Defense demand is building an industrial base robotics could inherit

1. Industrial base first — ▶ 1:03

   A defense industrial base cannot exist without an industrial base beneath it, and the investors say the decades-long hollowing-out of that base has had real consequences on the battlefield for American and allied soldiers. This year's AD50 list is organized around three lenses: inputs from the ground, the factory that makes the thing, and the end system. They argue that few people think about these companies end to end, though their portfolio companies increasingly do. They see conflict, maturing capital markets and a willing entrepreneurial base converging at once.

2. Defense as first buyer — ▶ 5:09

   First-of-kind production is expensive, and a defense buyer willing to pay a price the consumer market would not changes the unit economics from the start. The investors cite a US PCB fab: if the military subsidizes the capex, the same boards also serve robots and hearing aids. DJI is offered as precedent. It moved from consumer drones into defense production, and its drone-motor expertise carried over to robot actuators, which they call the hardest part of a robot. The consumer market then pushes products to become more capable and cheaper.

3. Integrate when no base exists — ▶ 8:38

   SpaceX and Tesla vertically integrated because no US industrial base existed to support them, including battery supply chains and drivetrains. The investors agree companies should use high-quality suppliers where they exist, though exquisite systems like missiles may require integration. Mariana Minerals shows the logic: it acquires, develops, mines and refines assets itself, because incumbents would not adopt technology fast enough and ownership captures the value. Owning the process end to end also means capturing data that feeds back into it. In the physical world, they say, showing is believing.

4. Line changes in 20 minutes — ▶ 14:12

   AI is not a silver bullet in physical industry. The reality, one investor says, is that it cannot turn a wrench yet. The gain comes from many process improvements across a factory, from quoting to inspection. The sharpest example is low-mix high-volume versus high-mix low-volume manufacturing. Technicians who are not software engineers can now use tools like Claude to reconfigure a line in about 20 minutes, versus three months and hundreds of thousands of dollars through a system integrator. That speeds prototyping and the move to rate production.

5. Building ahead of requirements — ▶ 18:04

   In traditional defense procurement the buyer is not the user, and the appropriator of funds is further removed. Primes waited to be told what to build, so soldiers received systems designed a decade earlier. Private capital now funds products before the government commits. Castelion, building a hypersonic missile, spent roughly its first three years on venture money before the government bought missiles. The investors expect better products and rising market share for new entrants. Castelion's founders also resisted pressure to become multi-product, concentrating on one hard problem for two and a half to three years.

6. Hardware becomes venture-scale — ▶ 22:05

   The category was always important to national security and daily life. What changed, they argue, is that it can now produce the most valuable companies, which makes it economically worth pursuing for ambitious founders. Palantir, Anduril and SpaceX are offered as proof, along with the observation that most of the world's top 10 most valuable companies are hardware companies. The work is brutal: remote sites, weeks away from family, tough supplier negotiations. The investors say that is why missionary founders gravitate to it, and why site visits repeated every three months show how fast a buildout moves.

7. Subsystems cross into robotics — ▶ 28:45

   Military systems built at volume require actuators, motors, valves and switches, the same subsystems future robots need. Building them to quality and cost could make robotics economically viable without reliance on China, where an investor claims a robotics company has a 99% chance of buying most of its parts. Supplier ecosystems are already maturing. One portfolio company once ordered 2x the PCBs it needed, expecting to discard 50% for quality failures. After new US suppliers emerged, the same company reports 70-90% passing testing. The lesson is that co-location with customers raises quality.

8. The manufacturing middle — ▶ 32:05

   Subtier suppliers have spent 40 years in receive mode, building to spec. The example is Honeywell's landing gear for Boeing's 737, which is rational to reuse because the revenue is the same whether the gear is 50 years old or new. As end systems change, aging subsystems become a constraint. The investors expect a new layer of suppliers that co-engineer with OEMs and push design upward. They call it the manufacturing middle, an American manufacturing asset class. They distinguish it from private equity, which has long improved assets' costs without changing what is built.

9. Demand that cannot reverse — ▶ 36:39

   The investors anticipate a pessimist's objection: not every new military system will be adopted, and energy demand may not rise 4x. Their response is that it may not matter. Once a top-down demand cycle begins and companies redesign from first principles, they say, the new designs often cost less, are more efficient, easier to service and easier to manufacture, so buyers will not revert. Even delivering half the projected requirement would be enough to avoid going back. The grid is their example, shifting from static mechanical systems deployed in the 70s and 80s to software-native power electronics and batteries.

10. Capital without industrial policy — ▶ 41:16

    The US advantage, the investors argue, is that trillions of dollars of private capital follow technological and production success without a state-directed industrial policy, so money can fund things before they are obvious. They say JP Morgan has committed trillions of dollars to the physical world, and private equity and credit firms are now active. Their figure: first-half 2026 risk dollars into defense exceeded all of last year's combined, which was itself many times the previous year's. They say the defense valley of death, the wait for a contract to fund capex, is starting to disappear.

