For most of the past two years, the silicon carbide (SiC) substrate business was described with a single word: oversupply. Capacity announced during the 2022–2023 shortage arrived almost simultaneously, demand growth in electric vehicles slowed relative to the ambitious forecasts that justified that capacity, and 6-inch substrate prices fell below USD 500 in mid-2024 — close to the production cost line for many manufacturers. The correction was real, and it reset the cost expectations of every device maker in the value chain.

2026 is where the story turns. Across substrate, epitaxy, and device segments, the industry is reporting firmer pricing, rising utilization at leading fabs, and a demand mix that no longer depends primarily on automotive. For engineering and procurement teams, the practical question is no longer whether to qualify 8-inch SiC, but how quickly the transition can be completed before the next capacity wave arrives.

1. How the industry got here: two years of price erosion

The 2024–2025 downturn was a classic capacity-cycle correction rather than a failure of the technology. SiC MOSFETs continued to displace silicon IGBTs in traction inverters, onboard chargers, photovoltaic inverters, and fast-charging infrastructure throughout the period. What changed was the supply side: aggressive capacity additions, particularly in China, pushed 6-inch substrate pricing down by roughly 30% in a single year, according to CIC consulting data cited across industry coverage.

By late 2025, the correction had produced an unintended side effect. Inventory across the chain — substrate, epitaxy, and device — had been drawn down to unusually low levels as buyers deferred purchases in anticipation of further price declines. Because SiC crystal growth and epitaxy are inherently slow processes, that thin inventory buffer left the industry poorly positioned when demand began to recover.

2. Three demand engines behind the 2026 turn

2.1 AI data center power: the fastest-growing and least price-sensitive segment

The most consequential shift is that SiC demand is now being pulled by AI infrastructure rather than by vehicle production. Rack-level power in AI deployments has moved from single-digit kilowatts toward tens of kilowatts, and power supply units (PSUs) above roughly 5.5 kW increasingly require SiC to meet efficiency and power-density targets. Industry estimates put AI-power-related SiC revenue at around USD 300 million in 2026, rising to USD 500–600 million in 2027.

What makes this segment strategically important is its buying behavior. Hyperscale data center operators and their power suppliers prioritize efficiency, reliability, and delivery assurance over unit price, which means they absorb capacity that would otherwise serve price-sensitive industrial and automotive customers. The result is a market where pricing power concentrates in AI-adjacent supply, while automotive pricing remains constrained.

2.2 Automotive: SiC moving from flagship platforms into volume vehicles

Automotive remains the volume backbone of the SiC industry, but the growth mechanism has changed. Penetration in premium 800 V platforms is already established; the next phase of growth depends on whether SiC can reach mid-range vehicle platforms, where the cost sensitivity is far higher. That, in turn, depends almost entirely on 8-inch economics: the larger diameter delivers roughly 1.8 times the usable area of a 6-inch wafer, and the cost-per-die reduction is what makes mainstream adoption arithmetically viable.

2.3 Energy storage, solar, and grid infrastructure

A third engine is emerging in stationary energy. In 1500 V string inverters and 2000 V storage power conversion systems (PCS), low-resistance SiC devices enable efficiencies above 99% with reduced thermal derating, allowing equipment volume to be cut substantially. Multiple analyses point to 2026 as the year SiC penetration in storage PCS approaches roughly 10%, with solar, grid, and industrial drives forming a demand base that is less cyclical than consumer-linked segments.

Fig. 1 — SiC epitaxial wafer surface. As device voltage ratings rise, epitaxial thickness, doping uniformity, and defect density become the binding constraint on yield.

3. The numbers: what actually changes at 8 inch

The transition from 6-inch to 8-inch is not simply a larger version of the same product. Larger diameters change the economics of every downstream step, but they also relocate the technical bottleneck — from crystal growth toward epitaxial uniformity and from device design toward wafer-level defect control.

4. Where the bottleneck moves next: epitaxy

For 650 V devices, the epitaxial layer is thin and the process is well understood. The difficulty concentrates at the high-voltage end. A 1200 V device typically requires a drift layer on the order of tens of micrometres, and 3.3 kV devices require substantially thicker layers, with doping concentration and thickness uniformity controlled tightly across the wafer. On 8-inch substrates, maintaining that uniformity while managing wafer bow, thermal gradients, and growth rate is the central process challenge.

