The Lord of the Rings, my favorite franchise ever. It's having a bit of a moment right now. There are reports of a new video game. There's a new film in the works. Anyone else excited about this? Hello and welcome to another episode of CNBC's The Tech Download. I'm Arjent Carpull, senior technology correspondent here at CNBC. Why was I talking about Lord of the Rings? Well, because the focus of this episode is Anduril. Anduril is a legendary sword from the Lord of the Rings and you'll hear a bit more about that in the episode because I speak to Brian Schimpf, who's the co-founder and CEO of Anduril. Now Anduril is a defense technology company. If we think about the last few decades, the defense industry hasn't changed at all really very much. It's dominated by a few big players building highly complex, very expensive systems over long periods of time, often costing billions and billions of dollars. But that model is really now being challenged by a handful of startups that are popping up. A new generation of companies, really mainly coming out of Silicon Valley with that Silicon Valley mindset, are trying to rethink how the technology is built and how it's being deployed. And instead of focusing just on building hardware, which is obviously a key part of the equation here, they're focused on the software side, artificial intelligence, and how to manufacture this kind of product cheaply and quickly as well. No doubt it has been a state of geopolitical flux. There have been multiple wars around the world. And we've seen during those conflicts that the nature of warfare itself is also changing from these large expensive platforms to more nimble, cheaper, smaller and faster, at times autonomous systems. Conflicts in places like Ukraine, for example, have shown just how important things like drones, data, speed have become. All of this really is pushing defense closer to the world of technology, and where iteration, faster iteration, software scale all matters a lot, and that's really where companies like Anduril come in. And as I said, today in this episode I'm speaking with Brian Schimpf, who's the CEO of that company, about how they're trying to reshape the defense industry and what the future could look like in an age of AI.
Brian, thanks so much for joining me on the Tech Download.
Hey, thanks for having me here.
So, Anduril, Lord of the Rings fans out there, Aragorn's sword in the Lord of the Rings, forged from the shards of Narsil, the blade Isildur used to cut the ring finger off of Sauron.
Well, we're all nerds. And the Lord of the Rings theme is very rich. But the sword, Anduril in Elvish means 'Flame of the West,' which we thought was a pretty good and appropriate name for what we're doing.
Yep. And we'll get into why. Do you have a favorite? I mean, Tolkien named a lot of things, everything from the horses to the trees. Do you have a favorite sword? Is it Anduril?
I think it would be Anduril would be the favorite sword. I'm not a huge expert in all the swords, but it's up there. It's definitely up there.
But he's my favorite character.
Well, so let's get into business now. Anduril has gone from sort of a startup to this $61 billion defense tech company in less than a decade. Over that time, what's changed? Is it the technology, the threat environment, the geopolitics, all of it? What do you think have been some of the biggest factors that have driven that growth?
Yeah, so go back to 2017, right? It was kind of the end of the counterterrorism era. The view was that when we started this, everyone was like, 'Well, there's never going to be wars again. Why would you even work on this?' The idea that China was an adversary was highly controversial. So it was a very different time when we started the company. And you fast forward through a couple of different things happening that I think all converged to make us accelerate a lot faster than we anticipated. Number one was the geopolitical shifts. I think this is probably the biggest. So the US pivot to China caused them to really look at whether the force structure and the way they were operating was going to be successful. And I think the conclusions when they went through that were kind of not great. Aircraft carriers highly at risk, a lot of the air platforms highly at risk, so it's a really hard problem. So there's a huge re-look on the military side. Then on the geopolitical front, you had the conflict with Ukraine, the prominence of drones and the Iranian Shahed drones, which are effectively low-cost cruise missiles that are creating a huge amount of issues. So you had all these geopolitical and military technology shifts. And then combined with that, you had two decades of consolidation and just frustration with the current system. A lot of the traditional contractors, if you were to run any betting market on this, what's the odds that this program is going to be over budget, late, and not meet what we wanted? Everyone knows about 100%. So the amount of frustration in the system that the government wasn't getting best value for money or getting things in a timely way was really high. So the opening to try something different was there. I think all those things converged in a way that we were able to be there. We started it when it was highly unpopular to do defense. All our VCs had to change their LP agreements to get people to allow them to invest in defense. It was not allowed. It was only a couple of years ago that we could even have European investors. So it's a very different time, and doing something when it's unpopular tends to be a pretty good time to do it because no one else is doing it, and you get the first mover advantage. So it worked out quite well. We've made an immense amount of progress, and I think fundamentally we've been able to just deliver. My read is delivery is the most scarce thing in defense. If you just deliver on time, do what you say, it's really good. It's really rare. So it's just a different environment than I think a lot of these competitive tech environments where just executing well and being in there at the right time made a huge difference.