11. Will, not technology — ▶ 44:52

    Asked about the chart of flat US power production against China's rise, the response is that it is not a technology problem. It requires soul-searching about the American people's will to grow and to let projects get built, after which technology and capital will follow. The investors add that physical projects take years and are visible to nearby communities, so participation in the upside matters. They hold up Hadrian as an example that works with small machine shops rather than cutting management, and say this is something the tech world still has to wrestle with.

The investors' central claim is that defense, energy, robotics and consumer hardware are not separate industries. They draw on a single supply chain, and the cheapest way to rebuild it is to let defense buyers pay for the expensive first units. Their reasoning is mechanical. First-of-kind production is costly, whether the product is a bespoke military system or a consumer device. A government willing to pay a price the commercial market would not changes the unit economics from the start, which funds more machines and capex. Their example is a US PCB fab: a defense buyer subsidizes the equipment, and the same boards later go into robots and hearing aids.

The precedent they offer is DJI. It began with consumer drones, built a large-scale defense and public-safety drone business, and now, they say, is the world's largest producer of robots. A drone motor and a robot actuator are similar, and the actuator is the hardest part of a robot. The investors read this as process knowledge moving from military to civilian use, with the government acting as a subsidizer of learning and the consumer then pushing products toward being more capable and cheaper. It is a tidy story, and it rests on a single case. The transcript offers no systematic evidence that the transfer reliably happens.

The second claim concerns who funds the learning. In traditional procurement, they say, the buyer is not the user and the appropriator of funds is further removed, so primes wait to be told what to build. The result is a soldier receiving a system designed a decade earlier. Private capital now funds products before the government asks. Castelion's hypersonic missile ran on venture money for roughly three years before the government bought missiles. The investors expect better products and growing market share for new entrants. This also explains why they treat capital markets as central. Systems like a Falcon 9 or a Blackbird missile are expensive to build, and a source of money that does not depend on a prior contract lets production lines get built first.

Software and AI enter this picture in a more modest role than in other sectors. The investors insist there is no single silver bullet. A factory or mine improves through 15 or 20 changes, from quoting an order to inspecting a part, and an owner starting from a clean slate can examine each node from first principles. Hadrian is their example for factories and Mariana Minerals for mining. One investor says plainly that AI cannot turn a wrench yet. The practical gain they describe is narrower and easier to credit: a manufacturing technician using Claude can reconfigure a line in about 20 minutes instead of waiting three months and paying hundreds of thousands of dollars to an integrator. That matters because low-mix high-volume manufacturing, common in the US, is hard to automate when every line change is costly.

Ownership and integration are handled with some nuance. SpaceX and Tesla integrated vertically because no base existed to support them, and the investors accept that companies should use good suppliers where those exist. Mariana, by contrast, owns the full chain from asset acquisition to refining and sells a commodity product, because incumbents would not adopt new technology fast enough and ownership lets it capture the value and the data. The deeper argument is about suppliers who do exist. The investors describe subtier suppliers in receive mode for 40 years, rationally reusing a 50-year-old landing gear because it earns the same revenue. If end systems keep changing, they expect a manufacturing middle to emerge that co-engineers with OEMs and pushes design upward.

The spillover into robotics is where the argument becomes most consequential, and also most speculative. Military systems built at volume need actuators, motors, valves and switches, which are the subsystems future robots need. If these are built in the US to quality and cost, robotics becomes viable without reliance on China, where one investor claims a robotics company has a 99% chance of buying most of its parts. There is some supporting evidence in the supplier base. A portfolio company that once ordered 2x the PCBs it needed, expecting to discard 50%, now reports 70-90% passing quality testing from new US suppliers. That is one company's experience, and the investors draw from it a general lesson about co-location with customers.

The most notable counterargument comes from the investors themselves. They acknowledge that not every new military system will be adopted and that energy demand may not rise 4x. Their rebuttal is that the redesigns themselves are the lasting asset: once a company rethinks an architecture from first principles, the result tends to be cheaper, more efficient and easier to service and manufacture, so buyers do not go back. The grid, they say, is moving from static mechanical systems of the 70s and 80s to software-native power electronics and batteries. This logic holds if the new designs really are better. It is weaker if the demand signal that justified building them fades and the supplier base cannot sustain itself.

Two other constraints get honest treatment. One is capital: they describe a loop in which emergent winners attract more money, but concede that at some point there must be a pot at the end of the rainbow. They also report that first-half 2026 risk dollars into defense exceeded all of last year's combined, which shows the loop is turning, not that it is sustained. The other is time. Moving dirt and understanding a supply chain takes years, and they call for patience from capital markets. Neither is a reason to doubt the thesis, but both are reasons to be careful about its timeline.

The last constraint is not technical at all. Asked about the US power gap with China, the answer is that this is not a technology problem but a question of whether Americans want to grow and let projects be built. Technology and capital, they say, will follow. The point extends beyond energy. Physical projects sit in someone's backyard, unlike software, which could remain abstract and concentrated in a few cities. The investors suggest companies must bring communities along, and they cite Hadrian's partnerships with small machine shops as a good model. If they are right that the technology and the capital are arriving, the binding question becomes whether local consent and visible benefit arrive with them.