Hadrian CEO Chris Power on How to Dominate US Manufacturing: GPU-Style Factories, Building Capex Ahead of Contracts, Ulysses Contracts, and Why Honesty Compounds

Chris Power, founder and CEO of Hadrian, has opened six factories since 2022, the latest a 2.2 million square foot site. In this interview, “How to Dominate Manufacturing in the US”, he explains what it actually takes to scale American manufacturing for aerospace and defense: factories that run like GPU clusters, billion-dollar capex bets placed years before the contracts arrive, the discipline that keeps judgment intact while growing fast, and why honesty with the government and investors compounds faster than any sales tactic.

TLDW

Power argues that US reindustrialization is a chicken-and-egg problem: there is no contract manufacturing capacity in America, so nobody will give you a production contract until you have already built roughly 1.2 times the capacity you hope to sell. That makes every capex bet “company killing” until a business reaches about $50 billion in scale, and he puts Hadrian’s risk of death at a steady 80% because growing at venture rates means always betting two years ahead of revenue. His answer is flexibility: standardized, software-controlled stations that let a high-mix, low-volume factory run like a GPU cluster, with 80% of capex shared across programs so one investment hedges several demand signals. He covers treating capex as a one-way door and paying 30% more for hedges, overbuilding foundations and electrical drops, designing factories with “off-ramps” for failures, a separate fast lane for new product introduction, surviving a rusted batch of 40 machines, standardizing on shared robot arms, scaling from 725 to about 1,300 people with small engineering teams and only two sacred cultural values, choosing programs by mission and “doors that will never reopen”, Ulysses contracts against hype-driven judgment, being the second-best person at every function, financing capex with a credible cost-of-capital curve, learning that most customers do not pay for speed, building government trust by saying “we are not ready”, maximizing velocity instead of forecasting, managing energy, and sending the hard messages right away.

Thoughts

The core insight is uncomfortable and clarifying: in capex-heavy businesses, the “safe” choice of waiting for a contract before building is really a 100% chance of never growing. Power reframes the billion-dollar factory bet as moving from certain death to a better than even chance of revenue, and he is unusually candid that his risk of company death has not fallen below 80% in two and a half years. That is not recklessness. It follows from the math of a two-year capex lead time on a venture growth curve, and he says the risk only retires when growth slows to around 20% a year. Anyone evaluating hard-tech startups should borrow that framing. The question is not whether the bet is risky but whether the alternative is worse.

The GPU analogy is the most transferable idea in the interview. Most production lines run linearly, like a CPU. Hadrian standardizes every station and runs it through software so a high-mix, low-volume factory can load-balance like a GPU cluster, and so 80% of the capex can be shared across programs. Paying 20% more for capex that serves three demand signals (turbine castings and munitions, drones and interceptors) turns a single binary bet into a portfolio. The same logic explains the seemingly wasteful choices, like $25 million of heavy foundations instead of $15 million, or electrical drops everywhere even though 30% will go unused: they buy months of layout freedom against lead times you cannot compress.

The section on judgment in the middle of the conversation deserves more attention than the factory talk. Power points out that a CEO’s intuition about what the company can handle is calibrated on their own ability and their ten best people, while the tenth program is run by people they have never met, and the consequences show up months later. His fix is the Ulysses contract: rules written while clear-headed that you do not break when tired, manic or winning. The hedge fund line about real estate cycles repeating because new 25-year-old investors have never felt the pain of losing their shirt is a sharp explanation of why hype, not fatigue, is what really kills companies.

The finance section is where the business model becomes believable. Power treats engineering as a subsidy of gross margin, the same way software companies accept R&D spending before efficiency arrives, and says the CEO’s job is to predict the capital cost curve over four years accurately enough that credit funds will eventually lend against it. Hadrian was underwater on financing costs until late in its Series B and educated every investor on manufacturing cost accounting. Alongside that is his most useful admission: he was wrong about speed. Outside one large rocket company, most aerospace and defense customers do not pay for being faster, they pay for flexibility, scale and total program cost. It is a good reminder that a startup’s first customer can be the exception, not the template.

