Investing in the age of AI, part 1: Vinod Khosla

0:00Starting a business comes with its share of ups and downs, which is why staying true to your vision is essential. a non-negotiable for Romeo and Milka Bregali, Capital One business customers and co-owners of Ra's plant-based restaurant in New York. Romeo and Milka took a leap of faith when starting their own restaurant, gutting an empty space and building it from the ground up, every pipe, every wall, every detail. But building from scratch came with a heavy financial burden, which is when they turned to their Capital One business card. With the flexibility of the card's no preset spending limit, they were able to spend more and earn more rewards while bringing their vision to life.

0:36Today, Raz's success is proof that with passion and the right support, it's possible to make your dreams a reality. Learn more at CapitalOne.com slash business cards. My life is mostly about an addiction to learning new things. Somebody comes, teaches me about fusion. Somebody else teaches me about cell therapy or AI. It's a wonderful life if you can just keep learning. Why does the market in 2025 affect my thinking? It doesn't. I don't care about tariffs. I don't care about what the stock market is doing. I look at the stock market probably once a month or five minutes. That's all. I've learned everything's possible, as flaky as that sounds, if you approach it the right way.

1:33Vinod Khosla is one of the leading investors and technologists of our time. In fact, he's been consistently on the Forbes Midas list, which is like the Oscars for venture investors. He was an early innovator in the world of computer chips and has since made his name investing in some of the most successful companies like DoorDash, Stripe and Headspace. Notably, he was an early investor in open AI. For the next two episodes, we're diving into questions around investing in the age of AI. And this week, we're kicking off with Vinod. We'll get into his investment thesis, his utopic vision for the future, and how founders can build a winning strategy.

2:16I'm Rana El-Khalyubi, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

2:33But before we get into my conversation with Vinod, I want to take us to a story that's been making headlines this past week. If you haven't heard, there's been some changes at OpenAI, including an announcement for a new business restructure and a push for further global reach. Here to help us break it down is Jeremy Kahn. He's the AI editor at Fortune. Hi, Jeremy. It's great to have you on the show again. Oh, it's great to be back, Ronna. So we invited you on the show today to talk about some news related to open AI and specifically their business structure. So I want to start there. What exactly is their structure?

3:13Are they a for-profit? Are they a nonprofit? And has anything actually changed over the past week? So I'll answer the second bit first. Nothing has changed yet. But what they are, and it's a little bit confusing, is they're a nonprofit entity that owns a for-profit arm. And the OpenAI company, all the employees, they all work for the for-profit arm. But the for-profit arm is controlled by a nonprofit, and the nonprofit has a board. And that board is ostensibly in control of the whole company. Now, they would like to restructure that and have been talking about restructuring that for some time.

3:53They want to convert the for-profit company into a public benefit corporation, which is also a for-profit company, but a particular type of for-profit company where the board of directors has a kind of dual fiduciary duty. They have a fiduciary duty to shareholders to make profit, but they also have a fiduciary duty to some sort of social cause that can be set by the charter of the organization or by the board to serve the public good in some way. For a long time, though, they were saying that what they wanted to do was have the for-profit, the public benefit corporation, not be controlled by the nonprofit.

4:31And what they announced last week is that they are abandoning efforts to have the for-profit entity escape the control of the nonprofit entity. They've said they still are going to convert to a public benefit corporation, but that the nonprofit will have a controlling stake in the public benefit corporation. You might look at that and say that's good for all of us. I think if you have been an investor in the for-profit arm of OpenAI, you might be less thrilled about that outcome because there is some tension there between the things you might do to make sure that this technology benefits everyone and the things you might do to maximize profit for investors.

5:12Yeah. I mean, one thing that is interesting about the structure is the board of governors of OpenAI are mostly independent directors and they do not have an equity stake in OpenAI, including Sam Altman, right, who only has a stake through his investment via Y Combinator. That is very uncommon and unusual if you think of tech companies, right, where the majority of the board is usually the investor. Yeah. So the structure has been very unusual. And one of the things that they're trying to do with this conversion to a public benefit corporation is actually make the structure more like other companies.

