind this, though. Google's search volume went flat back in 2017. They stopped reporting it years ago because it's flat, but the revenue just keeps going up and up and up and up. They say, “Well, look, we don't show any more ads than we ever showed before.” But what they actually do is take all the ads off the low-value searches, so baseball scores would be a great example, and amp up the ads like crazy on the topics that drive all the revenue.

That's auto insurance, mortgages, jobs, and travel. Just 1 or 2 other categories drive all the revenue. So now, as they roll this out, notice the examples they give are baseball scores. There's no revenue there anyway, so move that over to AI.

The other one is shopping, where they got crushed by Amazon. Something like 60% or 70% of all shoppable product searches go straight to Amazon. You don't even start on Google. So they're not winning that war anyway.

They're basically going to claim AI supremacy and try to move only the categories that are low-click-through-rate revenue over. They're very good at optimizing that balance internally, so this is their way of trying not to cannibalize too much while claiming more users than OpenAI has. And you know, there's something—

I think they're going to go after Amazon here on purchases, right? I mean, if I'm in AI Mode with Gemini and I'm talking about things, it pops up and says, “By the way, you can purchase it here.” Seamless shopping in the middle of conversations or in the middle of doing research, without having to shift over to Amazon, is going to be an invasion of Amazon's revenue engine.

It's a great point. And actually, in all of this PR war that's going on, with Claude coming out right on top of this event and everyone jumping on each other, Amazon is notably silent, even though they do have a ton going on inside AWS. I know they're working on it; they have a plan for sure. They just don't seem to be fighting the PR war.

Yeah, and then Apple's just given up altogether.

That is so scary, and it's so weird.

It is so weird. All right, let's jump back in here.

This is the future of Google Search, a search that goes beyond information to intelligence.

Our goal is to make Gemini the most personal, proactive, and powerful AI assistant. Gemini Live now includes camera and screen sharing. All of it is rolling out free of charge in the Gemini app on Android and iOS today. We're bringing our latest and most capable image generation model into the Gemini app. It's called Imagen 4. The images are richer, with more nuanced colors and fine-grain details. Today, I'm excited to announce our new state-of-the-art model, Veo 3. Veo 3 comes with native audio generation. That means that Veo 3 can generate sound effects, background sounds, and dialogue.

They left behind a ball today. It bounced higher than I can jump.

What manner of magic is that?

This ocean, it's a force, a wild, untamed might.

All right. Veo 3—grand-slam home run.

Mm-hmm.

Oh my God, extraordinary. Dave, you started playing with it when you got your $250 account. How easy was it?

Yeah. Oh my God, it's—well, it's trivial. You gotta get the account and just play with it. It's instantly mind-blowing. The integration of the audio, I think, really puts it over the top.

Mm.

But it's immediately obvious to you that the future of media is gonna completely flip, where it's on demand. This is what I want to see created on the fly for me. Every time you do a Veo 3, it's clipped at 8 seconds, and you immediately want a full feature-length movie. And you're like, “Why can I only get 8 seconds?”

Yeah.

The answer is there aren't enough GPUs on the planet; there isn't enough compute on the planet to meet the demand of everybody who wants this instantaneously. And that gives you a sense of how many data centers like Chase Laqueller Carusos are gonna succeed, because just that use case alone will use up everything, let alone customer service and code writing—all of these things combined. But it's amazing that Josh Woodward got this unit. This is the keystone. This is it.

Yeah. Josh—

And Peter, your buddy Josh got it.

Yeah, Josh was on stage with us at the Abundance Summit this year. He's a great presenter, but it was awesome. Salim, what are your thoughts on Veo 3?

I think it'll be interesting as people start. I'm curious as to how people are gonna use it at a personal level. There are obvious business uses, et cetera, but will I really use it to generate my own entertainment? I'm not sure I would. We watch movies particularly because people have curated an experience for us, right? So I think this will be great for professional filmmakers, and it'll be great for making pilots. All sorts of stuff will come through with this. I don't think it'll affect the end consumer that much.

I don't know about that.

Oh, I'm with you on that.

I mean, listen, Gemini 3, Gemini 4—one of these will have looked at every movie I love. It'll understand the context and nuance of what I really enjoy. If it's got a camera facing me, it'll understand when I'm smiling, when I'm frowning, when I'm reacting to it, and it can generate a— Imagine you want the next season of a particular TV show, right? I'd love to see the next season of the original Star Trek. You'll have that.

