深入 Thrive Capital：投资 OpenAI、Wiz、Cursor、Nudge 与 Physical Intelligence

Our philosophy at Thrive, which Josh taught me, is that the people who win deals are the people who want to win deals the most.

Philip, welcome to Sorcery.

Thanks for having me.

This is a very big day. We have a very stealth private fund here. Okay, so you've been at Thrive for the past 3 years. You joined there in 2022.

Yeah.

And since then, you've been involved in some pretty incredible investments, like OpenAI, Cursor, Wiz, Nudge, and Physical Intelligence. How did you get to Thrive?

1. The Path To Thrive

Yeah. I think at the end of the day, I'm just a technologist and optimist. I have always really wanted to study and be like the people who worked on really big technology products and projects that bent history in the right direction. When I was a kid, my heroes were Oppenheimer working on the Manhattan Project, Claude Shannon working at Bell Labs, or Elon Musk starting SpaceX when I was 4 or 5 years old.

I actually went to school originally to study physics. I really thought that was the right way, sort of in the mold of some of those earlier figures, thinking about the hardest problems in technology. Now, the truth is, I was pretty good at physics. I probably wasn't quite good enough, and the road to making an impact on physics problems that are open in the field is very, very long.

I ended up switching to a more pragmatic route, which was studying computer science. Even so, I was always really inspired by this idea of technology being the right engine of history. After spending a lot of time thinking about having an impact there, I think there are 2 big options today if you want to actually be part of that technological engine.

One is that you can go start a generational technology company. To be perfectly honest, I didn't have any really great ideas about how to do that coming out of school. Or, second, you can be a partner to these sorts of companies and help them make the really important decisions. Over time, it became pretty clear that the second was the path that I wanted to take, and Thrive seemed like a really exciting place to do it.

I think Jeff Bezos had the same problem. He wasn't that good at physics.

He actually tried to do math and then physics, and then he did quant trading.

So how did you meet Thrive?

Yes. A little bit of a funny story here. To the point of being an amateur physicist at one point, I was really into semiconductors. I think they're one of the great examples of applied physics in the world. We literally have machines that zap plasma with lasers twice in order to imprint patterns on semiconductor die.

I used to maintain a little Substack about semiconductors, and my now-partner, Josh, who started Thrive back in the 2010s, was really interested in 2021 and 2022 in the idea of reshoring semiconductor manufacturing to America. These days, I think almost all advanced semiconductor manufacturing happens in Korea and Taiwan.

This actually is a big change from history. Back in the 1990s, about 40% of semiconductor manufacturing happened in the U.S. Today, it's something like 10%—a drastic decrease. Now, as you know, and as I'm sure all your viewers know, these are probably the single most important inputs to all information technology, whether that be computers, artificial intelligence, or what have you.

So the big question is: Is there a way to get America to be a major producer of these things again? Josh was exploring this and came across my Substack because a former colleague at Bridgewater, where I worked before ABC, was at Thrive at the time. He got in touch, and we started talking about semiconductors.

I'll spare you all the gory details, but it turns out it's very difficult to reshore semiconductor manufacturing, or at least it would require a little bit more money than Thrive had available at the time. But my reaction was that someone who had the boldness and ambition to think about an idea as important as that was really unique and distinct. If you go back to great technologists working on great technology projects, that felt really unique and distinct.

As it became clear that I wanted to move from the quantitative-trading hedge-fund world to actually investing in early-stage and fast-growing technology companies, I went to Josh for advice. You know what they say: Ask for advice, get a job. Now I'm at Thrive.

Simple as that. Make a Substack.

Make a Substack.

So, no investments in semis?

Not yet. We are investors in a company called Mesh Optical, which is in the data center space. If you think about the great AI capex build-out happening right now, there are hundreds of billions of dollars being spent each year on building out new data center capacity to train the latest generation of AI models.

Within a data center, there are a few important components. There are the semiconductors themselves. NVIDIA is a $4.5 trillion company as a result of being the greatest semiconductor designer in the world. But there are also a lot of other technological components.

One of these is called the interconnect, which is what moves data from one point in a data center to another. Mesh is a pair of founders from SpaceX, Travis and Cameron, whom I actually knew back in college. They went to work on the Starlink photonics team—the team that uses lasers in space to pass data from one satellite to another to give us the internet.

They are bringing that innovation to the interconnect space in data centers, and we led the seed for that company at the beginning of this year. That's where we've invested in data centers thus far. No chip companies thus far.

Okay. All right. Well, we'll cover the whole stack because I definitely want to get your take on where the value accrues in AI. But before we get there, let's talk about your—

Yeah.

—thesis around that. So what companies are in your portfolio, and what board positions do you have?

2. The Portfolio Takes Shape

Totally. I'm pretty lucky to work with a few really, really exciting companies. I think one of the great privileges of this job is that you don't just get to work on 1 really exciting problem; you get to work on a few with several founders with the potential to be generational.

A few companies that I work very closely with: Cursor—

Automating a lot of the drudgery associated with coding and being the ultimate AI pair programmer for a lot of developers.

I think when we invested, there were a couple of tens of thousands of users and low-single-digit millions in ARR. Fast-forward to today, I don't think the company publicly publishes revenue, but call it many, many hundreds of millions in run-rate revenue and many millions of developers using the product.

That has just been an incredible growth story in the span of 18 months. I think one of the really wild and exciting things about the AI era is that the timelines are getting compressed in a really radical way.