5:47So you're right. Right now, Sam Altman has no equity stake. And the other investors in the for-profit arm that, like Microsoft, also do not, they don't have traditional equity, but they do have these other things called profit participation units that give them a right to a certain amount of the for-profit company's profits, if it makes any. You have to remember that OpenAI is losing a tremendous amount of money right now. But if it ever makes any money, those investors are entitled to a share of the profits up to a certain capped threshold. What's going to happen in this conversion, if it actually takes place to a public benefit corporation, is that the investors would get traditional equity.

6:28If the company's value went up 1 ,000 times, the investors can make 1 ,000 times their initial investment. But also, if it declines, they would lose that amount. So it's much more of a traditional equity structure. So fascinating. So Elon Musk has been one of the most vocal voices against an open AI restructure, and he's actually suing open AI. He's been suing them for a while now. Any updates on that? Yeah. So that lawsuit is ongoing. Elon Musk's lawyers came out and said this didn't really change anything about their lawsuit. They still feel like this is an attempt to elude, ultimately over time, the control that the nonprofit would have, that this is not in the public interest.

7:11This can go ahead with this public benefit corporation setup. So they are continuing to pursue their lawsuit. So we'll see. Yeah. Yeah. So OpenAI's mission statement is to ensure that artificial general intelligence benefits all of humanity. I personally think that you can marry purpose and profit, but I wonder if there are any concerns about maintaining this mission in the current structure or if they convert to the public benefit structure as well. Yes. So look, I mean, there are a lot of concerns around how exactly you would marry these things in either construct. The big concern with the nonprofit has been, does the nonprofit care enough about the kind of for-profit mission, or was it just focused on the kind of research mission?

8:01And then it's always a question of like, who is on that board, right? So the nonprofit board initially had a lot of people on it who came from sort of the AI safety community. And that was the case up until the board of attempt to fire Sam Altman in November 2023. And then, of course, he was rehired. And part of the negotiations that led to him being rehired after the one weekend that he was ousted, part of those negotiations, they replaced a lot of people on the nonprofit board. And now you have people on the nonprofit board like Larry Summers and Brett Taylor who come from much more of a traditional board background and probably can't, you know, do have some balance of sort of the for-profit concerns and the public benefit concerns.

8:46Now, the new structure also has potential issues, as they've said that the nonprofit will initially have this controlling stake. But they have not said that like over time, if they continue to raise more venture funding, that that stake won't be diluted. So yeah, I think there's still some concerns even with this announcement about, you know, what the future will actually be. Yeah. Okay, so in other OpenAI news, there's an ongoing global race for AI. And OpenAI announced that they have a new global initiative to partner with countries to build data centers for localized versions of chat GPT. And Sam Altman actually spoke at a Senate hearing last week saying that mandating government approval for AI software would be disastrous for the United States.

9:33So where does OpenAI stand in this global race for AI? I mean, I guess it depends how you judge the race. But if you judge it by the broad capabilities of frontier models, then OpenAI remains at the front of this race. Maybe not. But unlike when they came out with ChatGPT in late November 2022, where they had this indisputable lead that looked somewhat substantial. While I would still say they're sort of at the forefront of that race, they are certainly joined up there in a kind of lead grouping with a lot of other companies, including Google and the same thing with Anthropic. And then, of course, we have these several Chinese companies that have done some very interesting things, including DeepSeek.

10:23So I'd say they're still in that lead pack, but they're not sort of the undisputed leader that they once were, nor do they seem to have a very substantial lead if they even have one. Yeah. So it sounds like we will continue to watch this space closely. Yeah, absolutely. It's going to be fascinating. All right, Jeremy, that's a wrap for now on OpenAI News. Thank you for joining us. Thank you so much for having me, Rana. Okay. After a short break, we'll dive into my conversation with Vinod Khosla on all things investing in the age of AI. Stay with us.

