I had the distinct pleasure of hosting Trae Stephens and Michael Kratsios on a panel in San Francisco in September on the topic of “Rebuilding the arsenal of democracy.” Both panelists have spent real time thinking and working on the topic, and I wanted to avoid asking the most basic questions, questions like:
Is China a national security threat?
Should we build the future in America?
Is American manufacturing cool?
Instead, I tried to dig a little deeper into the defense manufacturing base. How do the co-founder of a major defense tech startup and the former CTO of America think about building American manufacturing capacity? Let me know how I did.
Some background: Trae Stephens is a general partner at Founders Fund and a Co-Founder of Anduril, a defense tech company that specializes in advanced autonomous systems.
Michael Kratsios served as Chief Technology Officer of the United States in the Trump White House. He also served as acting undersecretary of defense, where he was responsible for research and engineering efforts at the Defense Department. These days, he’s managing director of Scale AI.
We discussed:
What’s wrong with the defense industrial base?
How can we use tools like the Export-Import Bank to beat China?
Can cutting Chinese tech out of supply chains hurt American companies?
Will we see more tech talent in the next administration?
This transcript has been lightly edited for clarity.
Let me start with you, Trae. Innovation spend at the DoD is something of a disaster. What specifically is going wrong?
Stephens: I think that the government, largely speaking, has good intentions. It wants to invest in new capabilities, and it knows it doesn’t have access to the technologies that it wants to have access to through the traditional defense industrial base. The problem is it also very deeply doesn’t want to pick winners.
The result is that the money gets split up into thousands of tiny little pots, none of which have any real hope of transitioning into production. It ends up being more like a grad school research program rather than a manufacturing-for-scale process.
When you say they don't want to pick winners, I feel like a lot of people in this room think, “What do you mean? Look at Northrop Grumman! There’s plenty of picking winners.”
Stephens: They don't pick winners from commercial industry. The majority of procurement dollars go to what is generally referred to as the big five: Northrop Grumman, Lockheed, Raytheon, General Dynamics, and Boeing.
For all intents and purposes, you should think of those five companies as subsidiaries of the United States government. They're basically government agencies, just private sector ones. They're directly tethered to GDP growth; they all grow low single-digit percentages per year. They all have roughly the same margin profile, between 7–12 percent on cost-plus contracts. They’re paid for the labor they do with enough margin to cover operating costs, and the government very carefully meters out contracts so that they all continue to grow at roughly the same rate. And, as we all know, a centrally planned economy is not the most efficient way to work. But over the last 30 years, there's been very little appetite to shake out of that system.
Michael, what do most people get wrong about the defense industrial base? You've spent time inside DoD and in the White House. What's a common misperception here?
Kratsios: Unfortunately for many folks who have been part of this wave of defense tech over the last two or three years, they think there's a chance that there's going to be a lot of winners out of there.
The reality at the Pentagon is there have been long-established programs that were created many, many years ago to support small businesses in the United States. Those have been rebranded and recreated for all the different services, and these dollars are going out to a whole lot of companies around the U.S. At their core, these companies are on a very important mission, and oftentimes have founders trying to do amazing things.
But as Trae mentioned, at the end of the day, the conversion of those into real substantial material contracts is still very low. Now, I'm not a venture capitalist so Trae, I’m curious how you think of this newfound infatuation with funding defense tech companies over the last three years. I'm always a little bit curious about where all these investors think these returns are going to come from.
…Trae, where do these investors think the returns are going to come from?
Stephens: Well, on the positive side, I think there's a good cultural shift that's happening. It's becoming more acceptable for the most talented people to work on defense problems. That was certainly not the case when me, Palmer, Matt, and Brian started Andril in 2017.
I think it's good that the Department of Defense is starting to build a head of steam around bridge transition funding and giving more budget and authority to the Defense Innovation Unit, and launching Army Futures Command, whatever that will eventually mean. There's a positive angle to that.
The negative angle is that the way that the tech industry works is on a power law curve, not a normal Gaussian distribution of returns. Founders Fund has been around for 20 years, and we've invested in literally thousands of companies. Twelve of those thousands of companies have returned over 95% of all returns to our investors.