I had the privilege of speaking to Michael Dell a few weeks ago. He's one of the great CEOs out there who has lived through both the internet era and the AI era and played a big role in each. I asked him how he thought about the difference between the two, and his line to me was, “In some ways, they feel very similar in magnitude of impact, but whereas the internet era was like playing chess, the AI era is like playing speed chess.”

I think that's what we're starting to see with those companies. The founders at Cursor are some of the greatest technologists I've ever worked with, and it's really exciting to see them change how an entire discipline and way of working is done in a small window of time.

I work very closely with a company called Wiz, which does cloud security. If you think about the internet as having created the first vulnerabilities of web technology, where you could actually access documents, files, and servers through the open internet, we got firewalls.

The next wave of information technology was the cloud, so we moved all of our software workloads to the likes of AWS, GCP, and Azure. But in doing so, we created a lot of net-new ways in which software could be accessed and made vulnerable.

Wiz was started in 2020 with the goal of securing cloud software for the world. It was started by a really incredible team of Israeli entrepreneurs—Assaf, Roy, Yinon, and Ami—all of whom had actually worked on this problem for a little while earlier, but acquired by Microsoft, built out the cloud security business at Microsoft, and then built Wiz.

Another amazing success story in a very short amount of time, and Google announced the acquisition of that company earlier this year for, I think it was, the largest sum a startup has ever been acquired for. So, another really exciting one.

We talk about OpenAI, Physical Intelligence, Nudge, or others as well, but those are a few companies where I've had the privilege of being close partners to the founders.

So Cursor is super hot right now, if you didn't know that.

Yeah, it's news to me.

I would love to know: How did you meet Michael and his team?

Yeah. Michael I actually met when he was working on a different idea. This is, I think, one of the great lessons of early-stage investing: Ideas change a lot. People are really the constant that you want to be focused on.

I met Michael back in late 2022. He was actually working on a tool for mechanical engineers at the time, trying to bring AI to what's called CAD, or computer-aided design. Actually, the day I met him was the day he decided to pivot away from that idea.

And so most of our conversation was not about any particular company, but about the list of possible pivots. I think my one helpful recommendation at the time was that he was thinking about customer support, and I recommended that he might find something related to software more enjoyable. My guess is I didn't actually have that much of an impact, but I'd like to think maybe it was a little bit of a nudge in the right direction. But even in that moment, I think with great founders, there's this almost electric energy you can sense in the first meeting.

Shortly afterwards, Michael and Aman moved to San Francisco. They actually merged with another company, where they got their other 2 co-founders, Swala and Arvid, and they started working on the first AI IDE, or integrated development environment, which is sort of the text editor where coding happens. When they put that out, I tried it and was an early user of the product, and it was definitely one of those magical one-way-door moments. Once you walk through a door like that, you can't not walk through a door like that.

And so Myles, my partner, and I spent the next year trying to see how we could ultimately partner with them. Fast-forward to May 2024, and we finally convinced them that we were the right sorts of people to bring onto the cap table.

That's wild. And to the point of just going hard at it, I think you were cited in the Colossus profile for flying into a war zone. This was after the second company that you mentioned, Wiz. So what was that like? Tell us everything.

3. Winning The Wiz Deal

It's definitely one of the more memorable moments of my career. I think it comes down to this: we are willing to do anything for our companies, and we want to show our companies that if we're going to be the right partner, they should know that we're willing to do anything.

The story with Wiz is that I actually met them shortly after I joined Thrive for the first time. If you want a fun story, to join Thrive, Josh had this really great idea that everyone at Thrive should go to a city outside the core of venture—not New York, not San Francisco—and explore it for a couple of weeks. So I went to Tel Aviv. I had never been there in my life, and if you're in Tel Aviv in 2022, the talk of the town was this really buzzy, exciting new cybersecurity company, Wiz.

I spent 2 weeks trying to figure out the right way to meet this really cool company. Eventually, I got a meeting with Assaf, and Josh and I did this 3-way Zoom. I was actually back in New York at the time, Assaf was in Tel Aviv, and Josh was in Tokyo. So we were really spanning time zones.

To the point about electric founders, as I mentioned with Michael, it was immediately clear this was not your mom's security company. It was very, very different. I think many great security companies in the world are often a little bit more risk-mitigating—you might say “cover your butt” if you want to be a little risqué about it. Wiz is not like that. It is a software infrastructure developer platform.

I think one of the most shocking things I learned on that first call was that 50% of Wiz's users at the time were software developers, not security people. It was very different from how security is traditionally done.

I think we were pretty quickly sure that Assaf was this wildly humble, charismatic, and brilliant founder. He was working with 3 other co-founders who complemented him in incredible ways across product, technology, and marketing, and this was among the most unique security products we had ever seen.

Similar to Cursor, it wasn't like we wrote a check that day. We actually spent a year or so building a relationship with the company, trying to act like partners well ahead of actually being partners, because that's just how we operate. Then we get to December 2023, early January 2024, and Assaf gives us a call and walks us through the latest numbers. Josh and I look at each other and realize we absolutely have to invest in the company. It's one of the most beautiful software businesses we've ever seen.