That's a great example, Peter, because every Netflix series, From the Earth to the Moon or whatever, you always want more when you get to the end, and they just physically can't create more. Well, now they can. And whether it's passive or active—real time and created for you—or it's just the producers can now create the episodes that much more quickly and cheaply, you're never gonna let an audience like Harry Potter's just drop. You're never gonna let the audience just drop. “Oh, this is the last book, this is the last movie”—disappear. That's just crazy inefficient, to just let the audience fall off a cliff. So now that'll probably never happen again. Whether it's because production costs are much, much lower or because people can create their own, either way, that trend will happen. But I'd love to side bet with you on this one too, Salim. I think that movie producers have always dwelled on this concept that people are fundamentally lazy, but I don't think that's true.

I think the medium—the passive medium—if you go into a dark theater on a sunny day in LA and you come out of that dark theater and the sun's blaring, you feel like crap. You love the movie, but you feel like, “Where am I?” The ability now to keep the dialogue going—you can move around and keep it going, and you can guide it in the direction you want—that's gonna be so compelling for people. I don't think the producers have mastered the new medium yet. But when they do—oh, yeah.

I just Googled this, and it says: “What's the average shot time for a movie today?” A modern movie's average shot length is between 2.5 and 6 seconds. In an action film, it's 2 to 3 seconds per shot, and in a drama or art film, it's 8 to 10 seconds per shot. So if there are any physical limitations here, you're just gonna cobble these together. But Hollywood is decimated. I drive by all the studios out in Hollywood when I'm in LA, and how do they continue to exist 2 or 3 years from now?

Yeah, you're dead spot-on on that too, because look at Bollywood and how successful that's been. Why? Well, it's because the movies are culturally aligned and it just aligns with the audience. But why did all the movies pop up in LA in the first place? Well, it's because we need to shoot the scenes outside in good weather. It's way too expensive to move all the equipment inside every time it rains in New York. So we got a setup out here where we have good weather. But now this kind of content is weather-independent, global, and language-independent, so it's just talent-constrained. So, yeah, you're right. It should be democratized and moved across the world very quickly now.

All right. Let's jump into Google I/O. We're three-quarters in, with one quarter left to go.

Okay.

Our new SynthID detector can identify if an image, audio track, text, or video has SynthID in it. Based on our collaboration with the creative community, we've been building a new AI filmmaking tool for creatives. We're calling it Flow, and it's launching today. Let me show you how it works. These are my ingredients: the old man and his car. We make it easy to upload your own images into the tool. It lets you extend a clip too, so I can get the perfect ending that I've been working towards.

But what about emerging form factors that could let you experience an AI assistant in new ways? That's exactly why we're building Android XR.

Right now, you should be seeing exactly what I'm seeing through the lens of my Android XR glasses, with my delicious coffee over here and that text from Sharon that just came in. Let's see what he said. All right, it's definitely showtime, so I'm gonna launch Gemini and get us going. Gemini, what was the name of the coffee shop on the cup I had earlier?

Hmm, that might have been bloomsgiving.

Okay. Gemini, show me what it would take to walk here.

It'll take you about an hour.

Okay. I can get some steps in, and these heads-up directions and a full 3D map should make it super easy.

I'm excited to announce today that Gentle Monster and Warby Parker will be the first eyewear partners to build glasses with Android XR.

All right.

I would give anything to be at the next Apple board meeting because you've got Sam Altman saying, “I'm getting Jony Ive. I'm gonna do devices,” and I've got Google saying, “Apple, get out of the way. We've got much better AI-driven...” And what's next? Oh my God. They—I mean, this is where Apple should pay the price. They haven't done anything to even try.

Yeah. Crazy. Salim, XR glasses—it's about time that we've got some good AR up and running, but it's gonna change our behavior, right? So every day as I'm walking down the street, do I have education mode on? Do I have gameplay mode on? Do I have entertainment mode on? I mean, it's gonna become the major educational partner if we let it. Warby Parker coming in—maybe they'll look good, maybe they'll look fashionable, maybe you won't look like a geek if you're wearing them.

I don't think anything can help you not look like a geek when you're wearing those glasses.

But I was playing with the Meta glasses, Peter, that you handed out at—

Yeah.

A360 last year.

Yeah.

They were amazing. They look like good sunglasses, they act like good sunglasses, and they've got all this AI capability built into them. It was really quite fascinating to see. That was just version 1. The next versions are going to be 100 times more capable, with AR built into the middle of it. I think this is going to be quite a game changer. It will be really amazing to see the use cases, and the potential utility is off the hook.

