It's election night. It is election night. The early signs are the Democrats are making big gains. How would a Democratic majority change tech and the way you invest?
Well, I think the most important thing is we'll constrain Trump and handicap him. I'm actually a registered independent. I've never been a Democrat my whole life. I used to be Republican on strictly fiscal grounds and then on climate issues became an independent. I think the level of trust in the US that has been lost is probably the biggest and longest lasting damage Trump will have done, and hopefully the next two years will show we constrain him somewhat from random decisions.
What are some of the policy goals that you would want to see advance under a Democratic majority in Congress?
I think both parties have taken similar approaches to China, which I think would be very, very good to have consensus of both parties on, and I do think we had that. The Biden administration had similar goals. I think AI is one I'm actually happy with the Trump administration's policies on. Pretty forward-looking and very much recognizing the competition with China.
That's to say, a lighter touch approach.
A lighter touch approach, yes. But also a real commitment to winning against China in the AI race. We can talk about why that's important, but it is important.
But on open source models, that's an area where you would like to see more regulations. Is that right?
You know, there's a lot of misperceptions about open source. So, how many people think open source is cheaper to run? It's just plain wrong. Open source is way too expensive to run compared to closed source models. Open source is way more dangerous than closed source. In fact, it's the one area I've recommended to David Sacks and the White House generally to do a treaty with China. You can't trust China, but it's one area where constraining open source to not be able to do wild things like design another bad virus is in the mutual interest of both the US and China.
Do you view that as a viable or possible outcome? Is there a political appetite in both countries for that?
I think so. I think because it's in both parties' interest to not have models go wild in what they can do in the hands of individuals. You know, this is not just a nation state thing. An individual at home could design a bad virus and release it. And so, it's in both parties' interest to not have that happen. In open source, it's very easy to do. In closed source, you know, OpenAI, Anthropic, Google know what's going on, what people are using their models for. So you try and do something bad with their models, they know. Open source, you can run anywhere. So there's no way to prevent the safety and harnessing to be kept in place.
Is it too late at this point given how far development has come?
I don't think it's ever too late and I think there's always solutions. So if you constrain where open source runs on certified clouds where you know they can't be used for certain bad purposes, that can be done both in the US and China and in everywhere else.
Do you know if those conversations have started? Is that something that we're making tangible progress towards?
I don't know the answer to that.
No. To pivot a little bit, I want to talk a little bit about AI's impact on jobs. You've talked about AI performing 80% of tasks in 80% of jobs, and that seems like a lot of lost jobs. How do you see that playing out when public sentiment around AI is already pretty fragile?
Yeah, look, the fact is AI will be capable of doing 80% of all jobs or at least 80% of 80% of all jobs.
And that's by 2030, you think? What's your bet?
Yes. Or well before then. Now, capable of being done is different than being done, and that'll take deployment will take longer or adaptation will take longer. One of the interesting things is, and you see all the sentiment against AI, and it's real. It's because people are afraid of jobs being lost. I think what that will do will have the perverse advantage, good for Silicon Valley, of advantaging startups over big companies. If an incumbent can't lay off its employees, a startup can compete very cost-effectively with them. I was just talking to somebody with a $1600 million run rate in a financial services business and they had two people in accounting, two, and this is a complex regulated lending business. So if Bank of America can't lay off employees, but the startup doesn't have to hire them, it gives the startup ecosystem a real advantage if we prevent job losses.
So yeah, play that out. What do you see as the ripple effects of what does the American economy start to look like if that's the case?
I think it gets renewed with newer companies instead of older companies which are more constrained by politics and politicians. That's sort of what happens. Some areas like manufacturing, because you're talking about moving manufacturing from say China to the US, it's an advantage because you can do it without having the low labor costs that exists in other parts of the world. But in most areas where expertise counts, then new startups will have a huge advantage. So I think the rate, if I were to say by 2035, the Fortune 500 has had a steady rate of extinction. You look back last 20 years per year, extinction rate has been pretty consistent. That'll accelerate dramatically, I think.
And you've also put forward the idea of taxing, you know, when you think about some of the wealth disparities that exist and that may continue particularly as AI creates new fortunes. You've talked about the idea of taxing investment gains as ordinary income and getting rid of preferential treatment for capital gains. How would a pool of money that that would create, how do you view that as being dispersed? How would that work?
So the real question you were asking is how should we protect the people whose jobs will be displaced? One is to prevent them from being displaced, which is the AOC solution, and that'd be a really bad idea for the US economy and we lose comparative economic power globally. So what's the solution? The solution is instead of protecting jobs, you protect people, the people who are displaced. It's a very important distinction to protect people who have fear of being displaced instead of protecting the jobs they're being displaced from.
