Lux Capital’s Josh Wolfe & Axios’ Dan Primack

Hello again. So one of the best ways to understand any sort of technological revolution, and in this case, the AI revolution, is to follow the money. And there are few people who follow it better than Josh Wolfe, who is co-founder and managing partner of Lux Capital. Josh, join us. Thanks. So, Josh, I want to start kind of geopolitically a little bit. I feel when I talk to anyone involved in AI, whether it be policymakers or venture capitalists like yourselves, I always hear about this race with China, and it's imperative that the US win this race. Is that the right framing? And if so, what does winning look like?

We are in massive competition. It is true. I mean, if you look at 50% of all of the scientists that are being trained right now for AI are coming from China. You look at domestic workforce, I think we're outnumbered 38 to 37% now of Chinese foreign researchers that have come here that are working inside of our companies. Ideally, we have a brain drain like we did during the Cold War with Russian émigrés, and before that, with Eastern European Jews during World War II. It's better that we got the bomb than Germany did. But winning in the terms of restricting chip access is going to be foolhardy in that people are— necessity is the mother of invention. People are going to invent around it. They're going to develop better algorithms, different chips, different models. And so I think that that is a fool's errand that we will not beat them on. I think the way to beat China, so to speak, is by making sure that every single young person here that is in some way, shape, or form in touch with any kind of model, from visual to sound to audio to chatbots, et cetera, are super versed at a very young age and are using it more productively, because that is where the win is going to happen.

When Deepseek got released earlier this year, you called it a grave national security threat. What was the national security threat if, as you say, like, the tech itself is going to equalize in some way?

Well, some of it was the diffusion of the training and the fact that we're going to spend hundreds of billions of dollars between the physical CapEx on chips and data centers and clusters of hundreds of thousands H100s, which is not really sustainable in my view. And I think I have a sort of contrary take about what happens there. But in large part, it was just that they came from behind and suddenly had a performative model. There is another issue, which is if that model becomes the model that the rest of the world is using, it will approach an asymptote of truth but never really get close to it. So if you search for Uyghurs or Tiananmen Square today, being the anniversary thereof, you will never find that kind of stuff because it is censored. And so I've had these debates with some of the folks. Vinod and I were both in the White House with Jake Sullivan talking about open source versus closed source. Where you stand on the issue depends on where you sit in the cap table. We were very early in Hugging Face and believe in open source and the sort of nature of conjecture and hypothesis and criticism and correction. He was early in OpenAI and believed deeply that a single—

Did Jake Sullivan take a side between you and Vinod?

No, he was most interested really in the geopolitical implications to say what happens with Saudi and UAE and the location or co-location of data centers there and their investments overseas and and things like that. But, um, uh, I think he was swayed by both of us that one, one view for what's better for national security.

Well, I'm curious on the deep state question, where is your mind now? Now we are like 3 or 4 months later, and if you were an investor in some of the big US LLM company, uh, companies, would you still be worried about that? Because you say the, the quantums of capital are extraordinary that are going into this.

My biggest concern in being worried about the global competition is that other countries are using or developing on top of that. So wrappers and other things that they might develop are based on Chinese models instead of our models, because I believe that the models that we have, particularly open source ones, will be closer to the truth. And that to me is important for scientific development, geopolitical analysis, uh, culture, just far better than having Chinese models.

Uh, when it comes to Quantum Capital, uh, you were saying one of— you told me, uh, downstairs we were talking a little bit about data centers and data center buildouts. Do you think we are building out too much? Because the talk is usually we're not building out enough, and you know, every time there's a new data center gets announced people get very excited about it. We spending too much on this?