The problem is that the Israel–Gaza war is going on at the time. It's not super easy to get into the country, and so by hook or by crook, we make our way to Tel Aviv in 12 to 24 hours. We land, and Assaf and Yona are generous enough to get dinner with us for 4 hours. We talk about the future of the company and their ambitions, and it's immediately clear they're thinking about Wiz on a different level than just trying to build a tool—really, a different way to build cloud software.

By the end of the dinner, we definitely wanted to invest, so we shook hands on a deal, and we're lucky to be investors in the company.

I definitely have a bunch of questions off of this, but what was mentioned in the Colossus piece was that no other investor would fly there at the time. Was this a highly competitive round, and did you feel you needed to go there to close the deal?

Yeah, I think it's a great question, and I would frame it a little bit differently. Certainly, I think our philosophy at Thrive, which Josh taught me, is that the people who win deals are the people who want to win deals the most. So we always have a leave-it-all-on-the-field mentality.

At the same time, when we're partnering with companies or discussing with companies whether we're the right partner, we don't really do it thinking about other players on the field. You play against yourself; you don't play against others. If our founders were going to be in Israel during what was a trying time, we wanted to be in Israel and show them that that mattered to us, too.

And so I think it was much more a statement about the type of partner we wanted to be to the Wiz team versus the competitiveness of the deal.

And then walk me through diligence. You said that it had amazing numbers, but what does that mean?

Yeah. Wiz, I think, exemplifies one of the key things in enterprise software that I'm surprised more people don't talk about. Oftentimes, enterprise software has all the classic things you might care about—margins, retention, et cetera. Wiz looked 11 out of 10 across all of these. I think they had almost never churned; they churned 1 customer in all the time between when they started the company and when we were looking at this round.

But a really special aspect of the business is this: what is the size of the deals you can sell versus the speed at which you can sell them? I actually think often looking at this ratio is a really, really powerful metric of basically how many dollars you can turn as a company and how quickly you can change a market.

Wiz had, I think, the very best ratio of deal size to speed of implementation of almost any company I'd ever seen, and it makes total sense. If you're a sales rep at Wiz, you can get on a call with a customer and ask them for their cloud API keys. You can pull it into their side scanning product. They will scan your environment and give you a list within a few minutes of all the vulnerabilities you didn't know you had.

What customer is not going to look at those and say, “Oh my God, I have a bunch of vulnerabilities, but I'm not going to buy the product that showed me the fact that I was super exposed?” And so Wiz was closing on these very healthy 6-figure deals in a matter of weeks or months, which was just unprecedented. Normally, you have small deals and fast cycles, or you can have big deals and slow cycles. Wiz had more or less figured out how to cut the Gordian knot and get both. So that felt really special.

I think the second was that I mentioned this developer population on the platform. It actually felt more like a software infrastructure tool, like Datadog or Cloudflare or the like, than a traditional security tool. It wasn't a check-the-box compliance workflow. It was actually about how you surface issues to engineering teams, provide them with the context to fix those issues, and then remediate them. That felt very unique and special, too.

And I think finally, most products like to say they're platforms. Very few products are actually platforms. Wiz was actually a platform. If you looked at the number of products that were being adopted across different areas of the stack, they had code security on the left-hand side of the stack, cloud infrastructure security in the middle of the stack, and runtime security—so, actually looking at the live software workloads running on the platform.

Seeing all 3 of those get really healthy adoption rates in an end-to-end way for a company that was barely 4 years old at the time was really, really unique, too, and gave us a lot of confidence that the company was going to have not just an act 2, but probably an act 3 and 4 as well. I think the mark of a lot of great companies is that their first product gives you the right to go do and build some really exciting second and third products.

So these are both software companies, but I've heard you might be the hardware guy on the team.

4. Hardware Gets Easier

I do have that reputation a little bit.

Why? Physicist?

Yeah. When you're growing up as a physicist, you really come to appreciate hardware and like to tinker around with it. I also think if you're a technologist and an optimist in this moment, the barriers to entry to hardware are coming down a lot.

Okay.

Input costs are getting a lot cheaper. These days, lidar costs are a couple hundred dollars a sensor.

Back in 2014, so barely 10 years ago, there was something like $75,000 or $80,000 per sensor. So, a wild decrease.

Software is making it a lot easier to build hardware, too. You live in San Francisco. We have self-driving cars that I think increasingly outnumber human-driven cars on the streets, so we can build a lot more interesting things with the software intelligence layer than we used to be able to.

In addition, there is a generation of founders who have been trained at the first wave of really great technology companies. If you were Elon starting SpaceX in the early 2000s, or Palmer, Brian, Trey, and Matt starting Anduril in the late 2010s, there wasn't actually a huge population of preexisting, startup-hardware-trained talent to draw on. So they did amazing works of heroism and built really important businesses.

The nice thing today is that those companies have now trained a lot of really amazing engineers who know what it means to build software and hardware technology that can scale to very, very high volumes. Those founders can now look at this host of new problems that are available to them because of the changes that have happened in technology and go chase them.

As an investor and someone who's really excited about companies that can counterfactually impact the world—companies that can make the world look physically different in some meaningful way—I think the number of opportunities in hardware is now far greater than it has been at any time in the last 20 years.

I think you were quoted—and we talked about this before—as saying that you wanted to become the CEO of Lockheed Martin.

When I was a kid, this was a childhood dream. It's not still a childhood dream. I was born in the very late 1990s, and there was no SpaceX, there was no Anduril, and there was no Tesla. Were I born today, I'd like to believe I'd be CEO of one of those companies.