All right, let's hit a few more. I'm going to exit out of the Google I/O summary there. A lot of amazing stuff. And kudos to Sundar and Sergey Brin, the whole team there—

Of course.

—for the work they've done. They really came out on top this week. If you look at Polymarket—who's going to be leading AI by the end of May—the Polymarket results show Google at 80%, Anthropic at 19%, and xAI at 1%. Okay, seems rational and reasonable. Then let's take a look at what Polymarket predicts by the end of 2025. Google is still on top at 38%, OpenAI is back at 26%, and xAI is at 23%. That makes sense to me. This is a battle between the two of them. Then, of course, you've got Elon coming out of right field. Never bet against Elon—that's my mantra here.

Yep.

Any comments on this, Dave?

This is the most entertaining thing in the world to watch. You said Google's back on top, but look, they were number 3 in March on this chart. So everyone had written them off. You can actually put money on this if you want to make a living out of betting on who's leapfrogging whom.

At this stage, despite this chart, it's just a 4-horse, all-out race, with everyone going as fast as humanly possible. I don't think the short-term trend that vaulted Google to the top is it. It's going to flip many more times over the next year among these 4 players.

But there's this level of intense energy we haven't ever seen before.

Mm-hmm.

And it's what's accelerating our AI future.

It's also the most important race in the history of human innovation. The implications are so much bigger than the past races we've seen. The amount of money, the amount of energy, the short timeline—if you want to track one thing and enjoy it in your lifetime, this is it.

I'm going to hit on a couple more Veo slides. We saw this slide here, but you made this one, Dave. Tell us about it.

Well, it's my very first-ever Veo 3, so I just wanted to memorialize it and capture it.

Okay, let's play it.

Yeah, I don't know. It's just—

Your prompt here was—what was your prompt?

It's right there: “Make me a video of data centers getting overrun by dinosaurs.” I couldn't believe how good it was. No reprompting, no tuning—this is just what it comes back with.

I had Chase Laakmann and Crusoe, and Project Stargate in the back of my mind, because Chase has become a celebrity overnight. More power to him. He's our fellow MIT alum. Crusoe is the company building Stargate in Abilene, Texas.

I was watching the video of this insanely large amount of construction—the electricity, the pipes, the power generation—and now the chips are going to start coming in. That's worth following just as a storyline by itself. Chase is a great guy. He'll be another Mark Zuckerberg character for a long time to come.

We should bring him on the pod here. Yeah.

I don't know why I wanted dinosaurs to be overrunning it before it's even done, but it was just what was on my mind.

All right. There's one more Veo 3 I need to show, and I find it fascinating. All of us, in the evening, on TV, see these pharmaceutical ads that are ridiculous, and they spend a lot of money on them. Well, this is an example of a pharmaceutical ad made for $500 instead of $500,000. Take a quick listen to this one.

I tried everything for my depression. Nothing worked.

Every day felt heavy. I felt trapped.

Then I tried Pupperamin.

Our prescription helps your body secrete a special pheromone that attracts puppies.

The commercial goes on, and it's hilarious.

What?

I mean, for me, that would be the equivalent of an ad campaign. Never mind. I'm not going to go there. All right, so—

That looked jaw-droppingly real.

It did. It did.

It's hilarious.

All right. So this is my favorite Google I/O piece. For me, this is the materialization of Jarvis. I want to play this.

Mm-hmm.

Look at this, and then let's talk about holy-shit moments and how this is going to change our lives.

Can you go on YouTube and find a video for how to fix that?

Of course. I'm opening YouTube now. This looks like a good video.

Place a rubber band over the head of the screw—

Nice. Can you go through my emails with the bike shop and find which size hex nut I need?

Sure. According to the email from Bicycle Habitat, you need a 3/8-inch hex nut for your sprocket. I've highlighted the correct bin for you.

It seems like I need a spare tension screw. Can you call the nearest bike shop and see what they have in stock?

Yep. Calling them now. I'll get back to you with what they have in stock.

I think my brake pads are wearing down. Can you check the manual again and see if it talks about them anywhere?

According to the manual, if you look on page 24—

Hey, do you want to get lunch?

Yeah, give me 5 minutes. Can you finish what you were saying?

As I was saying, if you look on page 24, section 2, you'll see how to replace the brake pads.

Hey, any updates on that call?

