Chris Power: Building the Automated Foxconn of Record

Before we get into Hadrian as a company and what you've built, I think, pretty uniquely, you guys are shaped by the last 50 to 100 years of American industrialism. So let's talk about this history lesson. We've obviously got a dominant manufacturing power in the US in World War II. You've got this golden age from the '40s to the '60s. We're building prototype fighter jets in 140 days. We've got intercontinental ballistic missiles. I think at one point the DoD in the '60s was 40% of global R&D, a real resume. So the first question to start with is what happened? What missteps did we take to get to a place where our industrial base is so hollowed out?

I think what's also important in this is what did the US invent? People forget that we invented semiconductor chips. The Air Force invented CNC machines. We invented forging, die casting, sand casting. It's not just that we were the best and the economy was working. That core capability, we invented.

What happened in reality is that we traded seriousness at a cultural level for spreadsheet returns in finance land. When you look back at people who were sounding the alarm on this, Andy Grove from Intel, actually decades ago, basically said outsourcing chip manufacturing from his purview was the greatest mistake that we made.

What I think happened is basically after Watergate, right around that time, we made three or four large moves and threw the baby out with the bathwater. One of those was moving away from the gold standard. It all roughly happened in the same five-year period. Removing the gold standard, we removed a lot of draft conditions, we had a lot of change in policy around sexual health rights and a bunch of other stuff, and NAFTA.

Culturally, I think it meant we became a less serious country, in the sense that we broke four of the rules. Rule number one, the dollar is attached to a real thing, so it is a serious currency. Rule number two is men go to war. Rule number three is some of the original social contracts got broken. I'm not passing judgment on whether those were right or wrong, but the social contract got broken from the history of the '50s and '40s. And then NAFTA broke the rule of reserve currency being a proxy for military power, which is based on industrial economic power. So by doing these massive free trade agreements, we really broke all of the rules that made us successful in the first place.

In hindsight, what I think people will realize is that it's like, you're a really fit guy or gal, and because you're super fit, you decide to stop working out. At that human level, it's very obvious for people to understand: you stop working out, you're eventually going to get fat and sloppy and decrepit, and you can't do 100 pull-ups anymore. At a civilization or country level, it's much harder to understand that. But the reason I use that analogy is that's what we did. We removed the core cultural and economic muscle from the country that made us successful in the first place, and then expected nothing to change.

The problem is that at a country level, it takes decades for that to correct. One decision in real estate might take five years to show up in policy and economics. These sorts of decisions take decades to appear. What I believe that enabled is that people who were born in the '60s, '70s and '80s grew up in this mindset that you don't need to do any of those four things, because we were already the winners. Therefore, everything that gets built is a poor abstraction. It's like a children's crayon drawing of a Monet. And then over the decades, it just gets worse and worse and worse.

For manufacturing or industrial power, it's really simple. We won World War II because we had the best domestic production of anything. Once you remove the gold standard, the rules of economics get broken and thrown out. That plus free trade and '80s and '90s globalization led to basically outsourcing the core hard industrial power that gave you the reserve currency of the world and allowed you to financialize it using weird treasury instruments in the first place. You can't do that unless you're the reserve currency of the world, and those two things combined are really dangerous, because what you get is free spending, riding off a multi-decade coattail of hard industrial power. Then you outsource the hard industrial power and you're left with nothing that backs the currency in the first place, which ultimately is: are you an importer or an exporter?

Then over the decades, it's, well, manufacturing jobs are bad. Well, it's because we outsourced it, and you lose that hard muscle. There's this great podcast with the new CEO of Ford, and he's explaining how they're trying to get into self-driving cars. One of the problems with Ford getting into self-driving cars is that they have never written software in the last 20 or 30 years, something like that. The car equivalent of avionics controllers, their subcontractor wrote the software. The drivetrain software was someone else. The company doesn't know how to build software because they haven't been exercising that muscle for the past 20 or 30 years, and now they're really struggling to catch up with Tesla on the software side.

So we had this multi-generational, 50-year period. We thought it was a good idea because CapEx bad, spreadsheets go up. People in the '80s and '90s are trading abstract futures on whatever. What we did was effectively, we just shifted the problem right. We got a bunch of returns in the '80s and '90s and the Bush era and shifted the problem right 60 years. And now we're paying for the sin of trading robustness for spreadsheet returns, for the financialization of the country. Which, for the record, every great power makes the exact same mistake almost every single time.