At the time, if you're a hardware nerd and a physics nerd, what are the companies you're really excited about? They're the companies that take on really big technology projects. Say what you will about Lockheed Martin, but I think they took on some of the most ambitious technology projects of the 20th century: the SR-71 Blackbird, large amounts of the stealth-fighter program, and large amounts of the missile program.

Those were really incredible technological accomplishments. So probably what I really meant when I was in my early teens, a young kid thinking about that stuff, is that I just wanted to be in a place where I could push the envelope on how really great technology gets built.

Who knows? Maybe you'll take over Lockheed Martin and become CEO.

One day. It's never too late.

It's never too late. So it's interesting because you build software and you build hardware. I want to talk about the revenue between the two—the different types—and then also the business model. How come? You mentioned a little bit now that the window for hardware investing is a little bit more attractive. But on the business-model side, can you break that down? It's never been an attractive VC investment, so why is it attractive now?

Yeah, all those people who invested in the early stages of SpaceX and Anduril are really regretting it, right?

They did a horrible job.

Horrible job. I think you're pointing toward something that is genuinely true, which is that hardware is definitely harder than software. It's harder because you actually have to build more product. It's harder because that product is more expensive to build. It's not like writing bits, which are more or less free. It's atoms, which cost money to put together.

It's harder because there is often an operational aspect to hardware investing. I don't just have to build a product and press a “ship the binary” button on a computer. I actually have to have a manufacturing line and a deployment line that gets this out into the real world.

That said, I think hardware businesses can be enormously durable and attractive once you actually get them off the ground. It's really hard to imagine who is possibly going to disrupt SpaceX at this point, knock on wood, or who could really go out and disrupt Anduril. That's because when you get over the hump of these really, really hard industries and problems, you've actually built something at scale that is very hard to quickly replicate.

Whereas in software—not true for all businesses—the barriers to building software are probably lower than ever before, too. So I think there is a dynamic here of, yes, there are bigger fixed costs to getting hardware off the ground, but these businesses can be really unique and powerful once they do.

They often actually have a very geometric growth rate. I was looking at the revenue curves of Anduril, SpaceX, and Tesla the other day, and they are some of the only companies with actually accelerated growth at scale. That is to say, they are growing at a faster rate at multi-billion-dollar scale than they were when they were small.

Normally, in a technology company, you have a deterioration of growth over time. I think the difference with hardware is that the bigger you get, the bigger markets you can access and the more people trust you to take on big problems.

Take Anduril, for instance. When they were an early-stage company, they were really great about getting on the field. They won a program of record, which is a very big-budget, line-item contract in the National Defense Authorization Act, within 2 years—the fastest timeline since the Korean War.

Those deals were good, and they were on the order of tens or hundreds of millions of dollars. Now Anduril is being tapped for many of the great technology programs that the U.S. military is thinking about across all domains, from air to space to the Navy, et cetera. That means the number of contracts they're eligible for and the amount of revenue they can go after is quite a bit larger.

So, again, I think they're actually expanding the aperture of what they could do. Or think about SpaceX. For the first 15 years of the business, SpaceX was a launch business. It was a great business with lots of free cash flow, and it was the only provider in the U.S., maybe the world, that could bring mass into orbit at a competitive price point.

I don't know if you know this, but before SpaceX, we had to rely on the Russians to bring people to the International Space Station. It's good that we can have an American company doing it now.

That created the opportunity for them to build Starlink, which is now one of the most powerful internet constellations in the world. It's actually driving a very, very large majority of the overall free cash flow and revenue for the business.

So, again, I think these later-stage second acts of hardware companies can end up looking a lot bigger than the early stages, which is a very distinct characteristic. That said, great businesses are great businesses, and you do need to make sure that these things are still high-margin, have great products, and have great founders, et cetera.

There certainly has been a surge of funding that's gone in this direction.

Yeah.

And there are also bottlenecks that come with it, whether it's policy, government, or different kinds of regulations.

So, how do you see this current stage of hardware in a cycle? Do you think that we were in a bubble? Are we in a bubble? What's the temperature on the ground?

Yeah. I think the temperature on the ground is that the returns always accrue to a small number of companies. Venture is the actual asset class with the worst beta and the best alpha of any financial asset class, which is to say the best companies are going to give you the best possible return of any possible equity instrument you could invest in, and the index is probably going to give you pretty mediocre returns.

I think hardware is going to look no different than that. I am fairly confident that we are going to produce some of the world's most valuable companies in hardware over the next decade. Even right now, 5 of the 10 most valuable companies in the world are hardware companies: Apple, NVIDIA, Broadcom, Tesla, and TSMC.

At the same time, these businesses are hard. There aren't that many people who are really great at being combined hardware-software thinkers, and so there are probably going to be a lot of companies that don't work. That's just the nature of company building.

I think the really exciting thing is that the types of companies we invest in are built by N-of-1 founders, like the Camerons, the Travises, or the Palmers. They're also investors in Nudge, like you mentioned: Jeremy, who used to help run engineering at Neuralink, and Fred Ehrsam, who was one of the co-founders of Coinbase. They're working on these very unique missions.

I can't think of anyone else who is using noninvasive ultrasound to stimulate brain activity like Nudge and trying to cure neurological illnesses like depression, pain management, and addiction in ways that are going to be very, very hard to replicate in the future. I think those companies that can actually solve the problems we're looking at are going to be valuable no matter what.

