U.S. Think Tank: U.S. AI Initially Hit Power Bottleneck, Then Manufacturing Bottleneck, Now Material Bottleneck

¬ AI Supply Chain: U.S. Dependence on China for Yttrium

The future of the AI race will hinge on control over obscure but high-performance materials like yttrium—essential building blocks for artificial intelligence.

Beyond demands for water, electricity, and minerals, the biggest consequence of AI infrastructure development is the competition for gas turbines. Data centers, especially ultra-large-scale AI data centers, require massive power loads that many local grids cannot meet, forcing developers to build large-scale natural gas power plants in competition for computing resources. In the U.S., natural gas production reached a record high in 2025, yet turbine delivery lead times now stretch from five to seven years. It began as a power bottleneck, evolved into a manufacturing bottleneck, and has ultimately become a material bottleneck.

The problem lies in the machines themselves. Modern turbines rely on material systems capable of withstanding extreme thermal stress, and yttrium is a key element in the coating structures that enable this. Yttrium is also used to produce yttria-stabilized zirconia (YSZ), a ceramic thermal barrier coating that protects underlying metals from high-temperature damage. In fact, NASA notes that YSZ has become the dominant formulation for these applications, used across all new aircraft and ground-based power turbine engines. This critical role reveals a potential industrial logic: when a low-volume material occupies a performance-critical layer, with few substitutes and limited supply diversity, it becomes a primary bottleneck.

Today’s demand for exascale computing is intensifying pressure on yttrium, as it is also a key component in high-temperature superconductors. In March 2026, companies including Microsoft and Chevron announced large-scale natural gas power projects to support new data center developments. The growing demand for AI directly translates into more turbine orders, more advanced coatings, and greater strain on the complex internal material systems within turbines.

China’s Yttrium Is the Chokepoint in U.S. AI and Defense Supply Chains

This supply chain is a matter of national security. Even by rare earth standards, yttrium’s market concentration is exceptionally high. According to the U.S. Geological Survey, 93% of yttrium compound imports between 2020 and 2023 came from China. This is not a broad, liquid commodity market—it is a narrow mineral chokepoint, with mining, processing, refining, and distribution controlled by a single nation.

Beijing implemented export controls in April 2025. This move caused shipments of yttrium products to the U.S. to plummet from 333 tons to just 17 tons within eight months. For commercial turbine manufacturers like GE Vernova, which already have tens of billions of dollars in backlogged orders, this poses an existential threat to production timelines. Ongoing shortages of yttrium could delay future orders and put high-value contracts for existing data centers and utility projects at risk—threatening key sectors of U.S. advanced manufacturing.

This vulnerability directly impacts the defense industrial base and military readiness. The engines and certain components of F-35 and F-22 fighter jets, as well as turbines in U.S. Navy destroyers, all rely on the same yttrium-based coatings. A disruption in yttrium supply would undermine maintenance support and military readiness. Each engine overhaul requires inspection of turbine blades and reapplication of YSZ powder. Shortages of this material would create severe bottlenecks at maintenance facilities, grounding fighter jets and extending repair cycles at naval shipyards by months. The operational readiness of frontline U.S. forces would be directly compromised—not by combat, but by a rival nation choosing to withhold a little-known mineral.

Reaching the Physical Limits of Digital Power

This reflects an inherent structural fragility within the infrastructure of the AI economy.

While there are theoretical ways to reduce such risks, addressing them takes years—not months. Although the Defense Production Act (DPA) has been invoked to boost domestic rare earth projects, and companies like ReElement Technologies plan to begin producing yttria, increasing upstream output alone is insufficient. The real challenge lies downstream. Certifying a new material for performance-critical applications like jet engine turbine blades is a complex, multi-year process requiring proof of durability under thousands of hours of extreme stress. Supply chain resilience isn’t just about opening a new mine. It’s about building a fully functional end-to-end industrial ecosystem—from mines to processors to coating manufacturers—that can absorb geological pressures.

Political shocks no longer hinder critical production. The U.S. no longer possesses this capability.

The broader lesson is simple: the competition for computing power isn’t just about who can fund the next data center or pre-order the next turbine installation. It’s about who controls those obscure, high-performance materials—without which turbines simply cannot operate. The power struggle in AI may appear digital on the surface, but in reality, it depends on ceramics, coatings, and a direct pipeline to competitors’ mineral supplies.

Source: The National Interest

Authors: Morgan Bazilian, Jahaara Matisek, and McDonald Amoah

Date: April 9, Washington Time

Original: toutiao.com/article/1862015083192521/

Disclaimer: The views expressed in this article are those of the authors alone.