We Cannot Build Tomorrow's Weapons With Yesterday's Alloys

The alloys inside the last plane I flew in were designed in the 1950s, '60s and '70s. In aerospace, relying on decades-old materials is normal. Established materials have earned engineers’ trust through years of testing and service. But the next generation of military capabilities will demand performance that existing materials cannot deliver.

Washington spends considerable time on the defense industrial base: factories, workforce, and budgets. Materials discovery deserves the same attention. This year, DARPA kicked off its Next Generation Hypersonic Cruise Missile program and flagged high-temperature materials among the areas that need to improve. Hypersonic vehicles must withstand extreme heat and oxidation. Directed-energy systems, next-gen armor, and advanced reactors all require materials we don't have yet.

We don't have a manufacturing problem. We have a discovery problem. New materials take decades to find and qualify, and the programs that need them can't wait that long.

Part of the challenge is the industry’s structure. Universities discover. Materials companies scale. Defense contractors, equipment manufacturers, and primes build. Every handoff adds years to the timeline.

So engineers do the rational thing. They design around what already exists because a new material may not arrive before the program needs it. We end up engineering the future around the limitations of materials developed for the past.

That habit carries a strategic cost. Many legacy alloys depend on critical minerals with supply chains that run through adversarial nations. C103, a niobium alloy long used in rocket and hypersonic hardware, is roughly 10% hafnium by weight. Hafnium prices have risen 10 to 15 times since 2021, and China controls most of the supply. Under the traditional model, designing a replacement takes longer than most programs can wait.

AI alone doesn't fix this. Models trained on simulated data make good predictions inside known territory and unreliable ones outside it. In materials, the ground truth is the material itself.

What can change the timeline is pairing AI with a self-driving lab. AI agents review the literature and past experiments to propose new compositions. Robotic systems synthesize those candidates, characterize them and test their properties. Every result, including every failure, feeds back into the models overnight. The failures matter most. A well-characterized failure shows the system where the physics isn't understood.

This goes beyond automating a sequence of experiments chosen by a human. A self-driving lab uses results to determine what to test next, working toward performance targets set by scientists.

I’ve seen what this looks like in practice. In one 16-week campaign focused on high-temperature alloys for hypersonic applications, a self-driving lab synthesized more than 300 novel compositions. The top performers were independently tested by Purdue Applied Research Institute and outperformed the industry-standard material. For context, the entire field has reported roughly 3,400 alloys of this class over more than 20 years.

Discovery is step one. Those alloys still have to be scaled, processed, and qualified. The point is that with AI, robotics, and experimental data, the discovery timeline can be dramatically compressed. Work we assumed took years can take weeks.

Once discovery takes weeks, the sequence changes. Today, materials and products move in series. Engineers assess the available materials and design solutions that account for any performance shortcomings.

Fast discovery lets materials development run in parallel with the product roadmap, so engineers can design materials that meet the specs they actually need. When requirements shift, the materials development shifts with them. Engineers have long called this concurrent engineering.

None of this removes scientists. It changes where they spend their time. A metallurgist who once worked one research problem at a time can now direct ten campaigns at once. The machines do the repetitive work. Humans make the judgment calls: which results matter, which tradeoffs to accept, what's worth scaling up. This technology will make scientists more important, not less.

Closing this gap takes more than one lab. Federal efforts like the Department of Energy's Genesis Mission show the momentum is there. This is the moment for private industry and public research to build this capability together.

Two steps would get us there. First, when a program defines a capability, it should set the material property targets the same day and begin discovery in a self-driving lab alongside the engineering. Second, the United States should treat self-driving labs as national infrastructure and work with the national labs already building them, including Argonne, Oak Ridge, and Lawrence Berkeley. Materials can't keep arriving a decade after the design.

The United States and its allies are in a materials competition they don't always recognize as one. China has concentrated control of critical minerals and brings sustained focus to materials.

The next capability race won't be decided only by who builds the most factories or funds the most programs. It will be decided by who can find and qualify the materials those programs need, fast enough to matter. That isn't a technology problem. It's a national capability problem, and it's one that must be solved.

Joseph F. Krause is a materials scientist and the co-founder and CEO of Radical AI. He previously served as a Journeyman Fellow in the U.S. Army DEVCOM Army Research Laboratory.