Okay, let's talk about Nudge—

Let's do it.

Because you brought it up—

Yeah.

—and it's really fascinating. Tell me everything.

5. Engineering The Human Brain

Well, I'll tell you all the public things. I think the nature of early-stage technology companies is that you want to make sure that the correct secrets are kept secret.

The way I would think about Nudge is that the human brain is basically 1 of the last frontiers of engineering. For a long time, it's been considered a science problem—something where we'll study it in the clinic, we'll study it in the lab, but we don't really build products for the brain in the same way we might build products for other parts of the economy or even for human health.

Nudge is taking the approach of, what if we could engineer the human brain or bring engineering solutions to the brain's problems? For a problem like that, you want people who I describe as life's-work founders—people who, no matter how much money they have or what island you put them on, only have 1 missionary purpose in life. I think Jeremy and Fred are both that.

Jeremy, like I mentioned, was 1 of the early employees and key leaders at Neuralink, another really amazing neurotech company. Fred was actually 1 of the early investors in Neuralink and obviously built an incredibly successful company with Coinbase. I think they are both convinced that the most important thing they can do with their lives is go and actually build a solution to neurological illnesses.

So how does Nudge work, for those who haven't heard of it—which I imagine is most people right now, but hopefully not very many people in the future? Much like we use ultrasound to image pregnant women—

You can use ultrasound waves to actually stimulate brain activity.

They're very safe. If we're comfortable taking images of babies, we should be pretty comfortable using them on ourselves. What Nudge does is take ultrasound waves and direct them at very, very specific points in the brain that they've identified as being associated with some type of neurological problem. It uses those waves to stimulate the brain and actually change the neurological activity that's going on there.

Ultimately, the goal is to have a treatment for neurological illnesses that is easier than taking a pill. Imagine if we could treat really, really tough illnesses like depression and addiction by having you wear a set of headsets on your temples, and that actually doing meaningful amounts to cure them, versus struggling with an illness over a prolonged period of time.

It's a really inspiring mission. The company is early but moving really, really fast. If there are any engineers out there who are excited to work on engineering the brain, I think it's 1 of the best places in the world to work.

So, for most people, what's the difference between Neuralink and traditional clinical brain stimulation?

Yeah. Neuralink is a really amazing company as well. I think incredibly highly of D.J., Elon, and the entire team there, who have done huge feats of engineering.

I would think about Neuralink as starting with the read side of the brain—understanding the brain. They use an invasive neurological implant, so if you're a paraplegic or someone with ALS, they will implant a device into your brain with a relatively easy surgery that they've automated to a high degree.

As of today, this gives you what is called Telepathy, in classic cool Elon terms. Which is to say, you can now control any digital device with your brain. For people who have paralysis to that degree, this is a life-changing experience.

You can imagine that if you have ALS and are no longer able to speak, you can communicate with your loved ones for the first time. If you were paralyzed in a ski accident, you can now go back to work and do meaningful economic work for the first time by interacting on a computer.

They're on the read side. Over time, I think they'll actually do a number of things that are really helpful on the write side, too. Things from their roadmap, like Blindsight, will try to restore sight to the blind. You can imagine things that are going to try to restore limbic motion. Basically, biblical miracles. It's pretty amazing that we live in an era where this is possible.

Nudge is mostly on the stimulation side. They are not so much reading what is going on in the brain and interpreting it, but actually targeting specifically affected areas of the brain and trying to stimulate them.

The closest comparison that we have to this today in modern medicine is something called TMS, which is transcranial magnetic stimulation. This is really a version of what you might have seen in the movies called electroshock therapy, where people will more or less use electrodes to zap the brain. It's actually pretty effective. It is very unpleasant, and I have not personally tried it.

What kind of diligence is that?

I know. Totally slacking here. Did you try it before the podcast?

Yeah.

You did? Then you must be a big believer in Nudge.

I love Nudge.

Excellent. But in any case, oftentimes people have to do this treatment over many weeks. You have to go to clinics very far away from where you work.

What's called second-line adoption of this treatment—the number of people who will try this after first-line treatments for depression fail—is very, very low because no one wants to do this. Not only do we think that ultrasound can be more effective than TMS, but wearing a pair of headphones on your head is way easier than going to a clinic to get zapped.

I think that's hopefully evidence that this is not only possible, but that there's just a much, much better way of doing things.

Will you try it?

Nudge? I'd love to. I think 1 of the exciting things about Nudge is that you could imagine a world where everyone should use Nudge.

Today they're treating neurological illnesses, but if we can stimulate the brain to get desired outputs, why shouldn't I be able to stimulate my mood or my energy levels or maybe my focus levels? I think there is a world where, when we do this podcast again in 10 years, we're both wearing headphones on our temples, and they're getting us really dialed in to ask great questions to each other.

Everything is a computer.

Exactly.

That's really funny. I want to take a step back and talk more about the fund's thesis and the thesis on concentration.

Yeah.

Peter Walker, a researcher at Carta, recently shared this chart that showed spray-and-pray versus highly concentrated funds. It turns out highly concentrated funds, typically with about 20 companies per portfolio, outperform spray-and-pray. But spray-and-pray looks pretty good because you have lots of potential bets.

With a rumored—and perhaps reported—$25 billion in AUM, how do you guys think about where to concentrate your bets?

