IBM research notes discovering a new material can take 10 years and up to $100 million, whether it’s the alloy inside a jet engine, the coating on a chip, or the metal lining a reactor wall. That’s a decade between the first idea and the factory floor, and most ideas never make it that far.
That number should worry anyone watching the AI boom. Morgan Stanley estimates that hyperscalers and AI cloud providers will spend nearly $1.6 trillion in 2027, yet Accenture’s global CTO recently put the number of large companies truly reinvented around AI at “basically close to zero.” Software got faster. The physical world didn’t.
I’m a materials scientist, and I spent years at an Army research lab before starting Radical AI. What’s holding AI back right now has less to do with the models than with the materials we build everything on. I call it the decade tax, and as long as we keep paying it, the AI boom will keep outrunning the economy it’s supposed to transform.
The decade tax
The decade tax is the time and money between “we need a material that can do this” and a part that’s actually in production. Every industry that builds physical things pays it.
It exists for three reasons.
Discovery still runs on trial and error. A researcher designs a composition, makes a small batch, tests it, waits for results and adjusts. Each loop takes weeks. A good career might run a few thousand of them.
The scientific literature has grown past the point where any person can keep up with it. The AI research agents in our labs draw on 583,813 papers, more than one researcher could get through in ten lifetimes. Meanwhile, important findings sit in separate journals, and no one ever connects them.
The equipment was never built for automation. The instruments materials scientists trust were designed to be run manually.
The result is that in 22 years, the entire field of HEA research has reported the experimental study of 3,493 alloy compositions. With our self-driving lab we found 300 in 16 weeks. The scientists themselves aren’t slow; it’s the archaic process holding discovery back.
Why the decade tax caps the AI boom
Every big AI promise eventually turns into a materials problem.
Start with chips. AI runs on semiconductors that are hitting limits at the atomic scale, and the materials inside them are getting harder to secure. Hafnium, a critical material in every leading-edge transistor, has risen nearly 10x in price in the last four years.
Then there’s power. Data centers need enormous amounts of it, and the fixes everyone is counting on, from fusion to a stronger grid to better batteries, depend on materials that can survive extreme heat and stress. Many of those materials don’t exist yet.
Defense runs into the same wall. Hypersonic systems and next-gen jet engines are still built on materials designed decades ago.
Even the materials we already rely on aren’t guaranteed. China’s toughest rare earth export controls aren’t gone. They’re on hold, and the current truce runs only to January. If it switches them back on, chipmakers and defense contractors will find out what happens when a material supply disappears. The decade tax is why there’s no backup ready.
The hidden cost is bigger than any one of these. When a new material takes 10 years, engineers design around old ones instead. That cost never shows up on a balance sheet, because no one can measure the value of a product that was never built.
The real race
Every major economy wants to lead in AI, and the country that pulls ahead in self-driving lab technology will be the one that gets AI into physical products fastest.
Washington understands this. The Department of Energy’s Genesis Mission names AI-driven autonomous labs as one of its core challenges, and Radical AI signed a collaboration agreement in December 2025 to support the mission, yet most corporate R&D still works the way it did 30 years ago, with one experiment, one result and one decision at a time.
What cutting the tax looks like
The fix is AI that does more than predict. It runs physical experiments. It reads the literature, proposes a material, makes it, tests it, and learns from the result. Then it tries again. All day and all night.
We built that kind of lab in New York. In one campaign for hypersonic applications, our self-driving lab discovered a new material that outperformed the industry’s best. Along the way, the lab has generated 70 million experimental data points, a record no human team could build by hand.
That data matters more than anything else we do. In materials science, a model has to have learned from real experiments conducted in the physical world.
The bottom line
The AI race will be won by whoever turns intelligence into physical things fastest.
Think about what that speed could unlock: chips that run cooler while delivering more power, turbine blades that last twice as long, reactor walls that can withstand the heat of fusion. None of these exist yet. AI is ready to help build them, but the materials they depend on are still waiting out the same 10-year timeline.
The alternative is a rival country or foreign competitor cutting the decade tax first. American companies would end up licensing the very materials their own AI boom runs on, and the trillions spent on data centers would help build someone else’s industrial lead.
That’s avoidable. Companies already treat AI spending as a core investment, with budgets, owners and deadlines. Materials R&D deserves the same treatment: a timeline to cut, not a cost center to manage. The tools to do it exist now.
Trillions of dollars are pouring into AI, and the decade tax is what stands between that money and the real economy. Whoever cuts it first will shape the next era of industry, while everyone else keeps paying it.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