If you think about the big winners in the fund, we were the Series A lead in Facebook. We were the largest institutional investor in SpaceX, Palantir, Airbnb, and Stripe. In each of these industries, they were the winner. Like, if you were a space tech investor and you didn't invest in SpaceX, you probably lost money. If you were a social media investor and you didn't invest in Facebook, you probably lost money. If you're a crypto investor and you didn't invest in Coinbase, you probably lost money. Every single one of these industries has this sort of power law dynamic to it.
A lot of people look at the defense tech industry and they're like, “Well, Anduril made it work. I guess that means that there's a market space now that I can run into.” And, uh, that's great. As I said, I'm glad these companies are doing it. But what if it was just Anduril? What if that's the only company? Well, people are going to lose a lot of money.
We might end up going right back into this cycle where there are just lackluster returns, investors run away, and it takes another 20 years to reset and get people excited about the market again. These moments of hype and boom are not necessarily universally positive.
I don't want to spend the whole interview on defense procurement, but say a little more, Trae. What would it look like for the government to be picking winners consciously, to shift from its existing procurement model?
Stephens: This past April, there was an award of a contract called CCA, the Collaborative Combat Aircraft, which is essentially an autonomous fighter plane. The five finalists for that contract were Lockheed, Boeing, Northrop Grumman, General Atomics, and Anduril. In an old system, it would have been very obvious: they would give the contract to Lockheed Martin and Northrop Grumman. That’s what should have happened. but the Air Force really wanted to do things differently and they decided to give the contract to General Atomics and Anduril. Eliminating Lockheed, Northrop, and Boeing, which is crazy if you think about it.
So there's momentum toward empowering people to take risks that they might not otherwise have been interested in taking, but we need to see more of that.
And I think the most important thing is that there's a difference between a prototype or a pilot — some sort of research and development program — and a major weapons development program. These are generally called DoD 5000 Series contracts. Those are the ones that always go to the primes. That's just how it works. If this were working correctly over the next 10 years, maybe we'd see half of those 5000 Series programs go to tech entrants rather than traditional industry entrants.
That's really what we're building the entire company at Anduril around: being competitive to go out and win those really big contracts, and not worrying about whether we can do a prototype and a video that gets people excited.
Michael, you've spent a lot of time both on the defense industrial base and on other ways that U.S. policy can support emerging tech in the national interest. Talk a little bit about that.
Kratsios: Here’s one way to think about it: we as a country are trying to position ourselves to maintain our edge and continue to deter a Chinese threat.
That goes beyond just the tools and equipment that the DoD has, which are critical for the U.S. to focus on. A lot of emphasis can sometimes go on activities that the DoD can do to try to deter what China is going to do. I think a lot of the coordination efforts are equally important. If everyone's highly incentivized to not enter a kinetic war with China, how do you counter their economic warfare?
This goes back to a lot of the efforts that the U.S. does today on things like export controls on chips, and those are more “protect” actions, which the U.S. has to get right. But the other ones, which I think are equally important, are the “promote” actions: How can we export our technologies to countries around the world and make sure that we are creating the standards for the technology that is most important for the future?
When I was in government, we fought these massive “Huawei wars,” if you will. We were in this unfortunate position where some of our strongest allies had this nefarious technology baked into their infrastructure. We were literally begging them to rip it out.
With emerging artificial intelligence tech, we have a huge opportunity to not make that mistake again. I spent a lot of time thinking about and working on how we make sure that the standard for AI across the world is American. That’s just as important in many ways as building our defense base.
What are your tactical lessons from that Huawei experience? You have to go to your allies and beg, as you said. What does that actually look like? What works and what doesn't?
Kratsios: The challenge that we faced was that there was no U.S. alternative. It was often an embarrassing conversation. We would lay out a very detailed intelligence report on why Huawei was so bad, and they would say, “Okay, well, what do you want us to do about it?” And we would say, like, “Go buy some Nokia or Ericsson and maybe we can help you.”
You often hear the line that the Chinese come to these countries with a briefcase of money, and the U.S. comes with a lecture. It’s this running joke around how the U.S. conducts a lot of its foreign policy influence.
But we're in this lucky position with the current wave of technology: The U.S. is the home of the best and highest-end semiconductors. It is the home of the state of the art AI models. We have a homegrown stack that we can export to countries that are interested in artificial intelligence. It's incumbent on us to use the development resources that we have as a government, whether it be at places like the Development Finance Corporation, the Export-Import Bank, or even the Agency for International Development (USAID). These are pockets of money that are statutorily designed to support American exports and American development abroad. Those are places that we need to really, really focus on.
