The Startups Rebuilding American Industry | The AD50

Host Elena Burger opens by asking Erin Price-Wright and Connor Love why this year’s AD50 focuses so heavily on the American industrial base. Erin says the central lesson is that “you don’t have a defense industrial base without an industrial base.” Conflicts around the world have exposed the battlefield consequences of decades of hollowing out American production capacity, affecting both U.S. soldiers and allies. In his view, the country has received a wake-up call: it needs to start building again. Connor describes the moment as the convergence of several forces: geopolitical conflict, changing capital markets, and a new generation of entrepreneurs willing to build in physical industries. He argues that these founders are not only addressing national needs but can also create highly valuable companies. The AD50 therefore examines American dynamism “as a totality,” across three connected layers: the inputs that come out of the ground, the factories that transform them, and the end systems—such as defense platforms, energy infrastructure, and robots—that use them. Erin and Connor say each layer matters independently, but the opportunity is greatest when companies understand the entire chain rather than treating mines, factories, and systems as separate worlds.

Connor explains that American technology has historically advanced through government-led efforts such as the space program, the internet, and nuclear energy. What is different now is that private companies and private capital are increasingly initiating the work, even when government customers are an important early market. The resulting factories, materials, and components have many uses beyond defense. He cautions that “dual use” can be a distracting startup strategy when founders try to serve too many markets at once, but says the underlying industrial assets are inherently dual use: materials and components made for defense can also go into cars, toys, medical devices, and robots. They use DJI as an example. The company began with consumer drones, expanded into public safety and defense drones, and ultimately became the world’s largest robot producer because drone motors resemble the actuators that are among the hardest parts of a robot to build. Erin says this is a recurring pattern in which defense or government demand subsidizes process knowledge and technology that later benefits consumers. A new PCB factory illustrates the economics: a defense customer can help fund expensive machinery, after which the same factory can supply PCBs for robots or hearing aids. The government’s willingness to pay for early capacity can make factories financially viable at a scale commercial buyers could not support initially.

Connor and Erin discuss when startups should build their own supply chains. Connor points to SpaceX and Tesla, which had to vertically integrate because the U.S. lacked the industrial base needed for their first-of-a-kind products. That approach is difficult—an entrepreneur focused on an end system may also have to develop ball bearings, solenoid valves, batteries, power electronics, or drivetrains—but it can be necessary when reliable suppliers do not exist. Their conclusion is not to vertically integrate everything: companies should use strong suppliers where they exist, while controlling critical components when the available supply chain cannot meet the required cost, speed, or quality. They cite Mariana Minerals as an example of end-to-end ownership. The company acquired a copper mine and handles asset acquisition, development, mining, operation, and refining across multiple minerals before selling a finished commodity. Erin says existing mining companies were unlikely to adopt new technology quickly enough, while Mariana wanted to capture the value of changing the entire process. Connor adds that software and AI are applied at every stage—from planning and operating the mine to refining—and the resulting data feeds back into the process. The same logic applies to Hadrian’s factories, where quoting, manufacturing, inspection, and material transfer are treated as connected processes. AI is useful in the physical world, but Connor stresses that “AI cannot turn a wrench yet”; success requires understanding and changing the underlying physical workflow, not simply adding an AI layer.

Erin introduces the distinction between low-mix, high-volume manufacturing and high-mix, low-volume manufacturing. Traditional automation economics favor producing large quantities of one product because a highly customized production line is expensive to change. American manufacturers often need to make many different products in smaller quantities, which has historically made automation difficult to justify. The companies in the AD50 are changing that calculation by giving manufacturing technicians software tools to reprogram and reconfigure lines themselves. Erin emphasizes that these technicians do not necessarily have college degrees or software-engineering backgrounds; they understand the physical manufacturing process and can now use cloud-based tools to modify it. Instead of calling a systems integrator and waiting three months while spending hundreds of thousands of dollars, a technician may be able to reconfigure a line in about 20 minutes. Connor connects this capability to the broader systems companies in the AD50. Faster, more flexible production allows a company to iterate on prototypes quickly and then reach volume production sooner. It embodies “design for manufacturing” from the start: the end product, its subsystems, and the factory that makes them are designed together. They point to Anduril and Saronic as examples of companies building large facilities intended to produce thousands, tens of thousands, and eventually potentially hundreds of thousands of systems—something they say the United States has not done at this scale for a long time.

