Experts are experts in the previous version of the world, not the world entrepreneurs are trying to create. Try radical things, and then occasionally they fail very often, but when they succeed, they change the world. We might displace is not easy, and we fail often, but I always say my willingness to fail allows me to succeed.
Hello, fellow data nerds. My guest today is Vinod Khosla. Vinod co-founded Sun Microsystems and took it from zero to a billion dollars in sales in five years, and then became one of the most successful and most contrarian venture investors alive, first at Kleiner Perkins and then founding Khosla Ventures. He wrote the first institutional check in OpenAI. He spent 20 years making long-shot bets on fusion, biotech, robotics, AI. Forbes named him the number one investor in the world in the 2026 Midas list. Vinod, welcome to Summation.
Well, it's great to be here.
Super excited. I was actually thinking about this, like going back to like Sun, right? You guys started it in like '83-ish, right? I think its first year is 8 million revenue. Second year you 5x'd it to 40. Next year you 3x'd it to 115. Next year you 2x'd it to 240. Then you 2x'd it again to 500-ish. Then you 2x'd it again to a billion, and that's all in like 1988. Like today, people go crazy when things are growing so fast, but I guess we've seen this movie before, like in the '80s. So there's a few companies like Sun that were kind of like the companies of today.
In fact, Sun started in '82. Very contrarian bet. Everybody advised me that people didn't buy big computers from startups. Of course, people didn't in those days, and we said we'd disprove it, and nobody believed in distributed computing. So nobody wanted a network; they didn't see any use for it. And we just jumped in and grew very fast in those 1980s dollars. Probably not only one of the fastest growing companies ever, the number of very large companies that spun out of Sun, Sun founders. The only other time I've seen that is the PayPal mafia. The Sun mafia started a lot of big companies, and then it was the PayPal mafia. So there's something about talent concentration that's really, really important and gives rewards for decades to come.
It's funny because we see like there's just something, there were a few other companies like Sun that were growing like insanely fast in the '80s. We had Apple, we had a few others kind of like growing insanely fast. We have a bunch of companies today growing insanely fast. There are just eras where if you can catch the wave, like it's not like what we're seeing today isn't new. Like we've seen this movie before.
Yes, we have. There had been growth, though nobody's grown to tens of billions of dollars as fast as OpenAI and Anthropic have.
Yeah, that's right. Like that's a whole different level of growth. But even with the dollars-adjusted time frames, you have said that 80% of all jobs you think is going to be done by AI in 2030. Like what does that transformation over the next few years look like?
Well, so let me be clear: what I've said is capable of being done by an AI, which is different than being done by an AI. What percentage gets done depends on how fast it's adopted by the people who could use that. I think that piece, adoption, is misunderstood and may take different amounts of time, and it will also be a huge function of politics. So the technology will be capable of it. But if we elect AOC as president and she says you can't displace any jobs, you'll have some very different consequences. By the way, not all bad. If she says Amazon and Meta can't lay off people, guess what? It advantages startups. There'll be a new Amazon and new Meta with five times the revenue per employee because they have fewer employees, and they will have a huge cost advantage. So lots of configurations possible. My bet is with the jobs will be capable of being done, and then adoption will depend on who gets elected president in 2028 and 2032, and that will say does it advantage startups who have no employees or big companies that have lots of employees and lots of customers. I'm betting startups will have awesome opportunity.
Yeah, it seems like even in like the services businesses, it seems like startups are able to like attack everything nowadays. What do you like, with of the things that are being worked on right now, like what do you think is going to be the most impactful five years from now?
Well, you know, the technology is going to be exponential. My bet is in two years we see more progress in AI, a bigger jump between now and two years than we've seen between now and five years ago. So think 2021, when AI wasn't really a thing. You know, honestly, ChatGPT was less than four years ago. So AI wasn't a thing. We'll see more of a jump between now and two years from now than we've seen. Now, we'll see it in various configurations. The models will get better. The models will become the best scientific researchers. Agents will do their things. Forms of agents will do their thing. So there's many, many dimensions in which we will see progress.