So then we can talk about the second- and third-order impacts of that, but that's the basis of what happened. Post-summer of love, we broke a bunch of rules, and then for the preceding four decades, those rules came to bear in a bunch of ways that the government or people at the time didn't predict, nor probably cared about. Now we're here facing down this situation where all of the leverage is in the CCP's hands, in the complete wrong direction. Now it really matters, but you haven't been to the gym in 40 years. So it's not like you can just decide to spend money and be a bodybuilder again. You're just catching up. And that's a really hard thing for a country with 350 million people to do.

You paint this perfect picture of why we find ourselves in this situation, which is the vast majority of manufacturing is happening in these mom-and-pop shops. They're very small, they're run by people in their sixties and seventies, and they're soon going to retire. We built something and it just aged with us forever with very little maintenance, and now we've got a $60 billion pot of spend that is supported by a really fractured, lower-quality pool of labor and capacity. Then on top of all of that backdrop enters Chris Power. You started in accounting. You worked in an e-commerce business in Australia, pre-Amazon, spinning up your own warehouses. Then you worked at a software company building workforce scheduling software. I've heard you say your background is really important to being able to go into these real-world companies that are very operationally complex and figure out where you should and should not build and use technology. And I know you get to this point of Googling "what is a private equity fund?" Should I roll up some of these machine shops and slap on some software? Eventually you get to this idea that, no, it has to be a full-stack effort. How do you go from these insights and this experience to "I need to build the new age factory of the future"?

That's a very good question. I think the vision now, by the way, is far greater than machining, because what we're finding is that the problem is all over the place, and the only way to catch up is basically with AI, software and robotics. So I think it was an evolution.

I had this macro thesis, and I'll have to dig it up sometime, because I wrote a 100-checkpoint spreadsheet. I got a lot of things wrong, but what I wanted to see on the ground floor in the US was evidence that my macro thesis was correct, and a thesis on what the right starting position to do this thing was.

A general rule, I think, for building systems is that it is possible to make an exquisite system cheap. It is completely impossible to make a cheap thing exquisite over time, in the sense that many people fail in manufacturing or hard tech startups by first building the ultra-commodity version of what the market is. It's very hard to change a system or software to do more complicated things. It's possible to take a complex system and over time simplify it, or provide a cheaper version of the thing. You see this all the time with software. It's almost impossible for someone building small business payroll to start selling enterprise contracts. It's just a completely different thing.

So my first starting position was there are many markets in manufacturing, and you should probably start with the highest-margin, top tier of that space. Companies have tried and failed by selling onesie-twosies to engineers that are not really high-tolerance parts. They're not that complicated, because it's easy to automate. But that is a market trap, because you never muscle your way up and build what you actually need to build to attack the high-margin part of the market. You want to do the whole market eventually, but you need VC funding, which means you need to attack the higher-margin part of the equation first.

Through that, I very quickly learned that that part of the market in manufacturing is basically only because of ITAR regulation. Everything else has largely been outsourced. So if it's not defense, space, medical devices, some industry where there is a structural regulatory or safety reason why it has to be made onshore, you can build a system that is slightly faster or something like that. And through that, basically, the standards of how good you have to be to work in the top tier of space and aerospace and defense as a contract manufacturer, which we are, is if you started a small manufacturing company and then worked your way up, the average time is seven or eight years to get the certifications, level of quality and level of approvals from the big primes to even begin working with them.

My conclusion, for right or wrong, was that you can shorten that a little bit, maybe, but the only way to really start that journey is by buying someone that's embedded in that regulatory system and then putting the technology into them. Otherwise you just run out of time. You're not going to get your first customer within the first two years, no one's going to fund you, game's over. And by the way, I'm still kind of a nobody, but back then I was really a nobody. Five years ago, no one was funding factories. People barely fund factories now, for the record. Thankfully the interest rate environment was such that capital was free and there was too much money sloshing around in the M2 of the old people. But that was where the private equity thing came from, because you have to buy your way into this regulatory barrier.

Now, 12 or 14 months into that journey, what I figured out was that if you're building autonomous factories at all, you have to simultaneously standardize the hardware while you're standardizing the software. Every time you fragment the hardware, you get a nonlinear complexity increase in the software. Most existing contract manufacturers, if they have 30 machines, whether we're talking about 3D printing, forging, casting, wire harness or CNC, the pattern is the same. They're so fragmented on the hardware side that to get any sort of software leverage, you basically have to replace all of the CapEx anyway. But I didn't know that.

