Gallium: The Inelastic Byproduct Powering AI Semiconductors (GaN, GaAs & InP)

Gallium enables the compound semiconductors critical to the AI buildout. Gallium nitride (GaN) power devices deliver the high-efficiency, high-density switching needed for data-center power delivery and fast EV charging, cutting losses and thermal load in AI-scale racks. Gallium Arsenide (GaAs) supports high-frequency radio frequency (RF) in 5G/6G and defense systems /Yole/, while Indium Phosphide (InP) photonics (often incorporating gallium-containing layers) provides the low-power, high-bandwidth optical interconnects required to scale massive GPU clusters without prohibitive latency or energy penalties.

Semiconductor-grade gallium must reach 6N–7N purity (99.9999–99.99999%) through multi-stage refining, zone refining, fractional crystallization, distillation, and emerging plasma-chemical methods to support defect-free MOCVD epitaxial growth using trimethylgallium precursors /ScienceDirect/. Nearly all primary low-purity gallium (~848t in 2024, ~900t in 2025) is a byproduct of bauxite-to-alumina or zinc processing, with China supplying 98–99% of global output. The United States has had zero primary production since 1987 /USGS/.

Expanding supply beyond current byproduct streams faces steep technical and economic hurdles. Gallium concentrations in bauxite are low (~50 ppm), and recovery depends on installing specialized ion-exchange or solvent-extraction circuits at large alumina or zinc plants /Minerals Engineering/, capacity that is unevenly distributed and concentrated in China. Current byproduct recovery already offers some headroom (potentially several times present output by optimizing cut-off grades and recovery efficiency without massively increasing aluminum or zinc production), but pushing further requires either scaling host-metal output (risking price crashes in those larger markets) or shifting to lower-grade residues, red mud, coal ash, or dedicated processes whose marginal costs can rise dramatically. Emerging solutions such as improved chelating resins /Ecolab/, plasma purification, and scaled new-scrap recycling aim to flatten this cost curve and reduce waste, yet deployment is slowed by high capex, process integration challenges, and limited non-Chinese expertise in extraction technology.

Demand is accelerating rapidly. AI data centers and high-performance computing drive GaN adoption for efficient power conversion, while exploding needs for co-packaged optics and high-speed networking favor InP and related photonics. A promising 2025 breakthrough from Fraunhofer ISE shows that high-quality InP layers can now be grown on larger, more scalable GaAs substrates (InP-on-GaAs), potentially cutting production costs for InP-based photonics by up to 80% while easing indium supply constraints for AI optical interconnects /Fraunhofer ISE/. Broader compound-semiconductor markets are projected to grow at double-digit CAGRs, with GaN power devices expanding especially fast in EVs, renewables, and telecom infrastructure.

Geopolitical concentration adds urgency. China’s near-monopoly on primary gallium, combined with export controls and past bans (partially suspended into 2026), creates supply-chain vulnerability for Western chipmakers /CSIS/. Key non-Chinese players include Japanese refiners (Sumitomo, Dowa), Canadian recyclers (Neo Performance Materials), European specialists (Umicore for precursors, Freiberger for GaAs substrates), and downstream leaders such as Wolfspeed (GaN), Qorvo (GaAs), and photonics companies. Promising diversification includes new scrap-recovery facilities (e.g., MTM Texas targeting early 2026) /PR Newswire/, alumina-plant integrations in Australia, and restarts in Kazakhstan and elsewhere, but these remain small relative to Chinese capacity. Accelerating allied recycling, co-located recovery, and strategic stockpiling will be essential to secure the gallium flows that underpin AI power and photonics infrastructure.