Introduction
Artificial intelligence is attracting historic levels of investment, yet most of that money remains locked inside software and data centers. The physical products that would actually put AI to work—chips, turbines, reactors, and defense systems—depend on materials that can take a decade and up to $100 million to move from concept to factory floor. This gap, often called the decade tax, is why the AI boom has not yet reached the broader economy.
What Happened
Research from IBM suggests that discovering a new material can require ten years and as much as $100 million, whether the goal is a jet-engine alloy, a chip coating, or a reactor lining. Meanwhile, Morgan Stanley projects that hyperscalers and AI cloud providers will spend nearly $1.6 trillion in 2027, while Accenture’s global technology chief has said that essentially no large companies have been truly reinvented by AI yet. Software has accelerated; the physical world has not.
The author, a former Army research lab scientist and co-founder of Radical AI, argues that the bottleneck is not the AI models themselves but the materials underlying everything they run on. The decade tax exists because discovery still relies on slow trial and error, the scientific literature has grown too vast for any one person to master, and lab equipment was never designed for automation. In one striking comparison, the entire high-entropy alloy field studied roughly 3,500 compositions over 22 years, while a self-driving lab identified 300 candidates in just 16 weeks.
Why This Matters
Every major AI promise eventually becomes a materials problem. Leading-edge semiconductors are bumping against atomic-scale limits, and critical inputs such as hafnium have seen prices jump nearly tenfold in four years. Data centers require enormous power, and the solutions being counted on—better batteries, a stronger grid, and fusion energy—depend on materials that can survive extreme heat and stress, many of which do not yet exist. Defense systems face similar constraints, with hypersonic vehicles and next-generation engines still relying on materials designed decades ago.
Supply risks add urgency. Temporary restrictions on rare earth exports from China are currently paused, but that truce is set to expire in January. If controls return, chipmakers and defense contractors could find themselves without ready substitutes because no backup materials have been developed. The hidden cost is even larger: when new materials take a decade, engineers simply design around older ones, and the products that never get built never appear on any balance sheet.
Government attention is growing. The Department of Energy’s Genesis Mission has identified AI-driven autonomous labs as a core challenge, and Radical AI signed a collaboration agreement in December 2025 to support that effort. Still, most corporate research and development operates much as it did thirty years ago: a cycle of single experiments, isolated results, and sequential decisions.
Key Takeaways
- Discovering a new material can take ten years and up to $100 million, creating a “decade tax” on physical innovation.
- AI infrastructure spending is projected to reach nearly $1.6 trillion in 2027, yet few large companies have been fundamentally transformed by AI.
- Slow discovery is driven by manual experimentation, overwhelming scientific literature, and non-automated lab equipment.
- Self-driving labs can dramatically accelerate materials research, testing hundreds of compositions in weeks rather than years.
- Critical materials such as hafnium and rare earths face price spikes and supply risks that could stall AI hardware and defense programs.
- Cutting the decade tax requires treating materials R&D as a core investment with timelines and accountability, not just a cost center.
Conclusion
The AI race will be won by whichever country and companies turn intelligence into physical products fastest. Self-driving laboratories that read research, propose materials, run experiments, and learn continuously already exist, and one New York facility has generated 70 million experimental data points while identifying a hypersonic material that outperforms current industry standards. Without similar speed in materials discovery, American firms risk licensing the very substances their AI boom depends on from foreign competitors.
The tools to shorten the decade tax are available now. Treating materials R&D with the same urgency as AI software investment could determine whether the trillions flowing into data centers build a lasting industrial advantage or simply enrich another economy. The choice is to cut the tax or keep paying it.

Discussion
Join the conversation
Thoughtful reactions, questions, and follow-up ideas help shape the next story.