The last part of the interview offers some of the most practical operating advice. Power says the way to plan for unpredictable, runaway demand is not to forecast better but to find every place the company is below maximum velocity and fix that, since you can always slow down but you can never quickly speed up an organization. His argument that trust compounds (spending two and a half years telling the government Hadrian was not ready, turning down contracts that would not create value) is the long-game version of the same idea. The small habits land best: know which kinds of work drain you, and when a hard conversation comes up, send the email or draft it that night instead of scheduling it for Thursday and losing the whole week.

Key Takeaways

Scaling US manufacturing takes three things: GPU-style modular stations that give you standards to automate against, a complex organization capable of real manufacturing autonomy, and a balance of physical AI software and a new American workforce.
Traditional production lines run linearly, like a CPU. A GPU-style factory standardizes and automates stations so a high-mix, low-volume factory runs as efficiently as a low-mix, high-volume one.
Like AWS starting with Amazon as its first customer, a manufacturer needs capacity before customers will trust it. Shenzhen has spare capacity. The US has almost none.
To win big contracts you need to build roughly 1.2 times your future sales capacity ahead of signed demand. What manufacturing sells is trust.
Getting some defense contracts takes about two years from the start line, so you have to build capacity without knowing when the contract will come.
Capex lead times always slip. If a vendor says 8 weeks, plan for 8.5 to 10.5.
With flexible software and workforce, about 80% of capex can be switched between programs, like shared GPU clusters serving different AI products.
Paying about 20% more for shared, flexible capex can turn a $300 million bet against one demand signal into a bet against three, for example turbine castings and munitions.
Until a company reaches about $50 billion in scale, every capex bet is company-killing. But not making the bet means a 100% chance of death or a small acquisition.
Power estimated Hadrian’s chance of survival at around 20% two and a half years ago, and says the risk of death is still about 80% because growth keeps pushing capex two years ahead of revenue.
Risk only retires when growth slows to around 20% a year and cash flow catches up to capex.
Capex and long engineering projects are one-way doors. Spend three of four weeks making sure the large, single-use items are right, and hedge on flexible items like robot arms.
It is better to spend 30% more on hedges than to receive nine months of capex and discover it is wrong on day one, putting you 18 months behind.
In long-correction businesses you sometimes “roll a hard six”: there is no way to manage the risk, you just have to be right.
Hadrian overbuilds heavy vibration-isolated foundations and electrical drops everywhere, trading extra cost for months of layout simulation time and fast machine installs.
There is no such thing as asymmetric manufacturing, only asymmetric products. Manufacturing advantage comes from scale, flexibility and agility.
Machine vendors are rarely ready for software integration. One German vendor assumed Hadrian were hackers when asked for API documentation, and half the API calls were not implemented.
Plan every project assuming vendors and equipment will fail you, and build hedges into timelines.
Most factories are designed only for the success case. Hadrian builds in slack capacity as “off-ramps” so one failure does not back up the whole line.
New product introduction needs a physically separate fast lane, because the cost is in the back-and-forth between manufacturing and inspection, not the cycle times.
A batch of rusting subcomponents across 40 new machines nearly killed Hadrian before a fundraise. Power pushed the vendor to fly in 30 engineers and fix it in four weeks instead of by Thanksgiving.
Company-killing events happen roughly once a month or quarter in manufacturing. That is the game.
Hadrian raised a Series C led by Founders Fund, a C2 and a Series D largely because its forecast kept rising. The pipeline grows by about $2 billion a quarter with a federal sales team of eight.
To handle uncertain demand, aggregate across contracts: you may not know which of 10 bids you will win, but you know you will need at least 100 robotics engineers.
Bet into the growth slope. You can always pause hiring for a quarter, but you can never quickly restart it.
Hadrian standardizes on a small set of robot arm sizes and forward-buys them in bulk so programs can pull them from inventory.
Operating gets harder with scale, but market fundamentals get easier: materials get cheaper, debt gets easier, and vendors put their biggest customer at the front of the queue.
Hadrian was at about 725 employees and planning to add about 600 in four months. It is organized into modular capability and program teams of 30 to 40 with strong engineering leads.
Teams interact through organizational “APIs”, so the weld team does not need to know who leads scheduling.
Cut cultural values down to two you will fire people over. Many cultural values are luxury beliefs of being small.
Invest heavily in standardized interview tests so new managers do not have to design hiring from scratch.
Choose opportunities by mission first, then by doors that will close and never reopen. Hadrian is turning down profitable low-cost interceptor factory work to focus on those.
A CEO’s judgment is calibrated on their own ability and their ten best people, but the tenth program is run by people they have never met.
Write “Ulysses contracts” while clear-headed: rules you will not break when tired or hyped. Hype kills more companies than fatigue does.
Real estate cycles repeat because each new wave of 25-year-old investors has never felt the gut punch of losing their shirt.
Anything is learnable by doing it. Power learned recruiting by doing 20 screening calls a day for his first 12 weeks.
A CEO of a complex company should be the second-best person at every function, good enough to judge excellence and make trade-offs in real time.
Only two kinds of capex-heavy business work: customers prepay for your buildout (like NASA did for early SpaceX), or you architect your way to long-duration infrastructure credit.
Engineering is a subsidy of gross margin. Hadrian was underwater on financing costs until late in its Series B and had to teach every investor its manufacturing cost accounting.
A large pool of capital or a low cost of capital comes from executing better, not from fundraising skill.
Power was wrong about speed. Outside one large rocket company, most customers value flexibility, scale and total program cost over fast delivery.
Hadrian spent two and a half years telling the government it was not ready while giving policy advice. Doing that honestly built the trust that wins contracts later.
Honesty is a slow way to build a billion-dollar company and a fast way to build a hundred-billion-dollar one.
Beware “door number B”: people who signal insider status and think they are in the room. Established families, governments and enterprises build immune systems against them.
Every round gets easier because investors watched 80% of what seemed impossible come true.
With infinite demand, do not try to predict. Maximize the organization’s velocity and let the market show how fast you can grow.
Manage energy, not just hours. Power can do complex product work every day but has about three days of high context-switching in him per week.
Act on hard, emotionally difficult decisions immediately. Power drafts difficult emails until midnight so the only thing left in the morning is to hit send.