You have to look back at fundamentally what AI is going to do over the next 5, 10 years. Really advantage capital relative to labor. Labor will have pressure because expertise and labor becomes cheaper and cheaper with AI, whether it's robotics or expertise, all trend towards free. So labor is at a disadvantage. Capital's getting great returns. Our system has a huge advantage for capital over labor already. What is capital gains tax? It's a redistribution of income away from labor towards capital gains towards the rich. It's literally redistribution upwards. So we said there's no capital gains, there's no ordinary income. It's all the same income. You tend to generate about my estimate about $400 billion or so a year. That plus a few other tweaks I can go into in some detail. I had an op-ed in The Information recently on this topic. You then have this large pool of capital that can be deficit neutral and protect people who are displaced. So simple idea: give three-year unemployment insurance.
I was going to say how do you see it being doled out and what is the period that you give people to readjust to that new economy?
Yeah. So look, it depends on the rate of displacement of people, but we create a fairly large pool of capital by equalizing capital gains and ordinary income.
And do you think this can happen in the United States? I mean do you think we'll get to a place where there's enough consensus to make a change like that?
Tonight is election night. Look at my skepticism. Look, Trump got elected on an idea people thought was ridiculous, which is immigrants are eating your pets was his fundamental tagline. As ridiculous as that sounds, he won it on that basis the last election. Why can't somebody say, I'm going to find a way to appeal to 75 to 100 million voters? First, 75 million people make less than $75,000 a year. If you forgive taxes for that, you'd forgo 75 billion in collections, which is a small part of what you could gain by having 400 billion extra collections on capital gains being the same as ordinary income. So, you can appeal to 75 million people that way. So, I was counting votes. I see. Say, how do you get 75 million people on your side? And frankly, the fundamental notion that capital gains should be taxed at a lower rate is unfair because 40% of capital gains is paid by people making more than 10 million a year. So why are we advantaging those people? And when we want 75 million or 100 million votes in our favor, you can work the math easily. I started with how you work with the right votes.
Should a 2028 presidential candidate adopt this policy that the rich shouldn't have an advantage and a lower tax rate?
Absolutely doable, more feasible than immigration would have been four years ago or two years ago.
Okay. So turning to the AI race, you were one of the first venture investors in OpenAI. It recently did a first close of a round that didn't quite meet its expectations. There's been a broader tech sell-off in the market. Are you worried at all about the durability of the AI trade at this point?
No. There's two separate things you have to look at. One, what is the use of AI and what do stock prices mean in AI? You know people mistake what bubbles are. Are there bubbles in usage of the technology? And I'll give you a very precise example. Or are they bubbles in people's perception of what the valuations are? Perception of valuations isn't reality. And investors bounce between fear and greed all the time. So why are we paying attention to that? If you look at the dot-com bubble, the last big bubble, if you look at stock prices, you can tell the bubble. If you look at internet traffic, you can't tell the bubble. You can't detect it. So internet usage is what counted and it kept going up. I measure one thing: is the amount of inferencing, AI inferencing going up or down? And is it a bubble? The answer is absolutely not. I can bet my life on it'll keep going up in a steady way. What people associate with that in terms of stock prices, optimism and pessimism will go up and down. But that's not reality. That's Wall Street perception. It's mostly unimportant in the picture when you look at 2035. So if I'm looking 10 years out, will it matter where stock prices are up or down? No. Will it matter, and maybe it will to a degree in capital formation, but mostly not. Will it matter in the level of inferencing we use as a society? Yes. So that's the reality is number of API calls or inference calls, not stock prices. So let's not forget what's really important, which is the usage of AI, which is the number of API calls or inference calls.
And do you feel, I mean, if AI didn't get smarter than it is today, where are we on the curve of taking full advantage of it in your view?
Oh, I'd say even today most people who are using AI are using 10% of its capability. There's probably a small fraction that are using most of its capability. So we're very, very early in learning how to use AI despite the rapid growth in AI revenues, Anthropic, OpenAI and others. So you know it'll take 10 years before we're fully using the capability. And when I say AI is capable of doing these jobs, yes, soon. But when is it fully deployed? Not for a while.
I see. Anthropic's IPO is approaching. Did you have a chance to invest at any point? Do you feel that you've missed out by not?
No. We bet on OpenAI and we stayed with that bet.
But you've invested in competitors and other sectors.
No. This is a place where we've chosen not to do it.
Very simple reason. What matters in AI is having insights into where the world is going. By being close to OpenAI and being pretty clear I'm not investing in competitors, I have much more insight than I would have had because I get a lot of insight gets shared which I wouldn't have, and I'd rather have that insight than the additional profits from Anthropic.
Tell us more about that. What are we talking about? Is that where industries are developing or what?
Take the example of coding. Okay. Four years ago or so, I think it was three or four years ago, we invested in Cognition. At a very high valuation, I needed insight into the value of coding and how big a market could it be. It would only have made sense if a coding startup like Cognition could be worth a hundred billion dollars. Now, if I didn't have that insight, I couldn't have, I shouldn't have invested at two billion. But because I did, I was able to invest, and of course it's taken them, I think this is public data, to get over a billion dollars in less than two years.