Yep. I just got off with the bike shop. They confirmed they have your tension screw in stock. Would you like me to place a pickup order?

Could you show me some examples of dog baskets I can put on my bike?

Sure, I can help you with that. Just give me a moment. Here are some options. I think Zuka would look really great in these.

Jarvis, baby, is here. I want to free myself from holding the phone. I want to have my eyewear and audio pickups see what I'm seeing, but holy moly.

Yeah. This is a good place for me to give a shout-out to all our sovereign leader friends in the Middle East and elsewhere. What you just saw is not actually going to happen in your country, because you don't have access to the compute. You can do this for the next few months, but then all the other use cases that are bubbling up are going to overbid you for that data center space.

Your population is not going to be able to do what they just saw unless you have some kind of national compute plan. Like we said last time, about 180 out of 200 countries have no plan whatsoever. You have to get on that right now if you want what you just saw to actually exist in your population, and they're going to be—

Wow.

—screaming for it.

Wow. It is magical powers. It's superpowers. That's so true. We are so compute-limited. It is the currency of the future. It's oxygen for us.

Mm-hmm. And it's a basic human right. It's going to be more important than other basic human rights within a year or 2. Populations don't recognize that today, but once they experience it and then they're deprived of it, they'll be all over their leaders.

Wow, how quickly a miracle becomes something we expect and feel like we deserve. Salim, what did you think of that Jarvis moment?

The video's amazing. Once you can get all the billions of little YouTube clips activated and useful in this kind of seamless way, holy moly, that's pretty incredible.

I want to touch back on what Dave said, though. It's incredible what's happening in this realm, and then you look at country policy for most countries, and they're unbelievably backward. I was talking to the head of state of a country that's emerging, about to become a very big, very wealthy country, and their first idea was, “Let's build a call center.” And you're like, “No.” We're 4 generations...

Let's at least get it into the 19th century. Forget the 20th century, forget the 21st century. This is where I think smart people at the government level are going to need to be there to really change the game, because almost every country in the world makes policy defensively and reactively, right?

Only 2 places do it positively and in a forward-looking way: Dubai and Singapore. Honestly, they don't do it that well. Imagine you had a really forward-looking policy in any country—you could really change the game. So I think 1 or 2 or 3 are going to start doing it, and then we're going to see this general transformation in government.

But boy, I think this will be the forcing function that drives that transformation, so that's very exciting.

I want to add—listen, when I think of this as Jarvis, the countervailing partner to Jarvis is Tony Stark, right? Tony Stark is—

Mm-hmm.

This radical billionaire has access to this incredible AI capability, but guess what? This capability is now available to everybody. We're demonetizing and democratizing our vision of this future, and that's extraordinary. That's going to weigh in on you winning the bet here, Salim, versus me.

All right. We'll get into that another time.

At the beginning of this episode, I think we're done.

I don't want to give away too much. All right, let's finalize the bet first, and then I'll tell you why I'm going to win.

All right.

But I do, actually, Peter, I want to tell you: between Salim in New York, me at MIT, and you in LA, we're booting up a new company right now that takes advantage of what you saw—that a synthetic ad is just as good as a real ad now.

Yeah.

We ran some tests over at EverQuote using celebrity ads, and they work incredibly well. One of the problems you run into with LeBron James is that his peak click-through rate, or his peak impression value, is right during the playoffs, when he can't go to a studio and shoot a video.

Nice.

So we tested 2 things. If you make a synthetic LeBron James at peak season, what's the click-through rate? It's incredible. And if you make thousands of variants of the ad where LeBron is doing different things—maybe different languages, just different messages—that also drives up the value of the ad incredibly.

What we're working on right now is a company that makes it so the celebrity just needs to check a box and collect money. They don't have to show up and shoot anything. Everything else is done by AI. Super. That'll also get a lot of your LA superstar movie-star friends, whom you see all the time but who never show up at MIT.

Well, they don't show up for anything half the time. It's like they don't show up to their own movie openings.

Mm.

Yeah.

You'd be amazed. If you get Will.i.am, or anyone, to show up on MIT's campus, the students come out by the hundreds. They just way overreact to a celebrity, so I'm really excited about this project—dragging just a handful over to the campus.

I love it. Here's a quick, fun tweet from Dario saying, "2026, we'll see the first $1 billion company with 1 human employee. We've been predicting this for a while." I get it, and I agree. I don't know if you guys—

Peter, 2 years ago, you and I were launching the Exo 2.0 book, and we said it would take 3 employees and AI to deliver. So that's now shrunk to 1. It'll soon shrink to 0, by the way.

