The AI Boom Has a Rare Earth Problem Few Investors Fully Understand

Highlights

• A 2026 Fordham University study reveals AI infrastructure could face critical bottlenecks from rare earth dependencies, with 94.4% of exposure coming from IT hardware (servers and GPUs) that rely on neodymium, praseodymium, dysprosium, and terbium for cooling systems and permanent magnets.
• Neodymium-related exposure alone could exceed $90 million per gigawatt-scale AI campus, while advanced liquid cooling systems for powerful AI chips increasingly depend on dysprosium- and terbium-enhanced magnets amid concentrated supply chains dominated by China.
• The study introduces REX and SRX metrics to measure both financial exposure and supply-chain delay risks, warning that the AI revolution's physical infrastructure creates geopolitical vulnerabilities that green AI narratives often overlook.

---

A new 2026 study, “The Material Blind Spot of the AI Revolution: Rare-Earth Dependencies Undermine Sustainable Digital Future,” by Qi He, Fordham University, alongside collaborators Rui Shan, Chunyu Qu, Yuan Tang, and Yue Zou, warns that the artificial intelligence (AI) infrastructure boom may be quietly creating a major new dependency on rare earth elements (REEs)—particularly neodymium, praseodymium, dysprosium, and terbium. Drawing on industry reports, bill-of-materials data, U.S. Department of Energy supply-chain research, and AI infrastructure cost models, the authors argue that next-generation AI data centers could become heavily exposed to fragile rare earth supply chains dominated by China. Their core warning is striking: while investors focus on AI chips, electricity demand, and data-center powerconsumption, the hidden bottleneck may actually sit inside coolingmotors, optical systems, networking hardware, and high-performance permanent magnets embedded throughout AI infrastructure.

Rare Earth Exchanges™ has reported on this topic on multiple occasions verifying this general theme.

AI Data Centers Depend on Rare Earth Magnets More Than Expected

The researchers examined gigawatt-scale AI campuses—the enormous facilities powering cloud AI models and future AI services. They foundthat roughly 94.4% of rare earth exposure comes from IT hardware itself,especially servers and accelerators such as GPUs.

Neodymium and praseodymium dominate total cost exposure because they are widely used in NdFeB permanent magnets found inside cooling fans, pumps, storage drives, and thermal-management systems. Meanwhile, heavy rare earths like dysprosium and terbium appear in smaller quantities but may pose greater strategic risk because they are difficult to substitute and face highly concentrated global supply chains.

The study estimates neodymium-related exposure alone could exceed approximately $90 million per gigawatt-scale AI campus under baseline assumptions.

Why Cooling Systems Matter More Than Investors Realize

One of the paper’s most important findings involves liquid cooling. As AI chips become more powerful, future data centers may require increasingly advanced cooling systems using pumps, circulation units, and high-performance motors. Those systems often rely on dysprosium- and terbium-enhanced magnets capable of operating under extreme thermal conditions.

In other words, the AI arms race could unintentionally increase demand for the very rare earths facing the greatest geopolitical and processing constraints.

The study introduces two new metrics:

• REX (Rare-earth Exposure): Measures financial exposure to rare earths.
• SRX (Schedule-at-Risk Exposure): Measures supply-chain bottleneck risk and project-delay vulnerability.

That distinction matters because a material may represent only a tiny percentage of total project cost while still delaying billion-dollar AI infrastructure deployments if supply becomes constrained.

Implications for Investors and Policymakers

The implications are substantial. AI expansion may increasingly intersect with:

• Chinese rare earth processing dominance
• Heavy rare earth shortages
• Magnet manufacturing bottlenecks
• Critical mineral industrial policy
• Recycling and substitution technologies

The paperalso argues that many “green AI” narratives overlook the environmental costs of rare earth extraction and refining. Greater AI efficiency does not necessarily reduce material intensity—it may simply shift where rare earth dependencies appear.

Limitations and Controversies

The study is heavily scenario-based and relies on modeled assumptions, industry estimates, and projected AI infrastructure growth forecasts. The authors acknowledge that some supply-chain “schedule risk” scores were based partly on practitioner judgment rather than large empirical datasets.

Additionally, future technological changes—including alternative cooling architectures, switched-reluctance motors, recycling breakthroughs, or reduced rare earth intensity in magnets—could materially alter the conclusions. Critics may also argue the paper overstates AI-specific risks because many of these same rare earth supply-chain pressures already exist across EVs, robotics, wind turbines, and defense technologies.

Still, the study delivers an important message: the AI revolution is not purely digital. Beneath the software and cloud services lies a highly physical industrial system increasingly tied to rare earths, advanced magnets, and geopolitically sensitive supply chains.

Source: Qi He (Fordham University), Rui Shan (The Chinese University of Hong Kong), Chunyu Qu (Fordham University; Dun& Bradstreet), Yuan Tang (Carnegie Mellon University), and Yue Zou(Independent Researcher), The Material Blind Spot of the AI Revolution: Rare-Earth Dependencies Undermine Sustainable Digital Future, January 30, 2026.