6. The Case For Concentration

Yeah. I think there are 2 aspects that really matter here. First, you have to pick a strategy that is authentic to how you want to operate as an investor. My partner Kareem often has the line that mathematically, there are probably a lot of ways we can make money as a venture firm.

The question is what feels really aligned to the founders we partner with. And I think, above all else, I can talk about why we think concentration is a really great financial strategy, and I will, but the most important thing is that it allows us to be really deeply aligned with the founders we work with. If you're working with a life's-work founder, this is a portfolio of 1 for them, and we may have a portfolio of a few, but we really shouldn't have a portfolio of that many because all the wins should really feel great and all the losses should really hurt, and we should have real skin in the game.

I think the way we like to set ourselves up is that if we're going to invest in someone, they should really believe that the success and failure of the company really matters to us as investors. And so I think there's actually an emotional and ideological aspect to this. Now, there's also just a reality of the power law, which is that I don't need to document this for you. It is very clear that a small number of companies drives a disproportionate amount of the economic and social value in the world.

I think right now, I'd have to check the latest stats, but the Magnificent 7 and the Nasdaq—the top 7 stocks—represent 50%+ of the Nasdaq's market cap. That goes to show you that a few companies are really, really important. And so our concentration strategy just reflects that reality. We want to invest in a small number of companies that are on the right tail of the distribution of the power law, that are going to be the most important to how the world works over time. That's not to say there aren't other great companies that might make us money and that we can invest in with a more index-like approach, but the way in which we think we can deliver differentiated, correlated returns for our partners is through investing in the very best companies. And once we think a company is of that stature, concentrating as much capital into that company as possible.

Perfect tee-up. Okay, we're going to talk about OpenAI. I pulled from Patrick O'Shaughnessy's original interview with Josh way back when, and Josh said, “Concentrate on the most exceptional businesses and hold onto them over time”—case in point: concentration strategy.

I was also able to find some data on OpenAI rounds, reportedly led by Thrive Capital, dating back to 2023, and this was right around an inflection point—the ChatGPT moment. That started in January 2023 at a $29 billion valuation. The next step up was February 2024 at an $86 billion valuation. After that, October 2024, a $157 billion valuation. And then most recently, October 2025, hitting a $500 billion valuation. So how did you get involved in OpenAI?

7. OpenAI Research Advantage

Yeah. I think people probably overestimate how popular the company was in its early days, if we want to be perfectly honest. I think Sam, Ilya, Greg, and everyone else at the company worked on this project for 7–8 years before others realized how valuable it was. And so actually, if you want to go back to 2022, it's clear OpenAI was a really exciting company.

We had been spending time with them for many months. Josh had known Sam for a long time. We actually got a demo of GPT-4 in November 2022 before it was publicly released in March, and it was one of the most astounding pieces of technology we had ever seen. I think these days there's a lot of talk of what is AGI and what isn't AGI. We can debate that here if you want.

Debate it right now.

Right now. What does it mean? You and me both are stumped.

Okay. Great marketing.

But the thing that was definitely true is that the model passed the Turing test. There was all this talk for many years that the Turing test was this really important moment. You wouldn't be able to recognize it as a computer capable of acting like a human or having the intelligence of a human. GPT-4 was that, and we don't talk about it anymore, but that was a really important moment in technological history. And that felt really exciting.

At the same time, I think a $29 billion valuation or thereabouts is undoubtedly a high price for a company when ChatGPT hadn't even launched yet. And so when we invested in OpenAI in that round, I think we were one of 2 firms that gave the company a term sheet. This was actually more an artifact of an investment that required belief than of a competitive edge, so to speak.

Now, if we fast-forward 3-ish years from there, so much about the company has changed. We looked at our original investment memo. There was no mention of ChatGPT in there. Now it has more than 1 billion users and is on its way to becoming one of the most dominant and important ways in which people access technology, the internet, and AI. I think it's the front door to AI in many ways, and the company has totally changed in that sense.

As a result, I think when you see these new levels of value get unlocked and you're able to have real confidence in the dominance of the business and its centrality in the arc of technological history, your confidence in the company should increase, and you should invest a lot more in the company. I think the past few years have basically been a story of us having a front-row seat to that and getting to see this arc unfold up close.

I think one of the most impressive things to come out of it is the team, because everything that they've pushed out would seemingly be a tremendous force, and they continually do it over and over again. This is something I talked about with Apoorv from Altimeter. They're also investors in OpenAI.

It was not so much momentum as a moat, but speed as a moat—continually pushing things out and trying things. It went from voice-to-text to text-to-voice, all this kind of stuff. It seems like every month, every week, every day, there's something new coming out of that company. So how do you think about how they recruited and assembled their team and the talent around it?

Yeah. I think one of the great privileges of my career has been that I've been able to spend time with the OpenAI team—not just because they're great people and brilliant thinkers, but the talent density, as you're pointing out, is really unprecedented. And there are good reasons for this. If you're building what I think is one of the most important technological innovations since the atomic era—forget the internet—yes, a lot of the really brilliant people in the world are going to go work on it.

The really cool thing about OpenAI is they've attracted a lot of the star-power researchers. Ilya literally invented AlexNet back in the day, which kicked off the deep-learning revolution. But a lot of the really great researchers at OpenAI have come up through OpenAI and weren't famous or well-known people before. Alex Radford, who was at the company for a very long time, famously did not even really study AI when he was a student and was one of the most prolific and impactful researchers at the company.