Connor says defense procurement has historically been fragmented: the buyer is not the user, and the person appropriating funds may be far removed from both. Traditional prime contractors often waited for the government to specify and fund a product, resulting in systems delivered years later that reflected outdated technology and did not serve soldiers well. The newer model uses venture capital to build ahead of a formal government request. Castellan, for example, spent roughly three years privately developing a difficult, reliable hypersonic missile before the government began awarding contracts and buying missiles. Connor argues that this produces better products, faster iteration, and new defense companies that can take market share from incumbents. Erin stresses that these businesses are personally demanding. Mariana’s CEO, Turner, travels to remote locations without cell service or electricity, spends weeks or months away from his family, and works with suppliers whose culture and timelines differ from those of software startups. Building physical things is “rough,” not merely romantic. Yet both investors say founders are drawn to problems large enough to justify that effort. Companies such as SpaceX, Tesla, Palantir, and Anduril have shown that industries once considered merely important to national safety can also produce some of the world’s most valuable companies. They note that most of the ten most valuable companies globally are hardware companies, and that ambitious founders now see rebuilding energy, manufacturing, defense, and mining as economically worthwhile as well as nationally important.

The speakers describe a contagion effect in which building defense systems creates demand for shared subsystems—motors, actuators, valves, switches, electronics, and other components—that can later make robotics and other commercial markets viable. Connor says most robotics companies still source the majority of their components from China, but defense-driven production could establish domestic alternatives with better cost, quality, speed, and geopolitical resilience. Erin gives the example of a portfolio company that once ordered twice as many PCBs as it needed because roughly 50% failed quality tests. New U.S. suppliers using software and modern manufacturing methods have raised passing yields to 70%, 80%, or 90%. Connor argues that new end systems are also forcing a change in the middle of the supply chain. In the old model, a company such as Boeing could request another familiar landing gear from a supplier, which had little incentive to redesign a component that already met the specification. New aircraft, military systems, electric vehicles, and energy architectures have requirements that old components cannot optimally satisfy. The opportunity is therefore a “manufacturing middle” made up of suppliers that co-engineer with end-system companies, rethinking and redesigning components rather than merely receiving specifications. They say this is also occurring in energy, where static mechanical grid equipment from the 1970s and 1980s is being replaced by software-native power electronics and battery-storage systems. Once a new architecture offers better cost, efficiency, serviceability, or manufacturability, Connor says the market is unlikely to return to the old one—even if the original projected demand is only partly realized.

Connor says the United States’ advantage over China is its enormous pool of private capital, which can follow entrepreneurial and technological success rather than being allocated through a single top-down industrial policy. Venture capital, private equity, credit firms, and major banks are increasingly funding factories, energy infrastructure, defense systems, critical minerals, and other physical assets. He says this matters because a Falcon 9, a missile, or another complex physical system requires substantial upfront capital. Private funding can allow companies to build before government demand is fully guaranteed. Erin adds that in the first half of 2026, more risk capital went into defense than during all of the previous year combined, following several years of rapidly increasing investment. Asked why the United States has fallen so far behind China in new energy production, Erin rejects the idea that technology is the main obstacle. He says the country must decide whether it has the will to approve and build the necessary projects; if it does, technology and capital will follow. Both speakers warn that physical infrastructure takes years, not the speed of a software deployment. Erin also says communities must see themselves participating in the economic gains. A mine, data center, power plant, or factory is visibly located in someone’s neighborhood, unlike much software infrastructure. He praises companies such as Amca for partnering with small machine shops, preserving and growing their businesses rather than simply cutting staff. The long-term opportunity, he argues, is more jobs, prosperity, abundant energy, and less concentration of wealth on the coasts—but only if companies bring communities along.