So 16 months into that, there was this really awkward moment where I realized that, one, I'd come to really care about the mission and the industry in general, because once you see the problem, you can't turn away from it, no matter how hard it is. The second real thing was that you could buy a company, and unless you'd already pre-built the technology, it's a massively enormous journey and nearly impossible to upgrade a brownfield operation.

So I had this really awkward moment where I had to call everybody and be like, "Look, I think I've gotten that wrong. I'm really sorry. You put me in business." Some people were very angry at me. Some people understood. But it was really a mission-driven decision of, I came to this country 18 months ago and I have this thing which will make me rich for sure, without a doubt. But I realized that you're only ever going to get 20% more efficient, something like that, and it would just be impossible to fulfill the mission. The only way to do it was to take the bet on somehow screwing up the momentum and the funds to greenfield the whole thing, soup to nuts. And I just decided, well, I'm going to do this. I'm going to swing big, and I'd rather be poor and mission-oriented than rich, but knowing in the back of my head for a decade, "I know this is the wrong thing. It's not going to go fast enough." And three and a half years later, here we are.

It feels like one of those necessary formative experiences where, if a time traveler had come along and told you to skip this whole process and jump right to the full-stack factory, you wouldn't have the scar tissue to know exactly what you need to do. And it's a good segue to thinking about the product as this full-stack process you've built: as you've described it, a software, robotic, human, industrial complex, if you will. I know there are components like design for manufacturing, ingesting things into the process, CAM automation to get things ready to go, and CNC machines for the actual machining. And as I've gotten to tour your facility, there are these processes you've built to integrate all of this hardware and software. What's the framework you use to think about the product? If you were to go from A to B to C, what are the chunks, and how have you thought about tackling those?

Firstly, I think we're 50% of the way there, even after three years. The way I think about it is there are three layers to the stack. One is circus ringleader, central routing, which is scheduling and dropping buckets of tasks into the factory, to stations, whether it's a human doing something or a robot doing something, that give them a clear signal of what to do, how, when, why. But with enough operational flexibility that you're not over-scheduling things. There's loose command and control that draws tasks into queues, which is like this distributed multi-agent system. That process scaffolding is really important, because that's the number one thing that doesn't scale: what are the right sequence of events, so that someone, whether it's a robot or a person, doesn't have to think about the hill-climbing problem or the factory complexity. They just have to focus on, "I need to get this queue done today."

If you were to anchor that to a product, I think Flow is the ERP you guys talk about. Is that that initial process and engine?

That is one part of it. The second part is what I describe as an L2 layer, which is left and right in the process. So I make a part, you inspect it. What data pack or outcome am I handing over to you that your software can validate the input and output of? Then you have an incoming data packet or physical packet that you can go execute your workflow with, that there are software controls around. Because if I give you a bad data packet, you're immediately screwed.

So there's the top-down command and control routing. Then there's this L2 layer, which is left-right controls. And then under each of those L2 layers, which could be CAM programming, machining, quality, buying raw material, supply chain, each one of those things has task or workflow automation. Okay, you've got that scaffolding, but this takes 20 hours. How do we make that an hour? Or this takes 20 minutes. How do we make that a minute, or fully automated?

We currently have 60 engineers, give or take, but that's only been in the last year. In my opinion, this is a three-year, 200-engineer sort of budget, which we still don't have. So we try and make the trade-offs in, one, what's the best engineering return on investment? If you can spend an hour automating a 10-hour task, obviously do it. If there's a five-minute task that takes 100 hours to automate, don't do that.

And then for the overall error rates of the factory, what are we building first to control the error rates and the overall flow? The analogy I try and tell people is: what can we leave as a spreadsheet the longest? That's not necessarily cost. That can be errors or complexity, or something that is a cheap task. It might take an expert 15 minutes to do, but it's very hard to train anybody to do that. So we are constantly making these trade-offs in our roadmapping and engineering exercises. We know what the full system looks like, but how are we going to get there, minimizing the impact to customers while meeting our own internal goals of cost and how much money we can burn on VCs? It's a very hard trade-off to make, and it's a constant cycle.