11:00If you've spent any time building AI products or leading technical teams, you know this. Transformation doesn't fail because of ideas. It fails because teams can't move together. Enter Atlassian's Teamwork Collection. It has planning in Jira, documentation in Confluence, video updates in Loom, and now AI agents in Rovo, which connects the dots across your work so nothing gets lost. It's one AI-powered teamwork platform designed for how modern teams actually build. Learn more at Atlassian.com slash TeamChanger. That's A-T-L-A-S-S-I-A-N dot com slash TeamChanger.

11:50Hi, Vinod. Welcome to Pioneers of AI. It is a real honor to have you on the show. It's great to be here. All right, so let's dive right in. I want to quickly go over your origin story. So you grew up in a middle-class family with really no connections to business or technology, but then you went on to co-found Sun Microsystems, and it took a very novel approach to the chip industry, and you did really well. My first question, I guess, I'm curious, what was your biggest learning from that experience, and what advice would you give yourself back then? First, I was always interested in science and then technology following that.

12:31But what really was pivotal for me was hearing about Andy Grove, a Hungarian immigrant to the U.S. starting Intel. So I think this idea of role models is very, very important. And I realized how powerful they can be. Just reading about it as a 16-year-old kid really influenced me. And I do think it's pretty important, maybe the most important lesson. It's also true that I like to say most people are limited not by what they can do, but what they think they can do. And so you can do a lot more if you just assume you can do a lot more. And that's been the story of my life. So how was the experience of starting and exiting companies, how did that influence your investment philosophy once you kind of switched to the dark side?

13:30Well, first, I don't think it's the dark side. It can be the dark side if you take an investment approach. My approach is very different. In 40 years I've been doing this, I've never called myself an investor. I would say I'm a venture assistant assisting entrepreneurs building their companies. And what I learned, you know, most of what you run into, you not only didn't know when you started, you didn't know you didn't know. And so I applied that in helping entrepreneurs build their companies, think about it the right way. Most entrepreneurs are experts in one area. They know one thing. Maybe they know AI research, but then they have no idea what a CFO does or what somebody else, what marketing really is.

14:29And so bringing this broad thinking to an entrepreneur and complimenting them and helping them build the right teams becomes a critical part of what you do. And that's one very important lesson from having done it myself. The second lesson is I'm much more comfortable with ambiguity. So most startups, most big startups happen in areas that don't exist. You know, think of Twitter when Twitter started. who knew what it was or what it could do. It's hard to define, and I'm much more comfortable with ambiguity. And I think investors require a lot more. And I'm much more comfortable with the discovery process.

15:21So one, this ambiguity thing, and the other is how do you build a team to go after a venture and how to de-risk it? I would say two simple things. One, you have to engineer the gene pool of a startup to the key risks you're likely to face. The second thing I fervently believe is the team you build is the company you build, not the plan you make. Because the plan can change, and the plan will change if you have a great team. but the team you build is what will end up determining what your plan is. So a couple of very important lessons I learned from starting my own company. I find it hilarious the people I see giving advice to an entrepreneur on how to build a company when they've never built a company.

16:17Yeah, absolutely. There's a lot of shared experiences when you've walked down that path and a lot of learned lessons, right? Yeah. I am so curious about engineering the gene pool of a startup. Can you give us an example? So when you're doing a startup, you want to clearly define risks that you're going to face. Clearly define the opportunities also in that startup. But the thing that will cause you to fail are the risks. The weakest link in that chain and one weak link can destroy your startup. So what you do is say, what kind of background de-risks that risk? And so my process looks something like this.