Not only that, many of the assumptions may not be valid. My bet is if I look at 2030, we'll know how LLMs work. Interpretability will be a thing, and we've seen some recent results there where people are starting to decode how LLMs work. So this ambiguity will go away. We are probably one or two technical breakthroughs away from 10x the compute efficiency. If you're taking a billion or 10 billion to train a multi-trillion parameter model, it may cost 100 or 1,000th as much because you got more efficient on compute. Same with inferencing. My bet is the cost of inferencing and the cost of training will plummet with technical breakthroughs we are likely to see the next four, five years. So every assumption, every critique will go away.
Yeah, I mean, I recently just trained like a model on my home computer for a specific model. For me, it was like so cool that one could do that. And I don't know what the cost was, but it was essentially free. I mean, it's unbelievable the things we can do today.
Take a little company. We have this company called Prism ML. They can today put a 37 billion parameter model on an iPhone. Soon they'll put a 100 billion parameter model on an iPhone, so it becomes free. And I don't think a 500 billion parameter model is out of the question because they do a 15x reduction in the size while preserving something like 98% of the intelligence of, say, an open-source equivalent model.
It's pretty stunning. It's also just crazy if you just think back just to a year ago, like how much has changed in a year. I have never seen change like this in my lifetime. Like just the amount of innovation that's happening. It seems like almost every day there's something new coming out. It's really even hard to keep up with what's going on.
And frankly, a year from now, you'll say 2026 was tame compared to 2027.
Yeah, I agree. So you have said experts are terrible at predicting the future. Kind of like they're just kind of like extrapolating the past, especially when we're dealing with these like insane curves that we're having right now. How do you personally like tell a really make a real nonlinear bet?
Well, there's two things. You know, experts are experts in a previous version of the world, not the world entrepreneurs are trying to create. In 40 years of doing innovation and innovation only, I can't think of one big example of a breakthrough that came from a large company or an expert prediction. It just doesn't happen. So it's people who try radical things, and then occasionally they fail very often, but when they succeed, they change the world. And I think that's the model that has caused almost all innovation. Could you imagine Walmart doing what Amazon did when they started?
Could you imagine Lockheed or Boeing do what Rocket Lab and SpaceX have done in launches? Could you imagine a big pharma company doing a COVID vaccine like BioNTech and Moderna did? Could you imagine Hertz or Avis starting Uber? Or could you imagine Hilton Hyatt starting Airbnb? It doesn't matter if we had technical things like rocket ships or vaccines or business model changes. Almost all innovation that I have seen over the last 40 years has come from an entrepreneur, people in your audience. They do all the big society-changing stuff. Big corporations are great at incremental innovation. Intel can go from, or Samsung can, or TSMC can go from a 7 nanometer process to five to three to two to one and a half. But radical innovation only has come from startups the last 40 years. Can't think of one example of a large innovation.
If that's happening and assuming like the rents go to the innovators, why are these mega-cap companies just getting bigger and bigger and bigger?
You know, because they are taking advantage of the current opportunity. Everybody was surprised by how fast AI grew because of that. There was only one option for a GPU, which is Nvidia, so they can get 80% margins instead of the typical 30% in semiconductors. You're right. But when a chip like Jalapeno from OpenAI comes along, and I consider OpenAI a startup, it's one the cost of a GPU because it doesn't have the markup. It's also better tuned to the OpenAI model. So that's probably effectively one-tenth the cost of running an open-source model on an Nvidia GPU to run the proprietary model on a Jalapeno GPU. So one, and we can come back to it, open source will be way too expensive to run. A large proprietary model will be one-tenth the cost compared to a similar-sized open-source model to run, and then you can price it wherever you choose to.
But there will be five Nvidia competitors in five years. My bet is their market share will have to decline and their margins will have to decline. It just chips take five years to develop, three, four, five years to develop, and we haven't had time to respond. And so in AI there'll be startups, people like Cognition growing as fast or faster on the coding side than Claude or Codex. So this game is very early, and the response times are short, which generally favors startups.
Now, in some ways, you're an expert. You've been doing this a while. You've seen so many of these movies before. How do you stop yourself from making the mistake that many experts make?