One of the interesting components of your product is that it has a really important human component. I've heard you talk about how it would take a decade to automate 100% of every single process, so instead you focus on the 70% to 80% you can automate and the 20% where it makes sense to keep people involved. It feels like the first bucket of where people play in is knowledge extraction. You have these experts and machinists, and you have to get that knowledge out of their heads and into the software. How do you do that?

We start with software engineers and experts. In the early days, this was me doing very, very hyper-detailed product management to get a zero to one out. What we build first is not automation. It's simplification. The goal of the first product is not to take something from 60 hours down to five minutes. It's to simplify it so much that you could train anybody to do that in 30 days. What we've found over time is that starting with an expert and trying to automate their job never gets you to a product that a non-expert can use.

So the starting position with everything is five to six of the best experts in the world and a bunch of very context-heavy software engineers, and the goal is that none of the experts are doing thinking anymore. The starting point might take the same amount of time, but the workflow is consistent. They're using our software, not their own tribal knowledge. From there, it's actually very easy to take those chunks and automate them away.

What we find, basically, is if it's a 10-hour task, I'm using an example, we can get to 70% to 80% of that reasonably quickly: a small handful of engineers, six months. From there, we snap out the workflows between what is very cheap for an expert to make a decision on, that can make the rest of the automation very, very easy. A very dumb example of this is, how are you going to fix your part? It's very, very computationally hard, but an expert can eyeball something and click a dropdown in two minutes. That makes, one, the engineering 10 times less hard. And secondly, if an expert makes that one decision, a non-expert with our software can do the rest of it 10 times more efficiently, with our software checking and balancing. So that's the first principle of everything we do.

The second principle is we have to operate and we have to move very fast, and the edge cases of manufacturing are very, very wide. After that 80%, you would need two orders of magnitude of the resources, experts and software engineers, to burn down that last 20%, because the branches just get very, very distributed. So what we do is we build software that's always human in the loop. Our automation almost always generates an output and asks the user to validate it, and anytime a user overrides that, we use that to train the automation, to close that last 20% of enormous spread-out complexity.

Everything is human in the loop for two reasons. One is you just can't get to 100% automation. We have to sit here with 200 engineers and take six years. The second reason is the long tails of manufacturing decisions are so long that the only way to do it is basically to set up the system where it's 70% automated, have humans checking and balancing it, and feed that to models, which improve the automation over time as the factory grows, across the sub-processes as well as the master controller. That's how we think about those trade-offs.

Once you accomplish that, you get into new labor entrants. You've talked about how in almost every technical category there is a shortage of skilled labor, whether it's machinists or welding. And with that layer of automation, almost like a copilot, you can take folks who were working in hospitality, as UPS drivers, or at Home Depot, and within 30 days or so they can learn these processes. For a lot of folks, I'm sure that feels too good to be true. It's this idea of "once we automate truck driving, we'll just teach everybody to code." How do you ensure that those people really are capable and well trained? Have you had folks fail out? How do you think about that new labor force development process?

I would look at it as what we're doing is massively, massively, but not completely, de-skilling the thing, so you can learn it in 20 days, not in a decade. The other trick, frankly, is we have an entirely well-staffed workforce and training team whose entire job is to take people, mold them, shape them culturally, discipline-wise, and then how to use our systems. But yeah, we're at a point where, 100% of our factory roles, no more is there an expert on the factory floor, apart from operations leadership or when something goes wrong and an expert needs to intervene. But that took a while to get there.

I think it's really important, one, because I just don't believe you can have a fully automated factory, full stop. It's never going to happen. So this was just a necessity. And then secondly, it's really interesting, because what we see after about a year of someone being here is they want to go into leadership, or they want to go into software engineering, or they want to go into a more technical manufacturing role, up that skill tree.

Part of that is, firstly, you have to give the workforce a 10x advantage over China with software and robotics. Then you need a new workforce that is technology first. They can deal with our software engineers making changes to their workflow every second day, which is no joke. Our operations team is incredible, because half the tools don't exist and half of them are changing every single day. It's a huge part of the system. And it also allows us, as intelligently as we can, to let humans do the things they are good at and let robots do the things robots are good at.

