Invest Like the Best with Patrick O'Shaughnessy
· Palmer Luckey
[00:00:00] S1: Oh, this one is actually pretty easy. We decided to build Lattice, the AI core that powers all our hardware systems and a bunch of systems that other companies make now because that was a foundational platform technology that could be reused over and over again. Unlike, let's say, investing $1 billion into one hardware platform, where the value largely stays locked up in that hardware platform, putting $1 billion into something that is inherently cross applicable means you're basically multiplying the effect of that money. It means I get to now use it in 20 products. In a way, it's almost as if I spent $20 billion. Or at least in the current incentive structure of the DoD, that's what other people would have to do to make it happen because they don't want to reuse things where they can justify doing otherwise. Product-wise, that was an easy decision in terms of the things we build on top of that now and what areas we decide to go into. So how do we prioritize? The first step is actually just -- we actually don't have to make that decision is actually the great thing. These priorities are laid out. The joint chiefs have a list of priorities, things that are of utmost importance for the United States to get better at. This is one of the reasons we got into the rocket motor business. It's because there's been huge consolidation. There's only two vendors. Neither are very good. There are no independent vendors that are not owned by an existing prime, which is really bad for startups, really, really bad for the primes that don't own their own rocket motor vendors. And even the ones who do own rocket motor production capacity are not stoked with their ability to produce all of the different things that they need. So that was an area where we didn't have to come up with that thesis from the beginning. You had Congress saying, "We desperately need more rocket motor vendors." You had the Pentagon saying, "We desperately need more rocket motor vendors. Please, won't somebody do something?" And they've been doing so with minimal effect for many years. So when we do something, I think there's four things that need to be true. First, it has to be something that The Pentagon cares about. It has to be something they think is a priority because if they don't think it's a priority, it doesn't even matter if it should be a priority.
· Palmer Luckey
If they're not convinced it should be a priority, it's probably not something that is going to get fixed in the near term. If I think something is a priority and The Pentagon disagrees, that's not an engineering effort I need to deploy. That's probably a persuasion effort that I need to deploy. Don't start the engineering on something when you haven't successfully persuaded them. Most of the time, they know what their problems are. Two, it has to be something that Congress cares about. They have the power of the purse. They decide what is going to get funded. They decide what programs are going to get allocation. They decide which ones are going to get killed. And so if Congress believes that something is important, you're probably going to do better than if it's something that they think is a waste of time or the pet project of some billionaire. That's just the way that it is. This is one of those things where perception is reality. If they think it's not important, I have to act as if it isn't or try to persuade them otherwise. I can't just go out and build the Batmobile and say, "Hey, I built the Batmobile, check out how sick it is. You guys should fund this." That happens from time to time, but those are the exceptions that prove the rule. And then the third thing is it has to be something we can do well, that I know that sounds obvious, but doing well often means, "Can we reuse things we already have? Can we move fast? Is this something we can do without totally reorienting our whole business?" Those are going to be the most attractive things. And then finally, the last category is other people have to be doing a bad job. I don't want to spend our money rebuilding things that other companies are already doing well just because I think that I could do slightly better or that I could just make money in that area in a vacuum. This is probably the least rational part of the process. Could Anduril make a lot of money making small tactical quadcopters for surveillance, things like the short-range reconnaissance drone going on U.S. inventory? The answer is yes, we could definitely make a lot of money doing that. I think we'd be very successful at it. But there's a lot of great companies in the U.S. that are already doing this. Skydio is a great small drone company, BRINC, great company, Teal, great company.
· Palmer Luckey, Patrick O'Shaughnessy
So why would I spend my VC dollars, venture capital dollars, fighting their venture capital dollars, trying to solve the same niche that has already been pretty well solved by other good U.S. companies. I don't want to do that. I could use the opposite of the strategy that I have, what if I only went after areas where there are other good companies already doing that. Maybe that means that those are good markets. But how shitty would I feel if the net result of spending $1 billion in R&D is that I've just killed off 10 other companies that could have otherwise done a good job, and the DoD ends up with the same capability anyway. It's not what I want to do. So building things that other people are doing poorly is actually very important to me. I want to be entering areas where the existing people are not doing a good job. They're charging too much. They're working too slowly. Their systems are not highly functional. Beating them is something I can feel good about in the morning. Beating competitive players, not so much. [00:00:00] S0: Make the world a different place.