For those of us who are relative neophytes to how the Ex-Im Bank works, take two more minutes and explain how you use those big pots of money.
Stephens: You mean the International Bank of Boeing? 80% of Ex-Im dollars go to Boeing Corporation. Thank you, U.S. taxpayer.
Kratsios: Yeah, that's a great example. The challenge with these institutions is that for the entirety of their existence, they focused on the export of physical hardware.
The most famous examples are the Boeing ones: where there's a country that has a national airline that wants to potentially get some new planes, and they're essentially subsidized in some sense by the U.S. to buy Boeing planes or other American planes. But mostly Boeing.
But I think what I think really needs to be looked at very closely is, can these institutions be used to help export other things that aren't necessarily an aircraft, a wind turbine, or an oil pipeline?
The most valuable thing that the U.S. has to export is software, and that is something that I think needs to be looked at very closely by these institutions.
Trae, besides the DoD getting its house in order, what else would help industry rebuild the defense industrial base? Leaving aside federal procurement.
Stephens: Well, I think manufacturing is a huge part of this. One of the core lessons that we've learned from the war in Ukraine is that if we ever were to enter into a protracted conflict with another great power, we would run out of munitions in eight days. Not eight years or eight months. Eight days.
If you look at what we sent over to Ukraine in the early days of the conflict, things like Patriot missiles, Stingers, and Javelins, these are Cold War-era technologies that were manufactured, and we literally shut down the assembly line. When we wanted to go back and resupply that, Raytheon had to go and get some of their engineers to come out of retirement to rebuild the manufacturing lines. This is a gutting of the defense industry that's happened over the last 30 years where we ceded global leadership in manufacturing to China. It was supposed to be good for the American middle class, but obviously what happened in much of the country is that a lot of those jobs are offshored.
Really what we lost more than anything else was skilled labor. I started a consumer hardware company three years ago on the side. One of the things that we noticed quickly is that it wasn't just that Chinese labor was cheaper, and that these components were subsidized and made much easier to get from China.
It was that literally, you could not get them anywhere else. There was no alternative. You couldn't pay more to get it made in the United States. The labor force didn't exist to make it in the United States, regardless of how much you were willing to pay.
We're in this place now with defense manufacturing, where we're being asked to do things at scale, with high degrees of precision, and meanwhile, Boeing airplanes have doors falling off mid-flight, and our astronauts are stranded in space for nine months because they messed up one of the critical systems. It's like the Idiocracy version of manufacturing in 2024. We have to fix this from the ground up.
How do you go from zero to one on certain kinds of skilled labor? You have a cold start on being able to manufacture something we haven't manufactured in 40 years or whatever.
Stephens: There's a bunch of components to this. I think this is a great place for the tech community to contribute. Several really interesting companies are working in advanced manufacturing: companies like Hadrian that are bringing autonomy to manufacturing. There's a company called Layup Parts that we invested in at Founders Fund that's doing carbon fiber prototyping, which is really important for aerospace. There’s a company called Dirac that's automating assembly instructions: you pass it a CAD file and it generates the assembly instructions. There are things you can do that help get back on the horse with manufacturing.
But our approach to this at Anduril, aside from just integrating and partnering with those companies, which we've done, is to build a gigafactory. If you go back and look at, before Tesla had set up the gigafactory, GM was talking about its manufacturing of electric vehicles, and they had these crazy low estimates as to how many they would be able to produce.
And Elon came in and he said, “We're going to make 500,000,” I think was the first number he threw out. Everyone said, “Oh, you're crazy. That's never going to happen. It's not possible. The supply chain doesn't exist.” Elon said, “We're going to build a battery factory, and then we're going to build a car factory right next to it. We're going to streamline the supply chain. I'm going to add a bunch of autonomy from the ground up. We don't have any legacy systems that we have to deal with. We're just going to build it from the beginning to be flexible, to be modular, to be software-driven rather than legacy systems driven.” And he did it. He actually beat all the numbers.
I think we need to do the same thing in defense. We just announced this program three weeks ago at Anduril that we're calling Arsenal. We’re in site selection right now. We're going to build a 5 million square foot manufacturing campus somewhere in the United States.
