The third-generation semiconductor market — led by silicon carbide (SiC) and gallium nitride (GaN) — finds itself at a crossroads as the twin forces of electrification and AI compute reshape demand. While both materials promise efficiency gains in power electronics, structural differences in their properties and a legacy of overinvestment mean SiC makers face persistent overcapacity that surging AI data-center demand alone cannot resolve.
Physical traits decide applications. SiC’s high thermal conductivity and strong breakdown field make it the natural choice for high-voltage, high-power applications above 1,200 V, notably electric-vehicle main inverters and 800V architectures. GaN, with superior electron mobility and MHz-level switching, excels at medium- and low-voltage, high-frequency roles such as 48V/12V power supplies and fast chargers. Industry participants increasingly treat the two as complementary rather than interchangeable.
Automotive remains the central battleground. The global automotive-grade power-semiconductor market is forecast to exceed $20 billion by 2026, with SiC projected to hold more than 70% of that segment. Replacing silicon-based IGBTs with SiC MOSFETs in drive inverters can improve efficiency by roughly 5–8%, extending vehicle range by about 5–10% while halving system volume.
Tesla’s high-volume adoption of SiC in the Model 3 catalyzed follow-on deployments by automakers worldwide; within 800V platform models, SiC main drives are approaching standard status. Meanwhile, GaN is carving out roles in 48V mild-hybrid systems and low-voltage auxiliary supplies, leveraging high switching frequencies to shrink magnetic components and boost efficiency — albeit while contending with automotive reliability challenges such as surge immunity and dynamic on-resistance.
AI data centers offer a promising but limited incremental market. As per-rack power pushes well beyond 100 kW, efficiency shortfalls in silicon power supplies create openings for wide-bandgap devices. GaN-based LLC resonant converters can lift server PSU efficiency above 98% and raise power density, and SiC can enhance 800V HVDC rectification and PFC efficiency to over 97%. Some hyperscalers and GPU-heavy designs have adopted SiC and GaN in new deployments.
Nonetheless, industry analyses describe SiC usage in AI infrastructure as a “niche within a niche”: SiC devices typically become economical only at higher supply powers (above roughly 3,000 W), and a large share of data-center PSUs under 5,500 W still rely on silicon. Projections put global SiC device revenue at around $10 billion by 2030, with revenues tied to AI expected to be a marginal slice — estimates suggest single-digit-hundred-million-dollar scales today — insufficient to absorb current excess capacity.
That overcapacity is acute. Rapid capacity expansion over recent years pushed substrate prices down sharply and left utilization rates around 50% in some upstream segments. Industry players such as Wolfspeed and Rohm have reported production shortfalls and heavy impairment charges: Wolfspeed’s 8-inch substrate line has struggled to hit full throughput since commissioning and posted sizeable losses, while Rohm recorded substantial asset write-downs on its SiC investment. Temporary stock rallies tied to optimistic reports have not altered the underlying fact that volume growth from AI centers will take time to materialize at scale.
Electric vehicles, not AI, are positioned to be the principal remedy. Substrate-demand forecasts indicate EV-related SiC substrate demand could reach multiple millions of units by 2030 — several times greater than projected demand from AI data centers. Current global penetration of SiC in EV inverters remains under 20%, and 800V platform adoption is still expanding, leaving significant runway for uptake. For many manufacturers, surviving the present cycle depends on squeezing more demand out of the EV market and driving down costs via larger-diameter wafers, epitaxy improvements and yield gains.
China’s domestic players are pursuing capacity and cost plays. Firms in China are scaling 8-inch SiC substrate production and promoting substitution for imported parts; some are allocating major fundraising proceeds to capacity builds now in the hope of winning when demand rebounds. In GaN, domestic and global vendors alike are pushing 8-inch heteroepitaxy processes aimed at lowering unit costs and accelerating adoption in data-center and consumer fast-charging segments.
In short, AI creates attractive new use cases for SiC and GaN, but it is not an immediate panacea for industry-wide overcapacity. The next phase of the cycle will likely reward companies that simultaneously expand automotive penetration, achieve cost reductions through technological and scale improvements, and manage capacity rationalization to restore healthier utilization and pricing.