I think one of the other things OpenAI has done really well is they've actually looked for young talent in the AI era. I think this is a really important thing, and one of the things that Cursor has actually been really attuned to as well is that the people who almost hold the model weights in their brains because they've been brought up thinking about AI natively are really, really good at the creative thinking, experiments, and hypothesis testing that are necessary to push the performance of these models. Sora, a recent OpenAI product launch, is a good example. If you looked at the average or median age of that team, I think it's in the low 20s.

Wow.

It's pretty amazing. So, long way of saying, OpenAI has the benefit of, one, if you're a great computer scientist, you probably should be working on building some type of superintelligence, and two, they've been really risk-forward about trying to find people who could be great versus who are already great. I think that has actually allowed them to bring up and train several generations of great researchers in the company itself, rather than just needing to hire the people who've already been proven entities.

How do you think about the evolution of OpenAI and where it goes? I've gotten exposure to the AI research side of things, or an AI research accelerator—they help out with the models. And right now, they're noticing a severe shift over the last 1–2 quarters from imitation data and synthetic data to reinforcement learning. So I'm curious how you're seeing it on the technical side and then also on the application side.

I think AI is a little bit like semiconductors in a sense—if I'm allowed to go back to semiconductors, the great touchstone. And that's in the sense that there are many, many waves as we're trying to get better and better performance. With semiconductors, we used to have single-core semiconductors. Then we couldn't basically squeeze anything more out of the single core, so we went to multicore, and then we went to GPUs. And I think we're seeing something very similar with AI, which is that we started with pretraining back in the GPT-1, GPT-2 era and GPT-3. That took us to GPT-4.

Then we started using a lot of post-training techniques for ChatGPT: supervised fine-tuning and reinforcement learning with human feedback. This is what allowed the models to be less alien intelligences and more helpful assistants. Now that a lot of the low-hanging fruit for those gains has been squeezed out, even though there’s probably a lot more to do, reinforcement learning has been the next big paradigm shift in AI and where a lot of the resources are going. My guess is that there are going to be a couple more of these; this is not the end of the story.

We are basically figuring out in real time how to do the alchemy of turning pieces of metal into thinking machines, so to speak. That is a really tough and challenging thing, and I think the really exciting thing about OpenAI is that, yes, it’s one of the great product companies of today, but it’s also a research effort. These are not things that we have fully discovered and figured out yet.

So, I think the long way of saying it is that the ultimate competitive advantage for the labs, and for OpenAI in particular, is actually being able to constantly figure out what the next paradigm is, even as you’re productizing and implementing the last paradigm. I think one of the strongest testaments I can give to OpenAI is that every single major paradigm in large models, at this macro level, that has been very successful, to my knowledge, has come from OpenAI, or at least originated at OpenAI. I think that speaks really strongly to their ability to continue to push the frontier of research going forward.

One of the lines that comes out of Cursor is, “What comes after code?” What comes after OpenAI? That’s the question for you.

Yeah.

8. The New Software Substrate

I think the beauty is that, hopefully, OpenAI is going to be an enduring technology. So I’d be very surprised if, in the near term, we were talking about an after OpenAI. Is there an after Meta or an after Google?

I think the better framing of the question, if I could tweak it, would be something like, “What are the companies that can now be built as a result of OpenAI?”

Feel free.

I think there is a “what comes after SaaS” that is a more interesting way of thinking about this. We had an old way of building software. There are these cloud primitives, and we build these great digital apps on top of them—a lot of really amazing companies that we had the privilege to invest in. Now there’s a new substrate for technology: What if you could productize intelligence and call it via API or fine-tune a model with it?

I think that engenders an entirely new breed of software companies. So I don’t think it’s an after OpenAI. I think it’s an after SaaS that we’re going to see. We’re seeing the early signs of this with products like Cursor, where we’re actually thinking about products that can work with you as a collaborator in order to create work.

We’re seeing this with products like Harvey on the legal side, where they are doing work as well as just acting as an assistant. We’re seeing this with even incumbent products. Wiz, for instance, now has a security assistant to help you find and triage vulnerabilities. I think there basically is a new set of building blocks that we’ve gotten, and I think the right way to think about it is that software is going to look very different going forward. But the big, fundamental companies are going to be around for a while.

The other standard second-order question is the impact on jobs. Calshe recently had a chart out showing that up to 86% thought we would have more layoffs this year in tech than in 2024.

How’s that going?

Yeah. I definitely take the under. But public companies have been reporting more layoffs. It’s part of a natural cycle, et cetera, et cetera. Do you think that AI and these tools will create more efficiencies and create new jobs? Where are you in this whole discussion and debate?

I’ll give you my bottom-up and top-down answer. My bottom-up answer is that I’m an investor in a lot of companies that use AI tools. I cannot think, especially on the coding and engineering side, of a single one that has laid off engineers because of these tools. I think maybe it allows them to grow without adding quite as much headcount, but really what it has done is allowed them to go after a much wider range of problems.

You can actually make everyone the 10x, or proverbial 100x, engineer in a really exciting way. So thus far, I think just empirically, will there be some turnover in economic sectors? Of course—there always is. But it’s actually been much more of an augmenting technology than a substituting technology.

As we look into the future and look beyond just products like Cursor, to this notion of what it looks like to have very powerful, intelligent AI more generally, there are just a lot of problems that, as a species, we aren’t working on right now that we should be working on. Way more people should be working on cures in oncology. Way more people should be thinking about what it looks like to actually sustainably mine the world’s natural resources. What does it look like to actually make us a spacefaring species?