I think that framework is really refreshing. There are a ton of benefits to avoiding full automation: avoiding complicated science fair projects that don't add a lot of value and are just really difficult to do when a human could do them in minutes, and creating jobs in manufacturing instead of destroying them. But then I see announcements like this new 800,000-square-foot factory in China that they claim is 100% autonomous and can run 24/7 as a dark factory, making 10 million phones per year. The first question is, do you even believe that? It wouldn't be the first time that an operation in China has lied about its capability. And the second is, does the commodity of a phone make it easier? How do you think about the world of full automation that people are trying to create?

I think that is probably more true now than it was a decade ago in China. Certainly the first iteration of Foxconn was: there's the automated factory you take the white people to, and then there's the factory out the back that no one visits.

100%. Total classic China. Look at these well-built concrete buildings.

Totally. Look over here. Don't look over there. China is like the WWE of economics.

I think there's a huge difference between the complexity of what we're doing, which is make anything a customer throws at you, versus setting up an autonomous factory for a product. The latter is 100 times easier than what we're doing. We're working on some projects now with the DoD and a couple of customers where we are building automated production lines for a specific product. And that, I think, yeah, you can get to full lights-out 24/7. You probably don't want to, but you can. I think we're probably the only company in the US that could do it, which is why people have come to us for it.

I kind of believe it for phones or large-rate stuff, like an iPad. I think if someone gave us the contract to make an iPad in the US, we could do it in nine months. It's just far easier to set up automated production lines and inspection lines in that full system for 10 SKUs at massive volumes. I really do think that's possible. I think we can do it here in America. I think they're probably doing it in China now because their labor rates are getting pretty expensive, because the middle class is going like this. And compared to what we are doing, which is variable manufacturing, it's not that hard.

A couple of months ago, I saw an interview where you talked about increasing throughput rates by building your own automated machining cells. Is that the idea of finding these pockets where you can automate a huge portion of a particular SKU?

We did that because we could halve the CapEx cost of standing up our own robotics and going far, far more full stack than we initially thought. In terms of taking a product like an iPhone or a missile or a rocket engine and standing up a facility just dedicated to that one thing, you just don't have to deal with the variability. You can just discretely automate things, versus building a system that can handle whatever the day throws at you. That's more what it's about.

One of the things I've heard you talk a lot about is having engineers as close to the problem as possible, trying to cut out the middleman of product management that makes them go off in a different direction than the operations actually need. It reminded me of one of the famous principles from Kelly Johnson, the Lockheed Skunk Works founder, who said engineers shall always work within a stone's throw of the airplane being built. As you start to scale, do you think that approach scales with you when you have 10 or 15 factories? Or is that a build philosophy that changes?

I don't think the philosophy is going to change. How we orchestrate it is going to change dramatically. The important thing to maintain over time is engineering context to the real user problems, and removing all layers of communication. I don't know if you remember this, but Tony Xu from DoorDash, half of his engineers rioted, not during COVID, but during full ZIRP, free breakfast land, where they were like, "You guys have to go do a DoorDash delivery once a month." The reason they did that was because then there's not a product team that's writing Jira tickets. Your engineers are just DoorDash drivers. They're going to be like, "This is terrible. This sucks. I'm just going to build something." That's what you want to maintain for as long as humanly possible.

But I think the way we orchestrate that and provide that is going to be enormously different across multiple factories. You very quickly get into this world where, if engineering, product and design is only at one factory, the other factories always feel like the redheaded stepchild: "Oh, no one's building anything that I want." You have to load balance it a little bit, but it's going to get enormously difficult at scale to maintain that. I don't have any great ideas to do that right now, but it's a next-year problem.

I've heard you talk about things like a human-to-machine ratio of one person to four or six machines simultaneously, and improving from a standard of eight to 18 weeks for product delivery to ideally one to two weeks. Is that where you guys are at consistently? That one-to-six human-to-machine ratio, is that consistent, or is that with established processes?

Our person-to-robot or machine ratio is very consistent now. What's more impressive is that our productive uptime of our capital equipment is as good as automotive, where you're making 10,000 of one thing, in aerospace and defense, which is very high SKU count, very low quantity count. That's the real magic trick. We're effectively operating at automotive facility levels of efficiency in this highly variable, stochastic environment. The real thing for production for customers has become quality, delivery, cost, and we're now pretty close to so cost-effective that that's become product-market fit.

What does that translate into from a numbers perspective? In automotive, is that 80% or 90% machine uptime, whereas standard machining is 30%?