I spend most of my time with entrepreneurs. I spend most of my time being relevant to a 20, 25, or 30-year-old. I can sit behind my desk and talk at them, or I can just mostly meet a lot of them and listen. I never speak at financial conferences. I only talk to entrepreneurs. I think I'm in the most rapid learning mode I've ever been in my life. Even when I was 20 and in college, I'm learning more every day, every month, every year today than I've ever done. And I think the key is not to assume, you know, it's to keep learning rapidly, and that's what I do. So I would say I spend most of my time with entrepreneurs.
Yeah, and by the way, I think what you just said is really interesting, that you're learning more today than ever. Like it's so much easier today to learn more than ever that if one isn't learning more today, like you have a real problem. Like you don't even want to learn if you're not learning more today than you ever have in the past.
Yeah. Look, somebody asked me about a glioblastoma type of cancer yesterday, and within an hour I knew more than their doctor had told them.
Right, within an hour.
You know, it's just such an opportunity, such a wonderful time to be learning rapidly and expanding your horizons, and you have the best educators in front of you in ChatGPT or Anthropic or Gemini, but any of these models are just fabulous at this.
No, to me, when I think of you, you're just like, you're just an incredibly ambitious person. And when I go back to the '90s and I remember the '90s, like I remember there was like these two super ambitious venture capitalists at Kleiner Perkins. It was you and John Doerr, both super smart, both super, super successful. You've kind of just continued to be the best, while like Doerr, who's like, you know, not that much older than you, roughly, retired and kind of lost the ambition. Why do you think some people like continue the grind?
Well, John's a great thinker, absolutely, who was a great venture capitalist. He's stepped back from Kleiner some and is sort of worrying about other priorities, important priorities like climate and other things, which I also share. So I think the key to life is be interested in something and keep expanding. It doesn't have to be in the same dimension. I've chosen to do it in venture capital, but I've also done it in other areas. I still care a lot about climate, and I think we are very close to solving the climate crisis from a technology point of view, not yet from a deployment and implementation point of view. But we can talk about that separately.
I do want to come back to the question you asked earlier. AI capable of doing 80% of the jobs. I wrote a piece in The Information, which anybody can read, a couple of weeks ago, three or four weeks ago, and it's not behind a paywall. So they thankfully put it outside the paywall so anybody can get it, and it was about what happens in this economy. We might displace 50% of the jobs in this country by 2035 or 2040 in that time frame with unemployment or underemployment from those jobs. But I also define 50 million better micro-entrepreneurship jobs in this blog that we can create. That's an opportunity where people are working for themselves. Most of the jobs in this country aren't very good jobs. You know, if you're a farm worker or an assembly line worker, you're doing one thing for 8 hours a day, bent over in a field in 100-degree heat.
I mean, even if you're an office worker, you're doing like, you know, random, even something highly paid like an investment banker or something, you know, it's not always that glamorous.
Yeah, but look at more than 50% of the jobs are what I call servitude jobs to survival. It's servitude to survival. Not a meaningful job mounting tires on a car on an assembly line for 40 years or till your back gives out, a lot of disability. So if we can replace them with people working for themselves as micro-entrepreneurs because AI takes care of everything except their one skill, it could be as ambitious as surfing or baking banana bread or wood carving. Artists would finally have a way to make a livelihood.
So so many things get freed up. We could free humanity from the servitude of survival and make business much more efficient and abundant. Great. So I do want to stress that. I also want to, and I referred to it obliquely, I said the adoption of AI will be a function of politics. So in this blog I talk about how to address the politics. Today you take somebody like an AOC, she says let's protect the jobs. Wrong answer. I say let's protect the people who might be affected by the job displacement. That's a very different thing. And I define how in a deficit-neutral way you generate four, 500 billion, and then a trillion dollars a year or more to protect the people who are displaced. Easy to go to three-year unemployment insurance if you have that much resources. At the peak of COVID unemployment, we were spending 250 billion a year.