Well, uh, in aggregate, yes. Uh, I always try to understand what's the consensus view and what's the variant perception. The consensus view is that we need hundreds of thousands of H100 clusters. Jensen and NVIDIA would like you to believe that you definitely do need it early for training. Whether you need it or not for inference is another story. And that what any one individual company does is rational, but collectively becomes irrational. So all the hyperscalers from Amazon, Google, Meta, Microsoft, Salesforce, Oracle, et cetera, all building this out, uh, it will not necessarily persist. And one take that is related to that is the demands for energy, which is presumed that because you need all these data centers, then you need small modular reactors. And so you're getting speculative capital that's going into the energy provision therein. So I think that that whole thing is going to end in disaster, mostly because, as clichéd as it is, history doesn't repeat, it rhymes. And if you look through history, you had Global Crossing with the layout of fiber optic and dark optical networking that got laid and laid. The beneficiaries of the third world that got connected to the internet, not the people that were the early investors. And frankly, the distressed guys who came in post facto and picked up these assets for cents on the dollar. Same thing when you went from servers that were on-prem to colo to cloud. And I think that you're going to have the same phenomenon now.

So are you— so is Lux not investing in things like that? Not like, say, for example, the mini reactors, for example?

No, I am a huge nuclear bull and I've tried to rebrand nuclear energy, as you know, Elemental Energy. And I applaud all the people that are doing it, but I'm just a little bit speculative, certainly on fusion. Forget about it for me, but on small modular fission, Does not make a lot of sense. The thing that I think is interesting—

Doesn't make a lot of sense because you don't think it will work the way that people are hoping it will work, or because you don't think even if it did, that the need, that the demand isn't there?

I think if I was building out large data centers under the belief that there's going to be persistent demand for the next decade, I would be doing it in and around the Permian Basin and figuring out natural gas and how do you get distribution of energy, number one. Number two, I would be building large 1,000-megawatt or gigawatt power plants, not small 150-modular-megawatt plants. And we have a history of doing that, but there's a psychological and political overlay to that. And you're in this catch-22 between the NRC, even though the Trump administration has been super positive on small modular nukes, a lot of that is favoritism and paying back stuff. And I'm just not that bullish. What I am bullish on is another sort of contrarian take, which is if you don't believe that inference has to go to these large data centers, then where is it going to occur? And I actually believe that people have misread or maybe overlooked— there was a paper that Apple put out maybe a year and a quarter ago about models on device. Using memory. And so back in 2015, 2016, we had funded a company called Zoox, which had a pretty good outcome, sold to Amazon, became the center of their self-driving cars. It was then that I saw NVIDIA when it was a $15 billion market cap that then, you know, became what it was as the soul of the new machine for AI, because people back then thought it's just PS5 and Xbox generative image chips. Today, I feel the same way about the memory players. 50% of your inference, I believe, will be on device, meaning your cached emails, chats, phone calls, videos, health data, etc. will all basically be processed on device. but it won't be GPUs that are on device. If you think about SK Hynix, Micron, and Samsung as the 3 major memory players, SK Hynix in particular, usually coupled high bandwidth memory to the GPU chips. But I think that increasingly you're gonna see them on phones, on device, doing 50% of your inference with their other 50% going for cloud perplexity, ChatGPT, whatever in the cloud.

Does that mean as a VC, it's funny, I, God, I'm gonna date myself and probably get the dates wrong, but I feel about a decade ago there was a boom of memory chip startups and most of them ultimately failed. They kind of got to that later stage venture world and then got murdered. They were kind of the early unicorns to die. Are you seeing, do you see a next generation of those coming to challenge Hynix, Sony, et cetera?

Yes, whether, you know, intelligent or not, we are making a bunch of investments around sort of edge inference that are on-device, small models, and memory-driven.

Another thing you guys are doing a lot of investing in is defense tech. You're obviously an investor in Enduro, but a bunch of other things. Venture capitalists in the last 5 months have been flooding into defense tech, a lot of which is, obviously, there's AI as a piece of that. Is the flooding in just because it's become trendy? Is it because they think there's about to become an enormous amount of spending 'cause of Trump's golden dome? Why have venture capitalists who avoided this for most of the history of venture capital now running in?