I think industry is 20% or 30% on a good day. We've seen 12% at customers in their internal operations. In automotive, 80% uptime, yielded OEE, is gold standard. We run pretty consistently at mid-to-high 70s.

That's orders of magnitude better than what you're typically seeing. And it feels like that's translated into business performance. I've heard you say that when you started operations, maybe early '23, in the first year of operation you'd scaled to over $20 million of revenue. You're growing like a software company, which is basically unheard of in what you guys are doing. Is the efficiency what has allowed that pace of scale?

Actually, no. I think it's standardized processes. The hard thing about contract manufacturing is you don't know what you're going to bid on, and every thing that you sell to customers is different. What helps software companies scale, other than product-market fit, is that if the demand is there or you can create the demand, creating the n-plus-one deliverable to the customer is effectively spinning up a new AWS instance and doing some customer success. In what we do, every single new sale is different. A different part is a different process. It's a different tolerance band. It's a different inspection process.

So I think one of the most important things we've done is basically take that complexity and try and simplify it as much as possible, so operations is as consistent as possible every single time, to remove the variability of the inputs and outputs. We are far from where I wanted to be, but that has been one of the tricks of being able to scale: process standardization, and basically using software to remove the complexity. Why is McDonald's efficient? It's because you're making the same thing over and over again with a small amount of variability. So imagine if you had a custom burger franchise, but you found a magic trick where, for the person on the grill, it was the same thing over and over again. That's where we're trying to get to.

I've also heard Matt Steckman, the CRO of Anduril, talk about these systems where you have contracts and you don't know what you're going to be bidding on, or what volumes and pace it's going to happen at. It's not at all linear like a software company, but if you can pile them on to each other and very efficiently figure out your capacity, you can actually start to plan. "We are going to scale at a certain pace" is more knowable than you would expect for a company like this.

I think this is a significantly easier business to run at $100 million in revenue than it is where we're at today. Adding a million in revenue now is a lift. Adding a million in revenue in a year's time is going to be an order of magnitude easier.

What you've laid out is this massive discovery engine you've gone through to build the product, and then these standardized processes that make you run much more efficiently. So from a competition perspective, it's kind of difficult to even come up with a framework. On the one hand, if you're taking customers away from these mom-and-pop shops, you are competing with them, and you're going to do it dramatically better, which is the goal. But are there folks that pose any kind of legitimate threat to you?

I think the biggest risk is that we might be early, or we screw it up. I would not discount how sticky these businesses are, and I would not discount how painful it is to move a dollar of revenue away from someone that has been working with you for 30 years.

Totally. Very Lindy.

It's very difficult for them, even if they're doing a terrible job, no matter how terrible it is.

As we think about the types of customers that resonate with what you're doing, you've got startups, these newer primes like Anduril, and the larger, more established primes. The newer players are moving faster, they have leaner requirements, and they really like the things you're doing. On the other hand, that has often forced them to bring some of their manufacturing in-house. I think I've read that SpaceX brought something like 80% of their manufacturing in-house at one point because they were unsatisfied. Just a couple of days ago, Anduril announced their new 5 million-square-foot facility, Arsenal One. What's interesting is when I hear them talk about 90% of Anduril's products being buildable using commercially available components. It feels like for the SpaceXes and the Andurils of the world, even as they build up their own capacity, Hadrian becomes an input into that rather than a competitor to that. How do you think about that ecosystem as more of these guys get into the manufacturing game?

I think it's evolving rapidly. Firstly, what Anduril's building is like Tesla's gigafactory for their scale production products, the vast majority of which is going to be assembly lines for whatever products. And I won't say too much more, otherwise Steck and everyone else will kill me. Obviously we have a very long-standing partnership with Anduril. It's been a great team to work with.

What's interesting is what we're seeing out of the DoD and other startups is, "Hey, we want to buy your drone, but you don't have a billion dollars on the balance sheet, and we don't necessarily believe that if we gave you a contract to buy a thousand, you could actually produce." So we see this legacy segment where products have already been in production for 20 years. How are we going to go disrupt that? It's usually cost or a desire to massively simplify your supply chain. No one wants 400 vendors. They want one. Then there's the startups, which is, "God, I really want someone to help me get to the first prototype so I can raise a Series A." We actually just acquired a company to launch a product specifically for that. And then there's how you do all these enterprise partnerships with companies at different levels of manufacturing maturity, and at the top of the scale is your SpaceXes and your Andurils.