Detailed Summary

What it takes to scale US manufacturing

Hadrian has opened six factories since 2022, most recently a 2.2 million square foot facility. Power names three requirements beyond raw horsepower and pain: GPU-style stations that can be configured modularly inside a factory, giving you standards to automate against; a complex organization that can build real manufacturing autonomy; and a careful balance of physical AI software and a new American workforce. Most production lines run linearly, like a CPU. A GPU-style factory is so standardized and automated that high-mix, low-volume work runs as efficiently as low-mix, high-volume work, and peak capacity is managed like a GPU cluster.

Building capacity before the contract

Power compares Hadrian to AWS, which built cloud capacity with Amazon as its first customer. American contract manufacturing is a chicken-and-egg problem. A startup designing a low-cost interceptor would rather not build its own factory, but in the US there is nowhere to start, unlike Shenzhen where a new hardware company can tap existing circuit board capacity. So a manufacturer has to build roughly 1.2 times its future sales capacity before signing contracts, both for optics and because execution is what customers are really buying. Some contracts take two years from the start line.

Turning binary bets into portfolios

Power models risk as burn rate plus execution risk. Equipment always arrives later than promised. The bigger lever is flexibility: if capex and workforce are software-configurable, about 80% of capex can move between programs. Instead of building a $300 million casting facility for turbine blades alone and hoping demand is still there in year four, he breaks every product down to the manufacturing method and the specific equipment, down to pour tonnage by alloy and envelope size by floor space. Paying about 20% more for shared equipment lets one investment serve castings for turbines, munitions and more.

Why every capex bet is company-killing

For large companies capex risk is binary. Nobody gives a production contract to a manufacturer that has not already started executing, so not spending means a 0% chance of growing revenue. Spending the billion is still a company-killing event if contracts do not close, but it moves the odds of revenue above 50%. Power says this holds until roughly $50 billion in scale, and that Hadrian’s risk of death has stayed at about 80% because it keeps growing its capex and R&D balance sheet at venture pace. Deploying $2 billion of capex now is what makes $5 billion in revenue in 2028 possible. If that revenue does not arrive, there are no venture returns and the capex bill is crushing. The risk only retires if growth slows to around 20%.