How would you describe what you do? Because partly it's around weapons and hardware, but there's also a software component to what you do. So what's the best way to describe what exactly Anduril does and how you're a tech company or a tech defense company rather than just sort of a pure play defense company?
Yeah, so a big part of this is, play it forward 20, 30 years. It feels inevitably true that the vast majority of aircraft, of ships, of land vehicles, these things have to be more autonomous and less crude. You're just going to have this inevitable shift that occurs where it's going to be more robotic. And with that, sensors are getting cheaper. The cost to orbit is cheaper. The cost to build a new aircraft is going to go down even more with AI. So the ability to introduce new platforms goes up. The cost to manufacture them is going to come down. So the reality is you're going to have thousands and thousands of systems on the battlefield. That is an overwhelming amount of information. It is nearly impossible for humans to coordinate that and understand what is going on. So we've built a lot of software around it, we call it Lattice, but really the idea is what is that operating system and infrastructure to enable this modern connected battlefield really look like? And how do we fuse in autonomy and all these different aspects to it so that humans can do what they're really good at, which is make high consequence, high contextual decisions, and we take a lot of the more mechanical aspects and pull it all together. So we've infused AI into every level of this. How do we process sensor data? How do we surface up the right decisions for humans to make? We've made it all very resilient, very interconnected. So that foundational tech has really mattered and it enables us to bring really sophisticated systems to bear quickly, where we can make a relatively simple cruise missile, for example, but we can just load it with this advanced software immediately and constantly enhance it and refine it over time.
When people think of AI and autonomy when it comes to weaponry, people sort of have this idea of weapons that can carry out attacks on their own, for example, or very little human supervision. And there's this sort of idea around it when we talk about AI and autonomous weaponry. What exactly are we talking about from your point of view? So there's nobody wants the insane version of this. Nobody in the military is talking about it. Nobody thinks it's a good idea. But the reality is we already have lethal autonomous weapons. They exist. So you think about a torpedo. How does a torpedo work? You say here's a region, and if anything looks like it matches this acoustic signature in this region, that's a military target. We have anti-radiation weapons where the idea is if I shoot a missile at an area, if something is emitting in that area that looks like a Russian radar, that's a military target. So these have existed since the 1970s. This idea that you're going to have machines in a bounded way make decisions, but you know what they're going to do. They're going to be predictable, and you can constrain them in a smart way. That's always been true for a very long time. And AI enables you to make this more precise, more discriminating, more accurate. So the counter view of 'we shouldn't have AI in weapons' to me is a deeply unethical view where you're saying, 'Actually, we should have dumber weapons. We should have weapons that don't make smart decisions.' That seems like a strictly worse world to live in. But you want to put the right bounds on it, and you want to say these things are still controlled by accountable people, and we still have the mechanisms to say these are predictable, they're tested, we know what they're going to do. They just are smarter. So you're just talking about a sliding scale of how much discretion do we give it. The other dimension to this is you just look at the volume of strikes that the US had to conduct in the Iranian conflict or how much Ukraine is conducting. That is an overwhelming amount of decisions that are being made every day. Using AI where you have the best reviewer possible looking at your decisions and calling out errors in your decision-making, that seems like again a very ethical, responsible thing to do. But again, the idea is fundamentally the AI is there as a tool for the human decision makers. The AI is not a person. The AI is not making these decisions. It is about enabling humans to make better decisions. Because the military has a system that the battlefield commanders are accountable and responsible for what happens for their employment of weapons. And that is a good system. We like that system. And so AI can massively enhance the precision that they can operate at.
You said no one wants the insane version of this. No doubt. And so when you think about what you build at Anduril, is there a red line or something that you know we just won't ever build?
