Why Founders Are Betting Big on Factories, Mines, and Energy

One Stack, Rebuilt From the Ground Up

This year’s AD50 treats defense, energy, manufacturing, and robotics as a single stack, running from the critical materials pulled out of the ground, through the factories that process them, to the end systems they power. Erin Price-Wright and Connor Love argue that the hollowing out of American industry over several decades now has real consequences on the battlefield for American soldiers and allies. Geopolitical conflict, an evolving capital markets ecosystem, and a new wave of entrepreneurs willing to build in these categories have converged at the same time.

Dual Use and the Government as First Customer

For most of the twentieth century, the government drove foundational technology, from spaceflight to the internet to nuclear energy, and the commercial economy inherited the long tail of those investments. What is different now is that private companies and private capital are willing to be the impetus themselves. Defense demand remains the tip of the spear, but the systems being built, whether a factory, a battery, or an actuator, inherently have many use cases. The phrase ‘dual use’ is common in investing, and the speakers see it as literally true of any factory or energy asset being built today.

DJI illustrates the transfer. It began with consumer drones, built a large defense business, and is now one of the world’s leading producers of robots, because a drone motor and a robot actuator are very similar, and the actuator is the hardest part of building a robot. The federal government can subsidize the process knowledge and early technology, while the American consumer’s discernment pushes products to be more capable and cheaper than government procurement ever would.

Paying for the First of a Kind

The first unit of anything is expensive, whether it is a bespoke military system or a consumer product. A new PCB fab in the United States is a costly operation, but a defense buyer that pays for early capacity can make that investment work, because the same boards go into robots and hearing aids. The change is in the unit economics from the start: instead of building 100 things for a commercial market, a company can build 100 things at a price the commercial market would not pay, and that funds the machines and capital expenditure that later make physical products financially viable at consumer scale.

Vertical Integration and Owning the Process

SpaceX and Tesla are presented as the few modern companies that have built anything at scale, and both did it by integrating vertically. Where high-quality suppliers exist, they should be used. But SpaceX and Tesla were doing things that had no domestic industrial base to support them, so they had to build it. Tesla had to recreate the battery supply chain and build power electronics and drivetrains largely from scratch, drawing on papers from the 1970s that had sat unused for decades.

Mariana Minerals shows the same logic applied to mining. Existing mining players were not going to adopt new technology fast enough, so the company decided to own the full value chain: acquiring, developing, mining, operating, and refining the asset, across several minerals, and selling the final product. Once a company owns the process end to end and runs software at each stage, it captures data that feeds back into the process. Hadrian is cited as a case where a factory built from a fresh slate lets founders examine every node of production from first principles and ask whether a process that has run for a century is still the right one.

AI Is a Tool, Not a Replacement for Process Knowledge

The speakers argue that software and AI will look different in the physical world than in legal work or customer service. Digital products can be seen and transferred easily, while physical progress often requires owning the asset to change it. AI will be a powerful tool, but it cannot turn a wrench yet. Robotic models and autonomy are getting closer, yet the real advantage lies in a deep understanding of how a physical process works and how to change it in the world.

One concrete example is the choice between low-mix, high-volume and high-mix, low-volume manufacturing. Automation has traditionally paid off only for producing large volumes of one thing, and the United States does a lot of low-mix, high-volume work, which is hard to automate because lines must change constantly. Manufacturing technicians, who often lack software engineering degrees, now have the intuition and the software tools to reconfigure lines themselves. Work that once required a system integrator, months, and hundreds of thousands of dollars can be done by a manufacturing engineer using Claude in about twenty minutes.

Design for Manufacturing and Faster Iteration

Much attention has gone to systems companies such as Anduril, Saronic, and Castelion. But if a company cannot iterate on its design and prototype quickly, it will move slowly, and by the time it reaches production it must rework subsystems to reach scale. A transferable, adjustable production line lets a company build prototypes faster and then move each component to rate production sooner. This is the lesson from Tesla and SpaceX, now embedded in systems companies. Anduril’s facilities in Ohio and Baltimore and Saronic’s shipyard work are described as massive facilities that will soon produce thousands or tens of thousands of systems, something the country has not done in a long time.

Building Before the Government Asks

Defense has historically been one of the hardest sectors to build for. The buyer is not the user, and the appropriator of funds is far removed from both. Primes would wait for requirements, deliver systems that were often a decade behind, and soldiers ended up with equipment that did little for them. Now the military’s needs can be anticipated, and private capital will fund companies to build ahead of formal requirements. Castelion’s hypersonic missile is the example: it was private venture capital, not government contracts, that carried the company through its first years of risky development. The same logic is said to be emerging in energy, where the grid is being rewritten, though the speakers believe the country is only on the one-yard line.