One-way doors, hedges and rolling a hard six

Asked about sprinting through one-way doors, Power says capex and long engineering projects are all one-way doors. If a welding line must work in month 10 and equipment takes 9 months to arrive, spend three weeks making sure the large single-purpose equipment is right and make rough calls on flexible items like robot arms that can be reused elsewhere. Spending 30% more on hedges beats discovering a mistake when nine months of capex arrives. Software and sales mistakes can be corrected. Capex mistakes cost 18 months. Sometimes, borrowing from Battlestar Galactica, you just have to roll a hard six. On factory construction, he overspends on 20-inch vibration-isolated foundations ($25 million instead of $15 million) and electrical drops everywhere, buying the layout team months of simulation time. There is no asymmetric manufacturing, he adds, only asymmetric products.

Lessons from four years of factory building

Hadrian’s software platform, Opus, takes over the machines’ control software, which requires APIs that most equipment vendors never expected anyone to ask for. One German vendor thought the team were hackers. On the floor, two machine brands reported data inconsistently, half of one vendor’s API calls were not implemented, and integration took 12 weeks for one brand and two for the other. The lesson is to plan for unknown unknowns and assume the industry will fail you. On design, Power says factories are usually designed for the success case. Hadrian builds slack “off-ramps” so a failed part can be pulled aside without backing up five lanes of traffic, like a broken-down car on a freeway. New product introduction needs a physically separate fast lane, because the real cost is the back-and-forth between manufacturing and inspection, and software workflows tend to follow physical layout.

The rusted machines

A couple of years ago, with limited Series A money and a fundraise coming, Hadrian took delivery of new machines from a reputable vendor about four months before key deliverables. Engineers found rust they believed was terminal. Within 48 hours the team traced it to a subcomponent supplier the vendor had changed, affecting a batch of about 40 machines with sequential serial numbers. The vendor’s plan was to fix them one by one by Thanksgiving, which would have killed the company. Power spent a week pressing the vendor’s president and got 30 engineers flown in to finish in four weeks. Company-killing events like this happen about once a quarter, and one is always coming.

Raising the forecast every month

Referencing Zipline, where a new Uber deal pulled the timeline forward two and a half years, Power says Hadrian raised a Series C led by Founders Fund, a C2 and a Series D largely because the forecast kept rising, with about $2 billion a quarter added to the pipeline and a federal sales team of eight. That creates talent and capex problems. The fix is aggregation: across 10 competitive bids you may not know which you will win, but you know statistically you will need at least 100 robotics engineers. Bet into the slope, because you can pause hiring but you cannot restart it quickly, and recruiting itself has long lead times. On capex, Hadrian standardizes robot arms in a few sizes and forward-buys hundreds so programs can pull them from inventory. Operating gets harder with scale, but materials get cheaper, debt gets easier, and suppliers put you at the front of the queue.

Doubling headcount without breaking

At about 725 employees and planning to hire roughly 600 more in four months, Hadrian is organized into modular capability and program teams of 30 to 40, each with strong engineering leadership, around 80 of them. The first five engineers on a team set the bar. Teams interact through organizational “APIs” so nobody needs to know every other team’s lead. Power says to pick what you do not care about: cut ten cultural values to two you will fire people over, because many values are luxury beliefs of being small. Standardized interview tests mean a new manager does not need to design a great hiring process.

Choosing what to pursue

With more demand than capacity, Hadrian chooses by mission (reindustrialization and defense) and by doors that are open now and will never reopen, like programs where a Department of War customer or prime is letting a contractor in for the first time in 50 years. That means turning down profitable factories-as-a-service work for low-cost interceptors, where there will be many programs later. Execution risk is also measured in “Lego bricks”: a program that needs 20 capabilities Hadrian already has 19 of is easy to take on.

Judgment, hype and Ulysses contracts

Power says the hardest judgment is knowing how far over its skis the company is. A founder calibrates on personal ability and their ten best people, but the tenth program is run by strangers, and poor results only show up months later, like a podcast host doing ten episodes a week and seeing the YouTube numbers six months later. Hadrian’s first 18 months, when it raised $1.3 million and set out to build $200 million worth of software, felt like that, as did the last nine months. Like deciding in advance to leave the keys at home after a third beer, he writes Ulysses contracts with himself while clear-headed. A hedge fund investor told him real estate cycles repeat because young investors have never felt the gut punch of losing their shirt. Hype, more than fatigue, is what kills.

