The 2026 power semiconductor market is not in a single oversupply state; it is structurally split, with 8-inch legacy MOSFET/IGBT lines tight, AI-grade SiC modules short, and automotive-grade SiC facing price pressure after rapid Chinese capacity expansion [S2][S3].
Industrial sourcing teams should plan for 26–52 week lead times on branded 600–1200 V discretes, 20–30+ week waits on PMICs and VRMs tied to 48V/800V AI rack builds, and double-digit price increases announced by at least twenty vendors in the first half of 2026 [S1][S3][S5].
Why "oversupply" is the wrong frame for 2026
Aggregate oversupply headlines mostly describe commodity logic, DRAM, and NAND, where cloud digestion has flipped the cycle since 2023; the binding constraint for industrial buyers sits one layer down, on 8-inch analog/power wafers and discrete power lines, where 90% utilization was reported in mid-2026 by TrendForce and a third round of foundry price hikes was already being staged for late 2026 and early 2027 [S3][S4].
Foundry prices for 8-inch wafers rose 5–15% in H1 2026, while 12-inch mature-node prices for 28/40/65 nm processes were signaled 5–10% higher in Q2–Q3 2026 with order visibility stretching into 2027, which is the opposite of an oversupplied capacity base [S4]. A separate Aug 2026 source notes the global top-five power-semi capex declined for two straight years and only rebounded roughly 11% in 2026, with new fab output taking 2–3 years to ramp, structurally protecting the incumbent allocation [S5].
The "SiC paradox": automotive glut meets data-center deficit
Silicon carbide is the clearest case where "oversupply" is true in one vertical and false in another: after the 2022–2024 Chinese SiC capacity build-out, automotive-grade SiC is now under price pressure, especially in xEV traction inverters, while 800V/1200V SiC for AI server power stages remains under-allocated, which is why the market is described as a paradox rather than a glut [S1][S2][S5].
Per-currency parity between SiC and IGBT has narrowed: SiC MOSFET single-device cost sits at roughly 1.3× an equivalent IGBT, and the 800V platform share of new-energy vehicles reached 15% in Q1 2026 (up 21% year-on-year) per TrendForce, with global SiC main-inverter penetration at 29% of NEV inverter shipments, or 1.17 million units [S5]. That demand absorption is exactly why automotive pricing has eased even as the discrete power bind in 600–1700 V modules persists [S3].
Where the squeeze actually bites: discrete power and 8-inch analog
Three categories carry the residual risk: 600–1700 V IGBT/SiC discretes, 8-inch analog ICs (op-amps, ADC/DAC, isolated gate drivers, LDOs), and 40/65/90 nm automotive-grade MCUs, while commodity logic, DRAM, and NAND sit in oversupply and are not gating industrial OEM throughput [S3].
On the discrete side, branded parts in ≥100 A modules and 1200 V SiC carry 26–52 week lead times; automotive-grade 8-inch analog runs 20–40 weeks with last-time-buy notices already issued on several legacy op-amps, and second-source qualification typically takes 6–12 months; second-tier MCU lines still allocate to tier-1s even after the 2023–2024 normalization [S3]. Wholesale spot liquidity does exist, but at 2–5× contract on Asian B2B platforms, with single-piece MOQ on commodity SOT-363 discretes around US$0.14 and bulk 5000-piece BT136E-series TRIACs at roughly US$0.052 (July 2026 listings) [S3].
AI racks are the new demand floor
AI rack power density has moved past 100 kW per rack in 2026, with first-generation deployments around 200 kW, current generation near 500 kW, and one- to two-year roadmaps toward 1 MW, which multiplies the per-server power-semi content from under US$10 in legacy servers to roughly US$200 today [S1][S5].
The device-count math behind that dollar figure: an 800V HVDC AI server draws on more than 24 SiC devices, around 40 GaN devices, and 128 silicon driver chips, versus a handful of discretes in a 48V/54V legacy box, and the Nvidia Vera Rubin platform is targeted to complete 800V HVDC component stockpiling by Q3 2026 with broader rollout behind the Rubin Ultra launch in H2 2027 [S5]. Infineon projects its fiscal 2027 AI data-center power-solutions revenue at about €2.5 billion versus roughly €1.5 billion in fiscal 2026, which is the kind of demand pull that keeps 8-inch lines allocated to AI rather than released back to industrial [S5].
