Innovation & Tech

AI’s decade tax: $1.6 trillion meets a 10-year materials delay

IBM Research says new materials can take 10 years and up to $100 million to discover, while Morgan Stanley sees hyperscalers and AI cloud providers spending nearly $1.6 trillion in 2027.

By Amara Osei

4 min read

New materials can cost 10 years and up to $100 million to discover. That ‘decade tax’ is why America’s AI boom hasn’t re
New materials can cost 10 years and up to $100 million to discover. That ‘decade tax’ is why America’s AI boom hasn’t reDarkoStojanovic / pixabay

What's News

  • IBM Research: discovering a new material can take 10 years and up to $100 million.
  • Morgan Stanley: hyperscalers and AI cloud providers will spend nearly $1.6 trillion in 2027.
  • Hafnium, critical in every leading-edge transistor, has risen nearly 10x in price over the past four years.
  • Radical AI: a self-driving lab found 300 high-entropy alloy compositions in 16 weeks, versus 3,493 experimentally studied across 22 years, and has generated 70 million data points.
  • Radical AI signed a collaboration agreement with the Department of Energy’s Genesis Mission in December 2025.

By Amara Osei

IBM Research says discovering a new material can take 10 years and up to $100 million. The targets range from a jet-engine alloy to a chip coating or a reactor-wall metal. The commentary’s author, a materials scientist who spent years at an Army research laboratory before starting Radical AI, says most ideas never reach the factory floor.

Morgan Stanley estimates that hyperscalers and AI cloud providers will spend nearly $1.6 trillion in 2027. Accenture’s global CTO recently said the number of large companies truly reinvented around AI is “basically close to zero.” The author puts it plainly: “Software got faster. The physical world didn’t.”

Why does a 10-year discovery cycle constrain AI?

The author calls the delay the “decade tax”: the time and money between “we need a material that can do this” and a part in production. Every industry that builds physical things pays it, according to the commentary.

Three forces sustain the tax, the author argues. Discovery still runs on trial and error: a researcher designs a composition, makes a small batch, tests it, waits weeks for results and adjusts. A good career might run a few thousand of those loops.

The scientific literature has also outgrown any individual. AI research agents in Radical AI’s labs draw on 583,813 papers, the author says, more than one researcher could read in ten lifetimes. Important findings remain scattered across separate journals, with no mechanism connecting them.

Materials equipment compounds the bottleneck. Instruments scientists trust were designed for manual operation, not automation, according to the author. In 22 years, high-entropy alloy research has reported experimental studies of 3,493 compositions. Radical AI’s self-driving lab found 300 in 16 weeks.

What does the decade tax cost AI companies?

AI’s semiconductor demands now meet physical limits at the atomic scale, the author argues, while securing the materials inside chips has become harder. Hafnium, critical in every leading-edge transistor, has risen nearly 10x in price over the past four years.

Power demand creates a second materials test. Data centers need enormous amounts of electricity, and fixes ranging from fusion to a stronger grid and better batteries depend on materials that can withstand extreme heat and stress. Many of those materials do not yet exist, according to the author.

Defense encounters the same wall. Hypersonic systems and next-generation jet engines still rely on materials designed decades ago, the commentary says.

China’s toughest rare-earth export controls remain on hold rather than gone, and the current truce runs only to January, the author notes. If Beijing switches them back on, chipmakers and defense contractors will learn what happens when a material supply disappears. No backup is ready because the decade tax delays it.

The tax also hides a cost from company accounts. When a new material takes 10 years to develop, engineers design around existing materials instead. The balance sheet never records the forgone value of a product that was never built, according to the author.

How do autonomous labs shorten the cycle?

The author argues that the country that leads in self-driving laboratory technology will get AI into physical products fastest. The author says Washington understands the stakes: the U.S. Department of Energy’s Genesis Mission names AI-driven autonomous labs among its core challenges. Radical AI signed a collaboration agreement with the mission in December 2025.

Most corporate research and development still follows a one-experiment, one-result, one-decision workflow from 30 years ago, the author says. The proposed alternative links reading, making and testing. AI reviews literature, proposes a material, makes it, tests it, learns from the result and tries again, around the clock.

Radical AI built that laboratory in New York. In one hypersonics campaign, the company says its self-driving lab discovered a material that outperformed the industry’s best. The lab has generated 70 million experimental data points, according to Radical AI, a record no human team could build by hand.

The author says that data matters more than anything else the lab does. A materials model must learn from real experiments conducted in the physical world.

What could faster materials discovery unlock?

The author lists chips that run cooler while delivering more power, turbine blades that last twice as long and reactor walls that withstand the heat of fusion. None of those products exists yet, according to the commentary. Their materials still face the same 10-year timeline, even as AI tools become ready to build them.

Companies already treat AI spending as a core investment, with dedicated budgets, owners and deadlines, the author says. Their prescription for materials research and development is the same treatment: a timeline to cut, not a cost center to manage. The tools to do that exist now.

A rival country or foreign competitor that cuts the decade tax first could leave American companies licensing the materials their own AI boom depends on. In that scenario, trillions spent on data centers would build another economy’s industrial lead. The author calls that outcome avoidable.

The next contest is over who turns intelligence into physical things fastest. As the author puts it, “The AI race will be won by whoever turns intelligence into physical things fastest.”

Source: fortune.com

Share this article:

More from Amara Osei

Amara Osei

Show full bio

Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.

2,248 articles

Related articles

« Previous article