Some of it is people following success. So, you know, Peter Thiel and I were, 2018 or 2019, Reagan National Defense Forum, Palmer was there in his iconic Hawaiian shirt, cargo pants, you know, flip-flops. Everybody thought he was like a laughing joke, you know, amongst the 5-stars and whatever, but he was a very serious guy. They're passing around pictures of him in his cosplay costumes or whatever, and it followed that classic Gandhi thing of like, first they laugh at you, and then they fight you, and then they join you, and then you win. And that's what happened. My guess would be 1 to 2% of the people that were actually there at that Reagan National Defense Forum in California have since applied for a job at Anduril. It has become successful, not just in valuation, but in actual performance. You have a $2 billion plus top line. There was about $8.5 billion of demand for a $2.5 billion round that just closed at a $28.5 billion pre, and it's a very serious company.

And do you think that valuation would have been significantly lower if Biden or if Harris was in the White House?

No, zero effect of the sort of broader, you know, political or domestic political climate. They have done well under both Biden and under Trump. And I believe that when you just look at the model of funding the R&D and development and the production of products, not as a cost-plus 5-6% contract, but of actually like out of our equity dollars and everybody else's equity dollars to build this, and a warfighter that says I want that, and winning programs of record is just a step change from what the Lockheeds, Raytheons, Boeings, etc. have done. So they will be the next prime, and I think that they'll trade at a multiple that reflects the growth rates, the margins that are distinct from the traditional defense players. I will say, by the way, part of Palantir's ascension— and Karp is a friend— is, you know, uh, some of his, his vibe in the current zeitgeist of sort of anti-woke and speaking to that. But it's also, they've done extraordinarily well as a company. But it's also investor demand when there is a paucity or scarcity of pure-play defense companies. And so I think that there is this an early signal that there is a lot of pent-up demand for that. I will also say that there is absolutely going to be a bubble in aerospace and defense, and that bubble will be driven by pools of capital formed in part because you have— to your point, why are GPs suddenly excited about this? In part because you have success stories that people are following. By the way, not every company is going to end up like Anduril. Most of them are going to be somewhere between frauds and like just failed companies. And there will be 1 or 2 or 3 companies that are big, wherein Anduril, Saildrone, Hadrian, Varda, Impulse, etc. That I think are going to be winners, but there's going to be a long tail of companies that are just shitty or frauds. You have LPs though that are doing a very interesting calculus. Historically, they would say, well, we do ESG or we do DEI investing, or we are comfortable with defense or dual use, but we're not comfortable with offense. Anduril was the first time that we had to make a choice as a partnership to say, are you comfortable being involved in a kill chain? And are you comfortable with the moral decision of that? Which for me was one that said yes.

And certain LPs say we can't even touch it.

Now you're seeing that defense is actually part of ESG. Now, I looked at that back then, 7 years ago, when we first—

It's a little self-serving, but yes.

No, but here's why. Here's why. If you are making a discriminating choice looking at drone surveillance footage of a guy with an AK-47 going to shoot up a school of girls, and a guy with a pickaxe that's coming home from a mine to feed his family, that's a discrimination that you want to be able to make. My view of it is, are you reducing human suffering? And if you have weapons, in a world where there are bad people willing to do bad things against good people, you want good people with better technology that are willing to do bad things, Palmer would say, to commit violence in the name of some of the things that we value. And historically, people were not comfortable with that. What happens next is LPs that were like, no, no, no, we won't fund that, either as a direct company or a venture fund that is doing that, they are comfortable saying, we will invest in a fund of funds that invest in that. So you're two moral layers removed, which suddenly makes it acceptable, which means that you're going to have pools of capital that are formed to serve that demand, and you're going to end up with too much money chasing too few good companies, and you will get a bubble.

Something you said at the beginning of this, when we were talking about China and the US, you were talking about— you talked a little bit about students, and go back. At this point, 100 years to immigration. What, you know, I think of the stuff you invest in, you invest in a lot of deep tech, and so much of it comes from basic R&D that comes either out of universities, academia, or other sorts of things. Where's your head at right now with what we have seen, not just the NIH grants, the halting of foreign student interviews for student visas? Are we basically cutting off our own nose to spite our face right now?