Missionary Founders and the Reality of Hard Work

These companies are hard to build. Founders travel to remote locations without cell signal or electricity, spend weeks away from their families, and negotiate with suppliers whose culture is very different from software. Physical work is rough: it means moving dirt and understanding supply chains that take years to mature. The speakers say that the ambition of these founders is what drives results, and they describe the tradition of company sites competing to fly the largest American flag. Visiting a factory repeatedly over twelve months reveals how quickly a buildout accumulates, from an empty warehouse with a forklift to a crowded floor with power demands that outstrip the grid.

Shared Subsystems Connect Defense and Robotics

The subsystems in military systems, including actuators, motors, valves, and switches, are the same ones future robots will require. When defense builds quality and cost for those components, markets that were not viable become viable. Robotics companies building deployments today may buy most of their components from China, and onshoring those purchases is about quality, cost, speed of iteration, and geopolitics. A portfolio company is cited as an example: three years ago it ordered twice the PCBs it needed because half would fail quality tests. New startups applying software to PCB manufacturing now push its pass rate to 70 to 90 percent. Close co-location and co-development with end customers matter more than people realize.

The Manufacturing Middle

New systems such as Joby and Archer aircraft, new military systems, and new cars are all being designed differently, and their inputs must change too. Historically, Boeing has called Honeywell for the 737’s landing gear, and Honeywell has reused the same design for decades because that is rational for a supplier receiving specifications. But when requirements change, and new materials and designs make old components inadequate, that middle layer cannot change the requirement on its own. The speakers see an emerging group of supply companies that will co-engineer with original equipment manufacturers, rethinking not just the end system’s speed, weight, or mass but each subcomponent. They call this the manufacturing middle, and they describe it as creating an American manufacturing asset class rather than simply buying and optimizing existing assets. The energy grid is a parallel case, shifting from static mechanical systems built in the 1970s and 1980s to software-native power electronics and battery storage that require a rethought supply chain.

Second-Generation Founders and Doing One Thing Well

The most promising founders often come from a painful problem they encountered inside an organization such as Anduril, SpaceX, or Tesla. They see the problem is broader than their employer, build a company around it, and find a ready customer base in the ecosystem they came from. The speakers compare this to how Silicon Valley was built and find it encouraging to see the pattern in physical supply chains. The SpaceX-style focus of learning to do one thing relentlessly well is key. Palantir, SpaceX, and similar companies were overnight successes that took about twenty years, and the process knowledge gained from the first hard product makes the second and third far easier. Customer trust and manufacturing capability compound, but only if the first core product is delivered.

Capital Markets Without a Top-Down Plan

The United States has trillions of dollars of private capital that follow technology success and production success. Unlike China, the country has no top-down industrial policy directing money, and the speakers see this as a strength, because capital follows the entrepreneur and the success. What has changed is that capital markets now view defense, energy, and industrial categories as attractive. Major banks, venture firms, private equity, and credit firms are committing large sums to physical systems. The speakers report that in the first half of 2026, more risk dollars went into defense than in all of last year combined, and that the valley of death, where companies waited for government contracts to finance capital expenditure, is starting to disappear.

The Power Gap Is a Question of Will

Asked about the widely shared chart showing American power production flat while China’s new energy production climbs, Love argues the gap is not primarily a technology problem. It is a question of American will: whether the country wants conforming projects built and wants to grow. If it does, technology and capital will follow. Price-Wright notes that the snowball is rolling, and the open question is whether emergent winners will attract further capital. Physical infrastructure takes years, and patience is required, but the nation’s future in these sectors is, in their words, ours to lose.

Bringing Communities Along

Physical projects have a visible presence in the communities where they are built, unlike software, which was largely abstract and concentrated in a few cities. Residents of a town hosting a data center, power plant, or mine need a stake and a say in the economic upside. The speakers believe the capital markets will ultimately reward companies that bring communities along, citing Amco’s practice of partnering with small machine shops to grow their existing businesses rather than cutting staff. Because many of these companies will physically exist in middle America, the speakers see more jobs, more prosperity, more energy, and less concentration of wealth on the coasts as the likely outcome, though it will take time.

• Tags:
• Robotics
• Vertical Integration
• Critical Minerals
• Physical AI
• American Dynamism
• Defense Tech
• Advanced Manufacturing
• Co-Engineering
• Industrial Base
• Dual-Use Technology
• Energy Infrastructure
• Capital Markets
• Manufacturing Middle
• Hardware Startups