Yeah, so the fundamental thing is we have principles around how we think this should work, which is very much around human supervision of decision-making, robust independent evaluation that we have internally of how the systems are actually performing. There's a set of guiding principles around these things. Then our responsibility as technologists is to also say when something's just not going to work. So we've had people come to us and say, 'Hey, I want to do facial recognition from 5 kilometers away,' and we're like, 'It just literally won't work. It will not be responsible. The error rate's too high. This isn't going to give you what you want. This is not a good idea.' So our view of this is you do not want tech bros in charge of making military decisions. It seems nice to say all these things, but we're not actually responsible for the safety and security of the nation. And when you are responsible for that, the decisions are a lot more gray and a lot more difficult. So me being here and pontificating about all the ethics, it's talking from the cheap seats. The reality is we need to be responsible as technologists, and then the elected officials are actually the ones accountable for making these hard decisions, and we need to empower them with the best information. We need to do it responsibly. But fundamentally, our view is tech people articulating their philosophical positions on how national security should be done, it counts for about zero.
When people hear the phrase AI and war, they're often thinking about autonomous robots making these life and death decisions on their own. But that isn't actually where the transformation is happening. Increasingly, AI is sort of becoming this operating system of the modern battlefield, helping militaries process enormous amounts of information and coordinate these complex operations that humans alone simply can't. But the human is there, of course, making those decisions, being informed by those decisions as well. So if we think about AI, that could mean connecting thousands of drones, aircraft, ships, all of this different data, processing it, bringing it all together and understanding what's happening to allow militaries to respond far more quickly than before. And again, those recent conflicts we've mentioned before, including Ukraine, have highlighted why this matters. Cheap drones, constant surveillance, overwhelming volumes of battlefield data have made modern warfare dramatically more complex. And the challenge is no longer collecting the information. It really is sort of making sense of it fast enough to act. And I think as that complexity grows, that's where greater autonomy becomes less of a choice, really more of an operational necessity. Not because militaries necessarily want to remove humans from the decision-making process, but because they simply can't keep pace with what's happening at that scale of modern conflict. And of course, that doesn't mean AI is going to replace human judgment. Instead, I think one of the biggest questions facing militaries right now is actually where to draw the line. At what point does this sort of AI go too far? How much authority should AI have? Which decisions should always remain with humans? And ultimately, who is accountable when AI becomes part of that decision-making process? Those are exactly the kind of questions that companies like Anduril are now helping governments to grapple with.
So beyond AI in sort of the weapons themselves, how is Anduril as a company using AI in things like weapons development? There's been so many advances on models, and we're talking about digital twins, this idea of being able to create a digital world that represents and test things in that digital world before they get to real life. So many different parts of manufacturing design have emerged in the world of AI. How much AI are you using in terms of actually designing weaponry and future products?
So there's the obvious, coding agents and all that. We've been heavily adopting that. I was talking with another tech CEO who was just like, 'No, dude, I'm a coder again. It's great.' So I've got like three cloud code sessions ripping right now. It's great. So that side's kind of a no-brainer. But then we're seeing a ton of progress on both the mechanical and electrical design. The ability to very rapidly do all the analysis tools that would take like a cluster of machines two weeks to run, like computational fluid dynamics or anything, there are three different companies making AI enhanced versions of this that simplify the process and run it about a thousand times faster for 99% of the accuracy. So it's great. We're seeing a ton on the manufacturing side. There are three different startups that we're working with that are applying work instructions, one of the most tedious things to construct. You literally have highly trained engineers who are sitting there generating PowerPoints by blowing out CAD drawings. You're like, 'Okay, this is a no-brainer AI automation.' So the ability to rapidly introduce new products, to get all of the coordinated work instructions, bill of process, all those things on the manufacturing side is massive. So we're seeing it kind of across the board where we're seeing a ton of benefit there. And then the amount you can collect on just telemetry of how your systems are performing and then to be able to actually mine that efficiently, contrast it to what you have to do normally where it's like an engineer has a hypothesis that takes them two days to run down, great, now I can spin up 200 hypotheses that this thing is generating that it's constantly running down to assess, are my systems working, where are the reliability issues, anything like that. So it is just the volume of work that you can accomplish, the amount of leverage you can get on the engineers and production folks in terms of what they're able to do where they were just constrained by hours in the day before is just massive.