The idea is that, if we really need Stinger missiles, we can produce those at scale. If we really need next-generation Roadrunner interceptors, we can pivot the entire line to build those. You need to build fighter planes, you should be able to turn on a dime to build fighter planes. It will be a very expensive and longer-term project, but somebody has to do this. Otherwise, we're in big trouble.
Michael, I want to ask you a related question here. On the “protect and promote” framework that you flagged, you've been generally an advocate for a light-touch, pro-industry regulatory approach.
But you and others in the previous administration were very worried about Chinese supply chain infiltration, in basically everything: drones, cranes, chips, etc.
How do you make sure that you're not sticking those parts into American manufacturing without creating a massive burden for American companies?
Kratsios: Uh, yeah, it's tough. The first way you start is you have to identify which are the most critical technologies that you want to try to protect, and ensure that they don't have vulnerabilities. When I was at the Pentagon, we launched this program around US-manufactured drones to try to find a way to remove DJI [a Chinese drone company] from the DoD and you had these blue-certified drones that have components that are cleared.
In the Biden administration, you're seeing some activity around cranes. For these large ship-to-shore cranes that exist at our largest port facilities, the majority of them in the world are created by this Chinese company called ZPMC. A big chunk of them are actually remote-controlled and steered from offices in China, actually. To think that some technology like that doesn't carry some risk is funny enough in itself. So that's one that, that actually is being taken on.
I think one thing we need to think very carefully about is understanding that and I think it comes to where you draw the line on what are the most critical or or most sensitive technologies for the US to function.
Did you encounter challenges with prioritization? That’s been a common criticism of recent industrial policies: that if you prioritize everything, you're not prioritizing anything.
Kratsios: Yeah, it's, uh, it's hard. And I think where you draw the line again is challenging. There are some obvious priorities that started in the Trump administration and continued in the Biden administration. The most obvious is high-end, state-of-the-art semiconductors.
If you have the fundamental belief that the success of China’s development of AI that could be used for military purposes is contingent on their ability to access these high-end chips, then it makes a lot of sense to try to restrict their export to China. Being able to do that in a timely fashion with a technology that is constantly being updated is very hard.
Within Commerce, an agency called the Bureau of Industry and Security (BIS) is responsible for putting out these export controls. Tasking them with being on the ball and keeping up with the latest can be a little challenging.
It tends to be a leaky process. You could put an export control on some of these chips, but actually being able to enforce them and make sure that the chips aren't making their way into China via other means is also very hard. There are other back doors.
I just saw this FT piece a little while ago that was discussing iFlytek, a Chinese AI company that specializes in voice recognition and has been used for a lot of Uighur-related issues. They essentially found a way to access A100s through a U.S. cloud provider and were able to run their training on that. Even though they weren't able to buy the A100s themselves, they were able to access these clusters to train their models. So this is the stuff that you have to consistently play Whack-a-Mole against.
Trae, any follow-ups from the manufacturing side?
Stephens: I think the key piece to this is talent. Russia and China have civil-military fusion: essentially mandatory conscription of the most talented people in their society towards the biggest problems that they face. You could say the same thing with Israel: obviously, mandatory conscription there makes a huge difference culturally for the country.
In the United States, we don't have this, and I don't think that's bad. I like democracy and I like that we have freedom. But our elite class has basically decided that they don't want to work on any of the problems that matter. They want to make monkey JPEGs. And that's really annoying.
Whereas people like Michael left the tech industry to go work in civil service and take a huge pay cut to show up to the office every day and deal with a bunch of knuckle-draggers and actually move the needle for the country.
It’s something of a cop-out to say, “You have to focus on a limited set of things.” I think the answer is “No, we should have a country that actually works, and we should have talented people working in government that have accountability for making sure that our government works.” And, uh, right now it doesn't. [Gestures at audience] That's actually largely your fault, because you don't work in the government. Maybe you should.
Let me put this question to both of you. In the event of a second Trump administration, do you expect to see more tech talent headed to DC than the last time around?
Stephens: I think it's two parts. The government has to ask them to come and help, and then they have to be willing to come and help. In the Trump administration, I think the latter was a bigger problem. There was less of a willingness to go into it.