Mm. Yeah.

AI is that way.

With agents and crypto. Yeah.

Yeah.

Well, who gets the money? That's interesting.

The agent.

The agent. The bot.

This is the whole DAO kind of paradigm coming to life.

Yeah, the distributed autonomous organization. For me, probably one of the most exciting things that's coming is how these models are going to help us drive breakthroughs in math, physics, and chemistry.

This is from, I believe, Anthropic. This is the prediction on when disciplines will be solved. Pure mathematics will be solved by 2028. This means all of the unsolved math challenges out there are getting solved by AI.

Let's move down. Computational chemistry by March 2029. Medicinal chemistry leading to candidate molecules by October 2029. Materials science is being solved by 2030. This means you're able to say, "I need a material that has these thermal properties, these cost properties." That's insane.

Cell biology, core pathways, by May 2030, and climate and Earth modeling systems by 2033. This is the nonlinear inflection that just explodes all of our expectations about the future.

Look at that little note where it says most diseases are curable in 2.5 years, right? That vertical line. That's crazy.

Yeah.

I'd like to see, in pure mathematics or computation, which breakthroughs they expect to achieve. I'd like to see the discipline as "solved," quote unquote, but I'd like to see what specific milestones they're expecting to hit there.

So anyway, we can worry about that some other time, but this is amazing overall.

Yeah.

Yeah.

This is where we desperately need Richard Socher as a guest, because he's mapped out the timelines. It'd be great to get his opinion on whether these timelines are accurate or not, because they're mostly gated by simulation modeling and synthetic data.

That varies by different disciplines, but we've known that quantum computing is really, really important for materials science and for chemical reaction simulation, which is the full-cell simulator. So the timeline and the technology are probably somewhat predictable now, and we need to get his opinion.

This is why I love getting on this podcast with both of you guys to really consider what's just happened this past week. You really do help me contextualize it. And as we do every week, let's talk a little bit about crypto and Bitcoin. This was a big one: Bitcoin surpasses Amazon and Google in market cap.

Salim, are you happy?

I'm getting happier by the day as the thing goes. For me, this is the definition of democratization, because every single individual in the world can own Bitcoin, and it's very, very hard to own, say, gold, right?

This is, I think, a huge trend in the right direction, and we could have a whole dedicated episode just on the impact of this on fiat currencies. That might be with Jeff Booth as a guest, and we could just talk through the implications for how we've been doing things for 50 years, with fiat currencies having enormous stress.

We saw this week the struggle in Congress to pass a bill that's just going to blow the debt open. So there are some big implications for all of this.

Yeah, we'll see Bitcoin surpass Microsoft as well. It's got a while to go before beating out gold, but it will get there as well. Of course, this week we saw a surge, with Bitcoin surpassing $110,000 and hitting all-time highs. It's just an extraordinary time to be alive, as we say every single time.

We talk about the value of markets. $50 billion of Bitcoin gets traded every day. That's a staggering number. I mean, it's incredible.

Yeah. I know at least 1 person who's very happy, and that's Mike Saylor. I haven't seen MicroStrategy's stock continuing to peak at the same time that Bitcoin's been going up. That's interesting—that it's dislocated from Bitcoin in some fashion. But that's a different conversation.

Thank you, everybody, for joining us. Please subscribe. This is our desire to deliver to you the news that is transforming how we govern our nations, how we run our companies, our industries, even educate our kids and our families. Always grateful and appreciative to my Moonshot mates here, Salim Ismail, who's that way, about about five meters, and Dave Blundin, other side of the US in Boston. All right, guys. Can't wait to see what miracles are gonna happen next week.

Peter, I just wanted to thank you for inventing this podcast. It's getting far, far more views than I ever would have guessed, and for me, it's been life-changing. This is so much more efficient a way to reach out to people than one-on-one meetings or getting on stage, which I do a lot of anyway.

But this has just been life-changing for me, and I just wanted to thank you for inventing it and also for inviting me to join you in it.

That's okay.

It's part of the family here. I love your insights, brother. All right.

Oh, it's so fun.

Salim, say something really—

I mean, you're wrong about a couple of things on the democratization, but the tech insights coming from you, Dave, are off the hook. So, really great.

Well, thanks, Salim. All right, guys.

We'll battle it out. All right.

Have an awesome day. Can't wait to see what unfolds next week.