We could do much more than that to protect the people who might be displaced and free up those people to do new things they're passionate about, not the cubicle servitude or the assembly line servitude or the farm worker servitude. So I get very excited about the world that can be with great abundance and protection of the bottom half of the economic strata in this country, and that's where we can make the biggest impact, and we can make primary care free to everybody, education free to everybody. That costs almost nothing for what it costs to handle the bill at CMS for a Medicare patient, just the cost of handling the patient. My bet is you could provide free primary care with no billing for a whole year, just for the cost of handling today's one bill. That's sort of the magnitude of change we are talking about.
There's so many things where AI could make things deflationary, and that would like especially help people who are like the bottom income in the US.
I'm spending a lot of time thinking about the bottom half of the economic strata. There's 75 million people who voted for Trump to get him elected. Those 75 million people make less than 75K a year. And they pay 75 billion in taxes. That's roughly how the numbers work out. First, you should eliminate all taxes. 75 billion in the larger scheme is nothing.
Just to kind of stretch on that, you don't think it's helpful for everyone to pay like a little bit into the system or something? So you think we should have a scenario where like an Abu Dhabi, where like the vast majority of people just don't pay any taxes at all?
You know, for 75 million people to pay 75 billion, you're collecting $1,000 a year. My bet is it's costing a couple of hundred to process those taxes.
Well, with AI, we can get that down to a couple dollars, right? You know, we can get that down to 100%, I assume.
Well, but that's not what we are doing. All I'm saying is they get an immediate bump in income.
And those are the people we want to protect, and then we create micro-entrepreneurship opportunities because every one of those people is a human being with a passion interest. Let them take advantage of that instead of being trapped in a job where they have to pay their rent. They probably don't even have a mortgage. So I think freeing up those people and generating enough income or revenue, and I can get to in this thing I define how we do $400 billion of revenue generation in a deficit-neutral way to protect all the people who might be displaced. Think about it. That's so awesome. And it'll help the rich too, the upper half too, because there'll be less resistance to AI and faster economic growth.
Well, so I propose, you know, today AI is helping capital investments a lot. Look at the data center returns on trillions of dollars of investment, and hurting labor by suppressing wages and wages for young people. If we levelized all income taxes today, capital gains is a way to redistribute income upwards towards the rich. They get a lower tax rate, as Warren Buffett says, than the ordinary and then a secretary. If he said all income is the same, we're not going to favor capital gains over ordinary income, you generate enough money to pay for everybody's displacement and three-year unemployment insurance and a bunch of other things.
Just to jump in on that. So, you know, one of the nice things about people who pay capital gains is they only pay the gains when they take the gains. And, you know, for many of us, we rarely take the gains. So we don't pay the gains that very often, only when we need to, you know, buy something nice or something like that do we do that. So how would you kind of like square that circle?
Well, there's many, many ways to address that, but fundamentally, and we don't have enough time on this podcast to go into the details, but fundamentally we need to stop advantaging the rich. 40% of capital gains is paid by people making more than 10 million a year. They don't need that extra tax break compared to their assistants and secretaries and farm workers. We do not need that break, and we ought to find a way. In fact, this is the way to do it. It's a deficit-neutral way so we don't increase deficits.
What you're saying is that like we lower the income tax and we increase the cap gains tax?
I think we eliminate income taxes for people making less than 75,000 a year.
We leave the rest of the taxes where they are today for ordinary income and treat capital gains as ordinary income. Why should we give preference to one sort of income? It used to be when there wasn't enough capital and you wanted to encourage investment, you gave capital a break. I do not think we are in that situation anymore, and many of the places capital used to go, like China, it's not going there anymore because it's not a safe investment.
Let's say you live in California, so like, you know, let's say income tax in California is like roughly 54%. And so if you have to pay the income, like you're a high-earning, you know, doctor or lawyer, you have to pay it. But if you don't have to pay the cap gains, if you can defer it, like you're almost certainly going to wait and defer, like it's going to be like less likely you'll take the gain if it's at 54% than if it's at 30-ish% or something like that, right? So you can see a scenario where like the government's going to make less money too, right? Or do you disagree with me?