The question is, will there be a match between those two sides of the equation? The Biden administration showed us a problem in the other direction. There’s one member of the cabinet that has private sector experience, Gina Raimondo. That's it, one person.
I think we need to see in this next administration a willingness to say, “Hey, if we want Elon Musk to come and run a committee, we're gonna find a way to get him to do it.” I think that's a huge sign in the right direction. But I would hope that there will be people who aren’t necessarily politically minded who are willing to step into these roles where they can really make a difference.
Kratsios: Yeah. I think I'm optimistic that in the second go-around, there'll be a lot more willingness to come help. The way the country has changed over the last few years has made it clear to people that the government isn't solving the problems that it needs to. So I think on that side, I think there will be a demand.
Also, the pathways to be able to go into these important roles are much better understood by the folks that will be around this next go-round than they were previously. It's critical to get the cabinet right, and it's critical to get the deputy secretary and the undersecretary right. But at the end of the day, the vast majority of tech policy work is done at levels below that. It’s important to understand how to get very talented people slotted into those positions where they can help drive these policies forward, in an environment where tech and science broadly is always going to be a Tier 2 issue, and you have to be politically savvy in order to get it to a Tier 1.
Michael, each administration structures its White House differently. The managerial structure looks different from administration to administration.
What are your lessons from how it was structured in your time? What helped manage that full-spectrum federal approach to science and tech?
Kratsios: I guess there are two broad categories of technology policy: what I call national technology policy, and federal technology policy.
In the federal category, how can the government use technology to better do the work that it needs to in order to serve citizens? That type of work is generally coordinated and led by the federal CIO within the OMB in the White House. Those are decisions that individual agencies need to make: should they move to the cloud, what software should they buy, and so on. That work is most often extraordinarily nonpartisan, and just needs smart technologists to help the government do a better job operating as a government. That’s critical.
Then there's the second half, national tech policy. In our admin, the north star was to ensure American leadership and emerging tech. If you decide that we need to be leaders in artificial intelligence, what policies are in place to do that? How do you coordinate research and development across the agencies so the dollars that they're spending align with the priorities of the president? What do you do on the regulatory front?
A lot of conversations earlier today about whether or not to export control large language models. Do you export control individual chips? What regulations do individual agencies need to have on AI use cases? And so on. That's the work that's done on the national tech front. In the most elementary sense, the White House serves as a convening and coordination function.
What's unique about tech policy is that it has equities across a number of agencies. So take commercial drone operations: lots of people want to have drone deliveries to their houses. You might think, “We just need the FAA to approve that.”
But in reality, there are so many other equities there. The Department of Energy doesn't want drones flying around any of its sites. The DHS is concerned about the use of drones and in particular events that happen in the U.S. or for other safety reasons. DoD obviously has equities. So you have to get all these people around the table and get them all to agree on a path forward. And that's typically the role of the White House to do.
Trae, I want to go back to something you said right at the beginning, that folks in government generally want to do the right thing. The National Defense Authorization Act (NDAA) passes every year with supermajorities. It’s a yearly opportunity to change how we procure defense tech. Why doesn't that opportunity get taken more aggressively to do the kinds of things that you'd like to see happen?
Stephens: There are always policy changes that make it into the NDAA that benefit the tech industry or at least that are intended to benefit the tech industry. It's definitely something that the committees are thinking about as they pull this stuff together. I could reference all sorts of stuff that's happened over the last seven years since we started Anduril that were moments of celebration about things that happened in the NDAA.
It's less that it's not a frequently taken opportunity. I'm not sure policy is the thing that is really the driver. I think it's much more culture. You can have moderately intelligent people — I'm not gonna say super intelligent, but moderately intelligent people — who are really well-intentioned, who are patriots, who love the country, but don't have the level of expertise to make some of these complicated decisions.
That's not a critique of them. Generally speaking, I respect government employees and think that they're doing a great service to the country. But the world is way more complicated today than it was in the 18th century. You could make the argument that the founding fathers had read literally every important piece of literature that had ever been written in human history. That was actually sort of believable at that time.
In 2024, two professors in the same department at a university probably haven't even read each other's research. It's just infinitely more complicated. We're asking people with no tech background to make really difficult decisions about core platforms that are going to change the face of the world over 10 years. They're not able to do that effectively. This is both a fault of the government for letting the culture and the aspirational nature of civil service deteriorate over time, and also the responsibility of the tech industry to step up and be willing to fill some of these roles.