There are a lot of cell phones that are happening, and it's not just this administration, it's been happening for 20 years. Of the last 22 years cutting federal funding for NIH, NSF, et cetera. I mean, I literally, it was 20 years ago when Newt Gingrich was actually calling for tripling of the NSF budget, which was like unheard of for a Republican to do. But he understood the primacy of American science that can attract the best and the brightest around the world to come here, develop here, stay here, and for that science, which sometimes takes 5, 10 years, and amounts to nothing, and then there's some ancillary offshoot of it that becomes the next great company. You look at so many of our publicly traded companies that came from academic-derived science, from Genentech with Conan Boyer out of the UC system to Google itself. It's absolutely critical. Probably 10% or 15% of our portfolio are things that are totally derived from university-funded research. And we recently launched a new $100 million initiative within one of our existing funds.

This is 1-800-SCIENTIST.

Yeah, it was basically a science helpline, because we were getting inundated from friends and peers. And then we said, let's open this up to the country, where people are seeing grant cuts. There's politicization on academic campuses. Where researchers that have nothing to do with some of the stuff, whether it's fighting anti-Semitism or just dealing with administrative stuff, that they are being victims of. And these are people that are doing cancer research, people that are developing cutting-edge new materials. There's national security implications. There's neuroscience researchers that unfortunately had something like neurotransmitter in their grant name. And because the word trans was in neurotransmitter, it was part of a search, and they were getting cut. So there's crazy stuff like that, that the sort of sledgehammer approach, as opposed to a surgical approach. Is hurting American science. And we said, let's step in. And if we might normally do de novo companies, where we start companies from scratch with a scientist, and they're 90% mature, and we're taking 10% science risk, now we'll take 50% science risk. And we'll say, bring your work here, or maybe we can license it, or acquire it, or do a sponsored research from one of our existing companies.

I'm curious. I think— tell me I'm wrong about this— that you, at least in the past, have been friendly with Scott Bessen. Have you raised this with him, that the White House seems to be doing an economic or scientific side—

I have not talked to Scott about this issue. Scott, I think, is one of the great adults in the room. I will say Scott is a student not only of markets and securities and currencies, but also of human nature. And he understands who he works for and his boss. And I don't think that he will do something that upsets the apple cart in a way. And frankly, I don't think any of us want him to because we want him in that job.

Final question for you, and you have 1 minute to answer what is a very simple question.

I'm going to speak really slowly, slowly this one.

So there seems to be a fissure, and this is even within venture capitalists, but even within folks in AI, which is AI optimism and AI doomerism. I'm assuming you're somewhere in the middle of that, but which side are you more on in terms of the future, really the future for all of us?

I am so optimistic just about the human condition continuing to evolve. I mean, there are situations that I've heard just anecdotally of people that went to doctors and they couldn't get a diagnosis, and then they took all their medical records and put it into O3 and suddenly got a differential diagnosis that was more valuable. So I think that it will be used for enormous good. You know, I've got 3 kids and I grew up playing copious amounts of video games and watching a shit ton of TV. And I want them using every AI tool and being versed because that's where the competitive advantage is going to come.

Do you worry? Do you worry that when they're in their 20s that there aren't going to be white collar jobs for them?

I believe that the— okay, so I always like to look at things in the abundance versus scarcity lens. What is, what is abundant? Not valuable. What's scarce? Valuable. What is going to become increasingly scarce is this. Is human-to-human interaction, the care part of healthcare. I will get an amazing diagnosis, but I still— because I am human, because I still want a hug and want to hold hands and want to cuddle with my wife, all those kinds of things AI cannot replace. And so the human aspect of it will become increasingly valuable. I had dinner with Dario at Anthropic, and he was saying much the same thing. Blue-collar jobs are going to be safe. Plumbers are going to be safe. People that are doing Broadway plays are going to be in safe. We were talking to Joseph Gordon-Levitt about the nature of Hollywood and its evolution. We are huge investors in RunwayML, and I believe in them and the moral direction of them democratizing access to all of us to express our creativity. I can't draw for shit. I can't make a movie, but I can use prompts to express my creativity in ways that I never could. And that's a beautiful thing. At the same time, I value the primacy of the human that is performing on a stage and the artistic nature of that. So I would say on the whole, I am wildly bullish. But not ebullient with the ignorance of the things that get lost and fall apart.

That's a good segue, because we have both Joseph and Runway later. So Josh, thank you very much.

Muchas gracias.