17:07Find five companies that have worked on that particular risk. At each company, find three or five people who worked on that risk. Now, for each of your risks, you have 15 to 25 names, and you go after those people. First, you'll understand those risks better from having interviewed all these people or chased them, as the case may be. And second, you've assembled then a team that's specifically geared towards the risk you might face, and that you do for the known risks. So I hate the platitude you often hear of hiring great people. I've never had somebody come in to me and say, oh, they're only hiring bad people.

18:01Right. So what does it mean hire great people? How do you determine who are great people? That's a much more important question that's useful to an entrepreneur. But unless you live through hiring and making hiring mistakes, you're not going to know who the right person is. Yeah. Can we talk about what qualities do you look for in a founder that makes you excited to invest? Well, if I'm looking at even a YC founder, if I'm looking at a YC company and say, the most important question I ask of the partner at YC, how much has this person changed or evolved? Even over three months, you can tell a lot about what their learning rate is.

18:56So learning rate is much more important than absolute knowledge. And because over the next five years, they'll keep learning if they're good learners and open-minded learners. If they're closed and they think they already know the answers and they're not looking for feedback, they generally will fail to evolve as the world changes. And they will fail to adapt as their assumptions turn out to be right or wrong. The best thing you can do is find where you're wrong in your assumptions as quickly as possible. And slow learners don't do that. And rapid learners do that really, really well. You were one of the early investors in OpenAI.

19:40I believe you committed a$50 million check in 2018 and actually made the investment in 2019. It seems so obvious in retrospect, but I'm sure it was not back then. So why did you invest? I will tell you first how uncertain it was. It's the funny story. In the 20 years of Coastal Ventures, it's the only time I've sent an apology letter to our LPs when making the investment. And in 2018 or early 2019, I sent an apology letter saying, this makes no sense. It's a nonprofit. It makes no sense. There's no product plan. There's no business plan. There's no revenue plan. But we're going to make the investment anyway.

20:28I didn't ask for permission. I was just informing him. I knew it looked ridiculous. And many people told me it looked ridiculous. I won't name names. it'd be fun if I did. So why did I do that? I have to go back to another story before I get to AI. At Sun in 1982, we adapted TCP IP, which is the protocol of the internet, on which internet communication runs for your non-technical listeners. And I saw it grow over time. In 1996, I saw we had passed the flat part of the exponential growth curve in TCP IP. We started a company called Juniper to build a TCP IP router for the public networks. And every customer I talked to, from AT &T to Verizon, every single one, with one exception, which was a startup, said they will never use TCP IP in the public networks.

21:34Never. From 1996, nobody was, no major telco was planning on TCP IP being the core of the internet. But I saw the exponential. I didn't follow the experts because they were experts in a previous version of the world. I knew if the world was going to change dramatically, which the internet was going to do, and we'd seen the exponential. I was going by the data, not by what experts were telling me or big companies were telling me, exactly what happened. As a result, the$3 million investment produced a$7.5 billion return in an era where nobody ever got a billion dollar return. Why? Because we built it and they came, right?

22:25It was belief in your fundamentals and I was looking at the data. Why is it related to your open AI question? The year 2000 is the first time I mentioned AI in a public interview with the New York Times in a very vague way. I said AI will have us redefine what it means to be human. So I was tracking the data on the progress of AI and the exponential curve. In 2012, a dozen years later, I wrote two blogs called Do We Need Doctors? They were in TechCrunch, and Do We Need Teachers? The idea being these two basic human needs services, positions for everybody on the planet and teachers for every kid on the planet could be done by an AI.

23:16That was 2012. I kept seeing the progress in how AI was catching up to human levels of performance, even when it was far below human levels of performance. And by 2018, when I talked to Sam, I saw so much progress. It was clear we were going to have breakthroughs at some point. Didn't know whether it was in three years or 10 years, didn't matter If the effect was large enough, and I'd seen the effect of the internet being large enough, and companies like Google and Amazon created on that platform, it was clear it was going to be profitable if I was right that there would be breakthroughs and they would enable capabilities that we see today.