My folk understanding of military lessons from Ukraine is that we need a ton of cheap drones, way more cheap drones. Flesh that view out for us a little more.
Stephens: Well, one of the things that a lot of people in the media pointed out very quickly in the war in Ukraine is that Ukrainians were using a bunch of Chinese equipment. And it's true, they were using a bunch of Chinese equipment.
The reason they were doing that is that we were offering them expensive, exquisite systems that, if they were defeated using electronic warfare methods like GPS denial or something like that, you're crashing a multi-million dollar vehicle. And they're just not going to be willing to do that at scale.
And so what did they do instead? They bought 1,000 Chinese drones and sent them up and they were attritable. It's like if it's lost, who cares? I've got another 1,000 of them in a shipping container and they cost me nothing. It was never about China having superior military technology, necessarily.
It was really that we just weren't playing the same cost arbitrage. This applies to really high-end systems that are critical for conflict with great powers, like Patriot missiles. The United States pays a little over $2.25 million every time we fire a Patriot missile. Our partner nations pay about $4 million every time they fire a Patriot missile, and they usually shoot two, because you can’t afford to miss if you're actually trying to take down a target. The Iranian Shahed drone costs $180,000. Every time we shoot $8 million of Patriot missiles at a $180,000 drone, we are losing, whether or not we destroy that asset.
This is what Iran realized when they launched all of those drones against Israel. It wasn’t that they expected all of them to get through and cause kinetic damage to Israeli citizens. It was, “We're going to shoot a pile of heaping garbage that cost us $20 million, and you're going to counter it with a billion-dollar effort.”
You can only do that so many times before the economic impact becomes really damning. The approach that we're taking at Anduril is, how do we get low-cost, attritable systems onto the battlefield as quickly as possible, in a way that is scalable and manufacturable. That is the future of warfare. This is what our adversaries are doing. This is where they're investing their dollars, and it's where we need to be investing our dollars as well.
Maybe this is a dumb question, but we had 20 years of counterinsurgency wars: it's not like we just came out of a great power conflict and we're building battleships for that great power conflict. Why are we just learning lessons about attritable systems now from Ukraine and Middle East conflicts? That's so striking. Why wasn't this a lesson we learned 15 years ago?
Stephens: Well, as Kratsios has mentioned, this is really all about deterrence. Obviously we want to be able to win conflicts quickly if we find ourselves in a conflict. But more importantly, we would love to have enough of an advantage that conflict is unthinkable, because the adversary knew going in that they were going to get their clock cleaned.
This worked for a long time during the Cold War because it was believable. There was this credible threat of nuclear weapons. No one believes this anymore. I don't think there's a single person, in, I don't know, Iran that's thinking, “Well, if we launch another salvo at the United States, they're going to nuke Tehran.” We're not going to nuke Tehran. Like, that's crazy.
So you have to develop new methods of deterrence to conduct this next generation of warfare. The shifting sands of technology are leading us down into new paradigms, and we're still in many ways stuck in this old paradigm: We're gonna spend $15 billion on an aircraft carrier, $180 million on a fighter plane. That paradigm has passed us. It just doesn't really matter anymore. But that's where the majority of the dollars are still flowing.
One final question for each of you. Michael, you've pushed, successfully in some ways, for deeper science and tech ties with Europeans based on shared values.
But the Europeans hate innovation. So how do you do that?
Kratsios: The U.S. should just keep showing the way, and continue to be a leader in tech and hope for the best with them. The reality is, they are our allies and our friends, and we can just hope and pray.
Stephens: The Europeans don't seem to understand. Lots of own goals.
Trae, last question. This is totally for my personal interest. Why do people in defense always refer to “the American warfighter” and not just say “warrior”? The word already exists.
Stephens: During the world wars, everyone was a warrior: everyone was infantry, with a gun. They're out on the battlefield, they're fighting the war. People forget that the Air Force didn't exist: it was the Army Air Corps. The CIA didn't exist: it was part of the core defense apparatus. And so everyone was referred to as a warrior because they were all going into battle.
That's not modern warfare at all. There's intelligence, maintenance, people doing supply, logistics, etc. But not everybody is a grunt.
If you liked this interview, check out:
Share this post