Vendor price action through July 2026
At least twenty Chinese and international power-semi vendors issued price-increase notices in H1 2026, and a second-round wave concentrated around 2026-07-01 lifted many lists to 15% or more [S5]. China Resources Micro (Huarun Micro) raised full-line MOSFET/IGBT prices 10%+ in February, with a follow-on to 15%+ on July 1; Sien Microelectronics moved all product lines up 10% in March and to 15%+ on July 1; Yangjie Technology raised full lines 10–15% effective July 1, 2026; and StarPower Semiconductor raised IGBT and SiC MOSFET modules 15%+ [S5].
Internationally, Infineon triggered two 2026 increases, the first in Q1 covering power switches and related chips at 5–15% effective April 1, and a second notice on May 26 effective July 1, while Texas Instruments executed its fourth price hike since 2025 on the same July 1 date, and an Anhui power-semi plant is operating two shifts with a 4–5 month backlog while lead times have stretched from a baseline 8–12 weeks to over 30 weeks [S5][S4].
Wafer supply and the 8-inch bind
Global 8-inch capacity is forecast to fall another 2.4% in 2026, with the contraction extending at least into H1 2027, and TSMC and Samsung continuing to exit 8-inch mature nodes while Chinese foundries shift some capacity toward memory, which is the structural reason 8-inch tightness has outlasted the consumer-electronics squeeze [S5].
The long-cycle math is the same one that defined 2020–2021, except the bottleneck node is different: every kW of EV charger, servo drive, welding inverter, and power supply front end needs 600 V+ switching devices, and the wafer input to those devices is now the structurally tightest layer of the supply chain, which is why a power distribution cabinet design cannot be assumed to source the same IGBT part number at the same price it did in 2024 [S3].
How to read the market: comparison frame
The clearest decision matrix for a sourcing engineer is vertical-by-vertical: (a) automotive-grade SiC MOSFET in xEV main inverters: structural oversupply risk, 15–25% spot-discount pressure, AVL expansion is low urgency [S2][S5]; (b) 800V/1200V SiC for AI data-center power stages: structurally short, 30+ week leads, dual-source qualification advised [S1][S5]; (c) 600–1200 V IGBT modules for industrial drives and power tool inverters: still allocated, 26–52 weeks on branded parts, 5–15% list-price increases already booked [S3][S5]; (d) 8-inch analog ICs (op-amp, ADC, isolated driver, LDO): 20–40 week leads, several LTB notices issued, second-source qualification 6–12 months [S3].
The single largest practical lever is AVL expansion: secure pin-to-pin compatible alternates for any part that touches an AI-driven power path, and treat any sub-20-week quote on a 600 V+ discrete or 8-inch analog IC as a yellow flag rather than a market-clearing signal [S1][S3]. Related coverage of the wider IC squeeze, including ADI/analog line extensions, is tracked in analog and PMIC lead times through 2026, and the AI build-out is the demand side of the same equation driving data-center chiller allocations in data center chiller stock vs new capacity orders.
Limits, failure modes, and what to track next
Two failure modes dominate a bad 2026 forecast: assuming "SiC oversupply" applies to your data-center build (it does not), and assuming 8-inch tightness will ease by year-end (the announced Q4 2026 / Q1 2027 third-round foundry hike argues otherwise) [S2][S4].
Track three concrete signals over the next two quarters: (1) TrendForce's 8-inch utilization reading for Q4 2026 against the 90% mid-2026 baseline [S4]; (2) any revised Infineon AI data-center revenue print versus the stated €1.5B (FY26) and €2.5B (FY27) trajectory [S5]; (3) the Q3 2026 Vera Rubin 800V HDC component stockpile completion and the first published power meter and rack-PDU Bill-of-Materials changes that follow, since 800V native metering is a downstream indicator of how fast the rack architecture is actually turning over [S1][S5].