· Palmer Luckey
[00:00:00] S1: There are people who are not looking at it that way. I think there's been some defense startups where they are getting into the really crowded areas of the market. Or worse, they're working on things that they think are important, but The Pentagon disagrees. I won't name any names specifically, but I'd say like the entire category of defense startups building things that are for door kickers to better clear buildings. Yes, there is a need for door kickers to clear buildings. But The Pentagon doesn't see that as their biggest problem. They don't even see it as on the list of their top 50 biggest problem. They feel like, "We got really good at that. We have the tools. We have the tactics. We're the best in the world at clearing buildings with door-kicking soldiers, and we don't need to get marginally better at that." Same thing with Congress. Congress does not have an appetite to spend a lot of taxpayer money getting slightly better at something that they believe is irrelevant to a fight in the Pacific, that is irrelevant to a great power conflict with China across long ranges. And so I've seen people who say, "I'm a Navy seal and I know that we can do way better at this. The other people out there aren't making great stuff. And I know that this can be a powerful technology." Ideologically, I am aligned with them. I think that they're right, we could be building much better things for that group of people. They have historically gotten much worse gear than they should have. But as a practical person who wants to have impact, I know that's not where I'm going to have it. I know that I'm not going to have a huge many billion dollar impact on that industry given that the people in charge of running that side of things are not going to agree that it's worth investing in.
· Patrick O'Shaughnessy
[00:00:00] S0: So you built Lattice. Now the world is catching up. Everyone wants to build an AI platform system. What have you learned about building one? What are the components of it? What do you think about AI at large, the cost associated with compute and AI? All of these big things you've been working in for a while, and now it matters to everybody. So what would you contribute as the lessons that you've learned there so far?
· Palmer Luckey
[00:00:00] S1: It's been a double-edged sword, I think. We've been working on AI for defense since literally day one. That was the whole pitch. The second page of our pitch deck was a quote from Vladimir Putin. He was talking about artificial intelligence, and he said, "The country that wins in this sphere will become the ruler of the entire world," which I love. It's a very James Bond villain quote. On the one hand, it sounded crazier at that time because AI wasn't hot seven years ago. There were people who were interested in it, but it's obviously not even a 0.01 of the attention that's being dedicated to it today. The flip side of that is now that everyone is saying they're AI, all of a sudden our message is getting diluted or, "Hey, we've actually been doing AI for defense for almost seven years now." And now everyone is changing to say, "Yes, our systems are all powered by AI. It's all AI-driven." Some of that's true, some of it's not. Now it actually is less differentiated than we were. You have to now get very clear about what the difference between a real usable, fieldable AI system looks like versus strapping together ChatGPT with whatever your quadcopter thing is and saying that it's going to change the world. I do think what's been helpful to us, though, with members of Congress, people in The Pentagon, even investors has been the explosion of firsthand understanding of how powerful AI can be that's been driven by these large language models. Obviously, the things that we are building that fuse data from thermal vision and radar and signals intelligence processors, that then calculate optimal weapons pairing against that target, very different use of AI than a thing that you tell to write a poem about your car. But the fact that every member of Congress has been able to use ChatGPT, the fact that all these people in The Pentagon have seen and used systems, doing things that they never believed a computer could do has, I think, expanded people's minds in general towards the possibility that maybe people can be replaced by AI in certain use cases, maybe there are areas where computers really can do a job as well or better than a person. A lot of skeptics, I think, have changed their minds because they type something into ChatGPT, it did something for them, and they said, "Wow, computers sure are amazing these days.