I don't think so. I think you can impute taxes. You can put limits on how long you can not liquidate capital gains. You can eliminate breaks like step-up basis, which every rich person uses to pass stuff on to their kids. Their kids didn't earn the incomes; they did, so they should spend it. There's many structural things.
But here's the important thing: the politics of it works if 75 to 100 million people vote for it. Capitalism is by permission of democracy. So what we are optimized for is what 75 to 100 million people would vote for, not the rich people. What Trump did is win on one issue, immigration, even stories like immigrants are eating your pets. This is a much more serious and better issue for society that whether you're a Republican candidate or a Democratic candidate, you can say in a deficit-neutral way, we can create this pool and eventually a sovereign fund with a way to have enough income. If you're generating imputed income, you can borrow against it. Also, the federal government has low tax rates, low interest rates. So there's many ways to address this.
Sorry, I took you off on this other track, but I'm very...
Interesting. I was very, I didn't know we were going to have a touch on this. This is great, actually. It was kind of unexpected.
Let's go back to your questions.
Speaking of money, like I mean, you've seen so many people who went from nothing to be just all of a sudden becoming insanely wealthy. How do you think money changes people?
Oh, it depends on the persons. The fact is, I heard a number that at the Anthropic IPO, 600 people will make more than 100 million of net wealth.
You know, clearly they'll buy houses. There will be clear trickle-down economics. They'll hire nannies, and there will be a lot of that. I don't think that's a bad thing. Especially entrepreneurial-driven value creation is a very good thing for society. Makes the society richer. As to individuals, some do it very well. There's a couple of hundred people in the Giving Pledge who have said they'll give most of their wealth away, the majority of their wealth away, and others who want to hold it or use it in other ways, which is fine too. But I mean, I do think they shouldn't...
necessarily be able to pass it on to their kids, which, you know, they'll help their kids anyway. Their kids start on second base or third base. You know, if parents make $100 million, their kids do start on second base or third base.
Yeah. They've got all the things they need anyway to do well. Yeah. And also hopefully a lot of these people will live hundreds of years old too. So,
and to be honest, I'm not even talking about inheritance taxes, though. I saw an estimate something like over 30 years $75 trillion of wealth will be transferred from one generation to the next. I mean, that could pay off our deficit if we tax that.
Yep. Yep. So, but we shouldn't take away the incentive to work. I think that's very key. And most of what I suggest doesn't take away the incentive to work.
You've got this kind of like approach when you're betting on things like kind of like a black swan approach where you're making these like massive asymmetric bets. A large portion of them fail. Every once in a while one works in just an insanely great way. Like how do you think about that as a capital allocator?
I don't think of myself as a capital allocator at all. I've never in 40 years I've not even called myself an investor venture capitalist.
Ever. Other people have called me that but I haven't ever said I'm a venture capitalist.
You think of more like a technologist and just kind of trying to understand the future.
I'm a technology-based value creator. I like to create economic disruption with technology and that's both fun and a real contribution to society. So, you know, take OpenAI and I tend to think very long term. I'm sitting in a house I bought in 1986 and built it to my liking. I'm very long term. I planted trees I wanted to see in 40 years and I'm seeing them today at home.
And I love trees. Outdoors is my happy place. But when it comes to take something like OpenAI, I sort of ask what is the right thing? People focused on it was a nonprofit. They didn't have a business plan. They didn't have a product plan. They didn't have a revenue plan. I focused on if AI works and I was convinced it wasn't a question of if but when, what will be the impact. Whether I got a 100 or 500 or a 1,000 return over 5 years or 15 years, I'd still have an adequate IRR.
So I didn't, I almost never calculate an IRR on any investment. In fact, I can't say I've looked at an IRR calculation in the last hundreds of investments I've made, but I say what's the impact if AI works and then how can I make it work. And the same year 2018 when I committed to Sam to invest in OpenAI long before AI was a thing, before transformers was a thing, before LLM was a thing, I just bet on a great team going after an ambitious goal to create AGI and I could see the impact. In 2012 in TechCrunch I wrote about do we need doctors, my bet is medical expertise should be free.