Defect behavior matters just as much as thickness. Basal plane dislocations in the substrate can propagate into the epitaxial layer and convert into stacking faults under bipolar operation, driving the forward-voltage drift that has historically limited the long-term reliability of 4H-SiC bipolar devices. Surface defects — including carrot defects, triangular defects, and particle-induced downfall — become proportionally more damaging as die counts per wafer rise, because the probability that a given die intersects a defect increases with die size and count.

Fig. 2 — SiC epitaxy. Uniformity of thickness and doping across an 8-inch wafer is now the primary differentiator between suppliers.

The practical implication is that epitaxy specification has become a strategic purchasing decision. Buyers who specify only thickness and doping, without defining within-wafer uniformity, defect density limits, and verification method, typically discover the gap during device yield ramp rather than during supplier qualification.

5. What buyers should lock in before 2027

The current tightening is occurring against a known counterweight: substantial new substrate and epitaxy capacity is scheduled to come online through 2027. Several analyses expect the market to shift back toward oversupply once that capacity is qualified and released. That does not argue against the transition to 8-inch, but it does shape procurement timing.

• Secure 8-inch allocation early, but structure it flexibly. Long-term agreements signed in 2026 protect against allocation risk; volume commitments that extend deep into 2027 without price adjustment mechanisms expose buyers to the next downcycle.
• Qualify substrate and epiwafer together. Define thickness, doping concentration, within-wafer and wafer-to-wafer uniformity, and defect density limits in one specification, and require batch-level certificates of analysis with defect mapping.
• Keep a qualified 6-inch fallback. Dual-diameter sourcing preserves negotiating position and protects production continuity if 8-inch yield ramp slips at any supplier.
• Specify the verification method, not just the number. Doping concentration measured by mercury-probe C-V, thickness by FTIR, and defect density by candidate wafer inspection can produce materially different results; aligning on method prevents disputes at incoming inspection.
• Plan for reliability validation, not only electrical screening. For bipolar device architectures, request substrate and epi data relevant to basal plane dislocation density and stacking fault behavior, since these dominate long-term forward-voltage stability.

6. Risks on the horizon

Three downside scenarios deserve attention. First, the demand pull from AI power infrastructure is large in value but still modest in wafer volume; if hyperscale capital spending decelerates, the highest-margin portion of SiC demand would soften quickly. Second, 8-inch yield ramp has historically taken longer than announced schedules at every node transition in this industry. Third, capacity announcements are not the same as qualified output — the gap between the two has repeatedly produced both shortage and glut within a single cycle.

None of these risks reverse the direction of travel. The physics and the arithmetic both favor larger diameters, and the 2028 crossover point is now a question of execution rather than of debate. What they do change is the value of specification discipline: in a market this cyclical, the buyers who fare best are the ones who can move volume between suppliers and diameters without requalifying their entire process.

7. How PWG supports 8-inch and 6-inch SiC programs

Power Wafertech Group supplies SiC substrates and epitaxial wafers to research institutions and device manufacturers, with specifications defined against the application rather than against a fixed catalog. Typical support includes substrate selection across diameters and doping types, epitaxial structure design for target breakdown voltage, and documentation packages covering thickness, doping uniformity, surface defect density, and wafer geometry. For programs validating the move from 6-inch to 8-inch, comparative sample sets can be prepared so that uniformity and defect behavior are evaluated under identical device process conditions.

Data sources

• SEMI Silicon Manufacturers Group (SMG), quarterly silicon and compound semiconductor shipment statistics.
• TrendForce, 2026 global SiC power device market report (capacity digestion and cost-driven transition).
• CIC consulting data on 2024 SiC wafer pricing, as cited in industry coverage of the price correction.
• Hangjia Say Research, forecast on 8-inch SiC wafer shipments overtaking 6-inch by 2028.
• Public disclosures and investor communications from SiC substrate, epitaxy, and device manufacturers through 2026, including capacity and utilization commentary.
• Figures quoted on AI data center power demand for SiC are industry estimates and vary by research house; they should be treated as directional rather than precise.