Just do it, and be second best at everything

On things he did not know, like government sales, Power says everything in business is learnable and the way to learn is to do it, like lifting weights. In his first 12 weeks he had never recruited, read a few blog posts and did 20 screening calls a day until he got it. A CEO of a complex company should be the second-best person at every function: good enough at engineering, finance and M&A to judge excellence and make real-time trade-offs, because the best people want to work for CEOs who understand their craft.

Financing the buildout

There are only two ways capex-heavy businesses succeed: customers prepay the buildout, like NASA did for early SpaceX (the Pentagon, data centers and consumer energy do not), or the company designs its model to eventually access long-duration infrastructure credit. That means predicting cost of capital and duration over four years. Engineering is a subsidy of gross margin, and a good CEO should not be more than six months off on when it pays back. Hadrian went from venture debt to equipment leasing and was underwater on financing costs until late in its Series B. Power says every venture investor who understands manufacturing has invested, and each round meant teaching investors that Hadrian’s first-principles cost accounting was correct. Big pools of cheap capital come from better execution, not better fundraising.

What Power got wrong: speed

American manufacturing splits into print work (tight tolerance, high compliance, what Hadrian does) and no-print work. Hadrian started with a large rocket company as its toughest customer, assuming other fast-growing companies would pay for speed. There were no other startups of that scale, and most of the market does not value speed at all. That rocket company is the only space or defense prime that can convert speed into value. Most customers value flexibility, scale and total program cost, as long as you are not slow.

Building trust with the government

Power flips the question of building government relationships to win contracts. Create value first, and become trusted enough that customers bring you their hardest problems. For two and a half years, during the Biden administration, Hadrian told the government it was not ready for major programs while offering policy advice on manufacturing, to the point of frustration. Done honestly and not as a sales tactic, that built credibility, as has turning down contracts that would not create value. He compares it to looking for a spouse rather than posting photos with a Porsche. Beware “door number B”: people who signal insider status and believe they are in the room, while serious institutions have immune systems against them. Honesty compounds, every round gets easier, and investors who passed come back after 80% of what seemed impossible comes true.

Velocity, energy and hard conversations

Power’s priorities shift daily. Onboarding, once a known weakness, became critical as the company prepared to add thousands of people, because one extra hour of good onboarding multiplies across all of them. He says you should not try to predict an unpredictable future. With seemingly infinite demand, maximize organizational velocity and let the market show you how fast you can grow. You can always slow down, but you cannot speed up an org quickly. He praises Ramp for shipping at high velocity and announcing products when they are ready rather than to arbitrary dates. The meta-game is energy management. He can do complex product work every day but has about three days of high context-switching in him per week, and structures his work accordingly. His final tactic: when something hard comes up, act immediately. He drafts difficult emails until midnight so the only job in the morning is to send them, and sends a quick “let’s talk tomorrow” note with a calendar invite so he cannot put it off.

Notable Quotes

“The United States has no contract manufacturing capacity. So, if you want anyone to take a big bet on you, you have to basically build at least 50% more like 1.2x the percent of your future sales capacity ahead of having those contracts.”
Chris Power, on why US manufacturers must build ahead of demand

“Manufacturing, what we sell in manufacturing is trust.”
Chris Power, on what customers are really buying

“If you don’t deploy a billion dollars in capex, you’re 100% dead or you’re $100 million acquisition.”
Chris Power, on why capex bets are unavoidable

“Capex and long engineering projects are all one way doors and you’re better off spending 30% more to bake in hedges.”
Chris Power, on sprinting through one-way doors

“Sometimes, actually most of the time, you just got to be right.”
Chris Power, on rolling a hard six

“There’s no asymmetric manufacturing method where you magically going to make a tomahawk casting for a dollar.”
Chris Power, on asymmetry in defense manufacturing

“In a fast growing business, you can always stop hiring for a quarter. You can never restart hiring.”
Chris Power, on betting into the growth slope

“If you’re running a complicated company you have to be the second best person in the company at every function.”
Chris Power, on what makes a good CEO

“That is, I think, a slow way to build a billion dollar company, but it is a very fast way to build a hundred billion dollar company.”
Chris Power, on honesty with customers and the government

“When you’ve got infinite demand, the number one thing you should work on is where are we not at max velocity today and let everything accelerate past the business.”
Chris Power, on planning for an unpredictable future

Watch the full conversation with Chris Power here.