I think the beauty of AI is that we’re going to be able to reallocate a bunch of human brainpower, firepower, and creativity to these most important problems, and away from the things that have been great jobs to date but are not the highest marginal use of humanity’s creative and intellectual potential. I think there are just a lot of problems left to go solve, and AI is going to help us solve the most important of these versus less important problems.

Where do you stand in the AI bubble debate?

Hmm.

I think this is one where being concentrated allows us to feel a little bit better that, no matter how the world works out, we’re going to be in the right circumstances. It turned out that if you invested in Google in 1999, in 2001, in 2004, and in 2005, and you were a long-term investor in Google, you might have had some variability in returns. It was a pretty great investment. Likewise with Amazon, likewise with Netflix, likewise with Microsoft.

So we just think about what are going to be the enduring companies versus trying to time the market cycles. That’s my honest thought. I think a lot of value will be created out of AI. I actually think that, in a lot of the rounds we participated in, they felt eminently rational to us. But I think the most important thing is to be in the right companies and trust that, if you’ve picked the right companies and want to be the multi-decade holders of those companies, the rest will wash out.

And where do you stand on the exit environment for this? OpenAI is one of, if not the highest-valued, private company. Do you go public? Can you? What do you think about this for private companies?

Yeah.

When we invest in companies, it is almost always with the intention that, if they want to, they can at some point be a standalone public company. I don’t know OpenAI’s particular plans for this, but I think we should expect that most great technology companies will be public companies over time.

It will probably take some companies, in some cases, a little bit longer than in the past and, in some cases, maybe shorter. But I think the beauty of American capitalism is that everyone gets to participate in the growth story of these really important economic engines, and I think we think that’s morally the right thing to do. We also think this is how most companies will choose to represent themselves over time.

I’m really curious about this question because Thrive is such a stealth fund, a huge fund, and you do a lot of different things, from incubations to funding very large rounds of companies. How do you think about performance as a team?

Mm.

How do you think about performance as an investor on a very small team?

We think about decision processes a lot more than near-term outcomes. When I worked at Bridgewater early in my career, Ray Dalio had this very famous line: “Focus on the swing, not where the ball goes.”

I actually think that’s very much how we operate. We want to have the shot on goal with the most exciting companies. We want to have the right conversations around those companies. We want to be, as my partner Miles likes to say, paranoid but patient—always thinking about what could be, but not necessarily rushing into things. I think that structure is really what we anchor to.

We have seen plenty of companies that didn’t look like they went anywhere for the first few years. If you were an investor in OpenAI when it was a nonprofit in 2017 and 2018, and even a capped nonprofit with profit-like shares in 2019 and 2020, you might have thought, “Oh my gosh, what is this thing going to do? Is it going to go anywhere?” Of course, it’s now one of the most valuable companies in the world.

By contrast, I'm sure we could think of plenty of companies that looked hot coming out of the gates and at some point looked less good thereafter. Which is all to say, when we make decisions as a team, no matter what performance looks like for this company in the next 1 to 2 years, we're going to be confident that we made the right decision because we looked at the right inputs for this. And so, I think the worst types of investments to make are investments where, if the numbers change 1 quarter or 1 year, you would all of a sudden start to feel bad about the company, because that suggests to me that you're not investing because of the fundamentals of the company; you're investing because of the momentum of the company. And I think that is rarely the right way to capture really long-term value.

We were joking about this in the beginning, but I don't know what day it is, and you don't know what day it is. I thought it was February. It's apparently almost November.

Yeah.

But what are you looking forward to in the next 12 months?

Great question. First off, I think we're in this moment where reinforcement learning, as you mentioned, is this new lever, but we've seen it make its way into very few new software products. I think there are a few products in labs; rumor has it that some of the coding companies, like Cursor, have done some really cool things that they've mentioned on their blog. But relatively speaking, there aren't that many RL problems and projects out there in the world right now in production. I'm really excited to see that change.

I think it's one of these really powerful things where we now have a lot of the inputs required to automate large amounts of really highly treacherous work, and I think we're going to see software companies start to productize that over the coming months and years. So that's one thing I'm really excited about.

I think the other is the potential of intelligent hardware. I think we're going to see a real renaissance of problems that don't just have to be solved with bits; they can also be solved with atoms. Sometimes, to make a real impact on the world, you actually need to move something in reality. We're investors in Physical Intelligence. They're doing this on the robotic side. We've talked about Nudge, we've talked about Mach, and we've talked about Anduril.

But I think we're going to see a new wave of companies that can actually address a large swath of global GDP—well north of 50%—where actually affecting the physical world really matters. And then finally, I think there are going to be some really exciting developments in applying AI to science.

There's Isomorphic Labs, in which Thrive is a very proud investor. It's Alphabet's drug discovery and life sciences AI portfolio. There's a new company called Periodic, which is working on materials science. We are not an investor in it, but it is doing really awesome work on new materials discovery.

And I think there's this cool question: At what point, similar to our discussion on Nudge, do science problems become engineering problems thanks to the power of AI? I'm really excited to see how that transitions and evolves over the next few months.

It's a lot to look forward to.

I know. We're in a really exciting time.

It's really exciting. Philip, thank you so much.

Thanks for having me.