24:06So you sort of build on first principles, not on what experts are doing, saying, or what other investors are investing in. You look at the fundamentals. Yeah. So I want to come back to this idea that AI will replace jobs, including teachers and doctors. Bring this to life for us. So it's been more than a decade since you kind of published that TechCrunch article. Where are we with that, and what does that look like? Well, so there's two modes in which you can use AI. The most famous mode is the Microsoft Copilot for software programmers. And Cursor has done that much better than Microsoft. But the idea that copilots help a human do their job.

24:49We generally, two years ago, switched to a different model. We don't try and do copilots as an end goal. We do workers. That's been our investing philosophy the last two years, three years. We switched to this worker, builder AI worker, not an AI co-pilot. We're making enough progress in the next five years. And let me explain it this way. Next five years, most of these AI workers won't be good enough. So when you get an AI accountant, we sell it as an AI accountant intern as if a CPA firm had gone to your local community college and hired an accounting major to go do accounting, like audit or close reconciliation of your books, whatever.

25:46In these firms, there's always a senior accountant supervising the junior fresh kids they hire out of college. Well, they can also supervise AI interns. And in five years, our AI interns will grow up to be senior interns. And as they get more confidence in these interns, they will let them do more just like they let humans do more. Same thing with doctors. You don't let an AI go diagnose and prescribe. In fact, it's not even legal. But you can give each physician in this country five AI interns who do most of the work, but they get to supervise the interns. They get to practice that the AI interns practice under their license.

26:35And so they're okaying the prescription or the diagnosis or whatever the AI is doing. I am seeing these new business models where AI companies are charging for like AI headcounts, as you said, like an AI healthcare administrator or a therapist. And even like I have a, we'll talk about robotics next, but I'm invested in a company where they literally have names for robots, right? I picture org charts where, you know, the org chart is a combination of humans and AI, right? Yeah. So I first talked about a challenge on Twitter that I retweeted about a couple of weeks ago. I said, when will we have the first billion dollar revenue company with only 10 employees, 10 human employees?

27:23It could be a thousand AI employees. I think that company will be started now. whether it's already started or starting in the next year or two, I think that will happen. It's very, very disruptive to think about that. We're going to take a short break. More of my conversation with Fanod in a minute. Stay with us.

28:00Meet Nicole Nicholas, Capital One business customer and co-owner of Ansett Uncles, a plant-based restaurant and community space in Brooklyn, New York, that got its start from a need for unity. The inspiration, it was born from the desire to create a space that felt like home, where we can connect community culture, good food, and come together with family and friends. That's how we birthed aunts and uncles. Nicole and her husband, Mike, were fulfilling their dream of bringing people together out of their home kitchen. But they soon learned that the demand for community was greater than they knew.

28:31It became overwhelming and we were like, we need home, but not in our actual home. We realized that there was also a need in our community for something bigger in our neighborhood. So we had to find a place. Moving from a home operation into a storefront was a huge next step. But Nicole and Mike were able to take it on with the help of Capital One Business. It's not for the weak. As a small business, finding resources is super important because that's the way you'll be able to manage and scale. We would have never done that without having Capital One to be able to help us along the way. The cashback rewards are very helpful.

29:07You know, it just gave us that runway to be able to breathe a little bit. Then you get to focus on the cooking of the food and making the experience great. To learn more, go to CapitalOne.com slash business cards.

29:22So let's talk about robotics and embodied AI next, because, you know, so far a lot of the AI we're seeing is mostly 2D, right? But we're starting to see a lot more physical intelligences, if you like. You predicted that there would be a billion bipedal robots by 2040, and they would play a larger role than the entire automotive industry. So what would a world with a billion robots look like, and what are the applications? So two things. Precisely what I said is somewhere in the early 2040s, we will get a billion bipedal robots. They will do more work than all of humanity does today, physical work.