But what's emerging really interestingly over the last six months, because people know we exist as the contract manufacturer, the automated Foxconn of record, is people now go, "Well, why don't I just do sales and design, and partner with Hadrian to go from prototyping to production?" Have Hadrian there doing design for manufacturing, simplifying the product, and be the autonomous Foxconn of all these new DoD-funded companies, because very few of them have the capital or scale or desire to be a mass producer. And now their customers, mostly the DoD, are saying, "Well, we want to award you this contract because we'd love 10,000 of these, but we don't really know." So we're stepping into a lot of those situations saying, "We can do it. Just plug us in. We've got it covered." That has been a really interesting evolution of the market, and I think it will happen over the next decade as the market chops and changes.

That makes a ton of sense for folks who are scaling very quickly and trying to put things into production. And the DoD is a better customer today than it has been for a really long time. But I heard you say at one point that when you started working with customers at the beginning of '23, you thought they would take until the end of this year before they came around. It feels like around this time last year, there was an acceleration of even some of the bigger established primes seeking you out. What do you feel motivated that eagerness, even from some of these slower-moving players?

I'm not sure. I think, frankly, a lot of them are patriots. They're like, "Yeah, this is going to change. This is terrible." And no one's really said, "Hey, I've got an idea of how to change this. Let's partner up." It's one of those things that's obvious when someone tells it to you, but if there isn't a solution, what are you going to do? We've seen massive acceleration on the production contract side. Usually that takes multiple years, and we're lucky enough that it all happened last year, which is great.

At Contrary Research, we talk a lot about key risks and key opportunities: the pre-mortem, pre-parade type of stuff. On future opportunities, you mentioned this goes beyond machining. There's working in alternative materials, where it's not just aluminum but moving into steel and things like that. How do you think about that surface area expanding? Where does Hadrian go from here?

More metals, for sure. We're in three metals now. We added steel and titanium. We are so overconstrained for the next year that we're probably not going to touch anything else, except for these big partnerships with large companies where we're custom building a factory, because that's just obvious. I think we'll wait and see. There's a lot going on in 3D printing and in autonomous manufacturing in general.

What I'm seeing, at least, is many startups building one-tenth of what we built: automation for a specific manufacturing process, which is great. I'm really thankful that there are 100 startups doing that. My guess is that not many of them in this capital environment will be able to come out of the gates with $200 million to build their first scale production facility, nor an ERP, nor a go-to-market team, nor the nine other puzzle pieces that have gotten us where we've gotten today. So where there are gaps next year, we will spin out R&D teams for more types of manufacturing, and where there are partnerships or acquisitions to be had, we will also do that. The goal being, what's the fastest way to be the one solution for our large customers? How we get there, I think, is totally up for grabs next year.

Are those specialized, almost white-labeled facilities, where you're working directly with one customer to build dedicated capacity, separate from and in addition to your core Factory 3 facility?

Yeah.

I want to talk about Factory 3. You had this 20,000-foot Factory 1, a proof of concept, then you shut that down and moved to Factory 2, I think 100,000 square feet. The way you've talked about it, if you keep your current pace, you'll fill Factory 2's capacity at the end of this year or beginning of next year. At one point you mentioned you were looking at six different states. Give us the update. When do you think you'll break ground, how big will it be, and if you've selected a state, when are you going to announce it?

We'll have to break ground Q1 or Q2 next year, and it'll come online late next year. In terms of states, we are down to eight or so that we like. Then we'll have to run a formal site selection process in Q4, which will have all sorts of incentives and where is the best labor pool for us to recruit from. We'll probably run something very public in Q4, or make a selection early next year, and then go live late 2025 is the pace that we're setting.

I don't need specific states, but the exercise you're going through is really interesting. If you extrapolate the dream of really rebuilding the industrial capacity in the US, and you 10x and 50x the capacity we have today, you have to imagine you start to see these industrial Silicon Valleys build up over the next five, 10 or 15 years. As you've gone through this exercise to qualify where the talent and the skill capacity is, where do you think some of these hubs are going to get built up?

I think it's going to be existing hubs expanding. I find it very hard to believe there's going to be a Shenzhen of the US unless the government steps in and massively subsidizes it. I think what you're going to see is re-industrialization and acceleration of places in the Midwest, and I think it's always going to be centered around customers. If and when we do something for shipbuilding, it's going to be on the East Coast. Munitions are all Arizona. Space is mostly California, and Texas has a huge manufacturing base. So does Utah. I think it's always going to be those traditional hubs where you can blend the old and the new. I find it very hard to believe, unless someone steps in with a $3 billion subsidy, that some new hub gets created, unless there's a market force. It's too big of a problem for anyone to incept.

You've described this as AWS for manufacturing, doing for manufacturing what Stripe and AWS did with plug-and-play infrastructure components that dramatically increase the speed of product iteration. When you've talked about Hadrian in the past, it's that over the next five years or so, if you can pull off this copy-and-paste capability, where maybe 30% of it breaks the first time and you fix it, you get to the point where you have 10 or 15 factories. What has to happen from where you are today to achieve that vision? What are the milestones?

I think there are two really important ones. One is staying laser focused on this current facility and not getting too far ahead of scaling before we're ready. The second one is, I think 50% of our value creation in the next 10 years is going to be financial engineering, because what you're talking about here is billions in CapEx that has to be privately financed somehow, where the IRRs of CapEx by itself don't work for venture returns. I hope I have a big announcement on this in the next two or three months, but we're pretty good at it, and we have to get extremely good at, non-dilutive sources of funding to buy and operate billions of dollars in CapEx. That, I think, is a huge focus for us next year. It's customers, workforce, automation. Financial engineering has always been the fourth pillar, but it's going to get pretty close to the top of the table of the important ones to tackle next year.

So you prove out a lot of these capabilities in a fairly expensive way, in terms of cost of capital, with equity. Then once there is a more demonstrated base, you scale that with financial engineering to figure out what instruments make sense to scale the engine.

Yeah. You can think about it like, why does Germany have such a good industrial base? Well, it's because it got completely destroyed and they rebuilt it with fresh CapEx. The majority of the CapEx in the US, 99% of it, is between 30 and 60 years old, because we had this massive boom cycle, and inflation was much less, and all these other factors. So the other thing you have to consider is how do you fund $2 billion to $10 billion in CapEx that doesn't meet venture returns? And the answer is really good financial engineering. So unless anyone from the government is listening and wants to build a re-industrialized America fund, which would be a really good idea, for the companies that are successful, I think one of the most important skill sets is going to be debt, credit and financially engineering their operations.

A lot of the risk is execution on your side, and there are broader macro components that turn everybody's heads. I've also heard you talk about things like cyber espionage and China stealing trade secrets. I remember reading something from US Cyber Command about a quarter of a trillion dollars each year being lost to cyber-related industrial espionage. Somebody called it the greatest transfer of wealth in history. Is that a key concern for you? What keeps you up at night outside of the execution risk?

Execution is number one. Cybersecurity is a big deal, which I won't go into any details on for obvious reasons, but we spent a lot of time on cybersecurity. There are many things that could still kill us. I think one of the answers is you've got to get big really, really fast, and the rest of it sorts itself out. We'll see. I think one of the other risks is, the DoD continues to spend money in new companies, the growth will be there. If they don't, for whatever reason, it's really hard, because things just don't move that fast. There are many risks over the long scale, but certainly number one is daily execution at this point.

I want to end on what I would describe as your existential motivations for building Hadrian. You've talked a lot about how you see manufacturing as the underpinning of freedom and safety, and how the next 20 to 30 years could really dictate the next thousand years of human progress. When you think about that threat and the reason for existing, if you were to think about what could improve our chances of the US coming out on top of that conflict, what are the biggest levers you pay attention to?

Energy. Cheap energy, too cheap to meter. We should be the number one development partner of Africa and Australia for mining, and not in a weird subjugation way. We should be throwing trillions of dollars. The third one is pharma, actually. People forget that in a war footing, we don't make antibiotics anymore, and in fact, we have no idea how to do it. So we should actually be heavily partnering with India, who's pretty decent at pharma and getting better every day, to do a national strategy for pharma manufacturing in India. And then the fourth one, I think, is we should have already given every TSMC employee $5 million and a green card and a bunch of SEALs to get them out of there.

Those are my four things. Energy too cheap to meter. Help Africa industrialize itself even more than it has and be a partner, not make them a vassal state. Help India develop pharma. And Operation Paperclip TSMC.