Energy / Power Semiconductors · Deep dive

Wolfspeed

The US silicon-carbide pioneer that spent $6B+ building 200mm capacity for an EV wave that arrived late — filed Chapter 11 on June 30, 2025, emerged September 30, 2025 with 70% less debt and 95% of the equity in creditors' hands, and continues to miss estimates as Infineon and Chinese entrants pull ahead on cost.

at risk

A US SiC pioneer whose $6B+ capex bet was funded before the EV slowdown, whose Renesas take-or-pay unwound in bankruptcy, and which emerged from Chapter 11 into a market where Infineon has bought the biggest 200mm SiC fab in the world (Kulim, Malaysia) and Chinese entrants (SICC, TankeBlue, Sanan) are dumping 6-inch wafers below $500 — the balance-sheet fix does not fix the underutilised fab.

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Wolfspeed is the largest US pure-play SiC power-semiconductor company and, until 2024, the world’s largest merchant SiC substrate supplier by volume. It grows boules, slices wafers, deposits epi and fabricates MOSFETs and Schottky diodes at Durham, NC; Marcy, NY (Mohawk Valley Fab, opened Apr 2022 as the world’s first 200mm SiC device fab); and Siler City, NC (the John Palmour Manufacturing Center). It filed prepackaged Chapter 11 on Jun 30, 2025 and emerged 91 days later with $4.6B of debt cancelled and ~95% of new equity to Apollo-led creditors, convertibles and Renesas — legacy holders reduced to a ~5% consenting-creditors gift. Q2 FY2026 (Feb 4, 2026) missed hard: $168.5M vs $199.4M, EPS -$6.11 vs -$0.63, GAAP gross margin -46%. Short interest sat at 46% of shares on loan by late Aug 2026. Balance-sheet fix worked; utilisation problem did not.

Founding story

Cree Research was incorporated in Durham in July 1987 by John Palmour, Thomas Coleman, John Edmond, Calvin Carter, Eric Hunter and Neal Hunter — six NC State alumni from Bob Davis’s materials-science lab who had spent four years cracking single-crystal silicon carbide growth (SiC sublimates rather than melts, which is why nobody had a repeatable process before). For twenty years Cree was best known for LEDs, funding a SiC power/RF program on the side. That flipped around 2015 as Tesla put SiC in the Model 3. In March 2021 Cree sold its LED business to SMART Global for up to $300M; on October 4, 2021 it renamed to Wolfspeed and moved to NYSE as WOLF. Palmour, then CTO, died November 13, 2022.

How it works

The physics is unforgiving. High-purity SiC powder is heated to 2,300-2,500 degrees Celsius in a graphite crucible under argon; vapour recrystallises on a seed at the top over one to two weeks, growing a single-crystal 4H-SiC boule. Wafer diameter scales device yield exponentially, so the industry has pushed from 100mm to 150mm to 200mm. Boules are sliced by diamond wire and CMP-polished to a sub-nanometre finish. A doped SiC epitaxial layer — the active layer — is CVD-grown on top at ~1,600 degrees. Only then does the wafer enter a MOSFET fab: trench or planar structures patterned, ion-implanted, annealed far above silicon’s melting point, and metallised. Every step is slower and more expensive than silicon because SiC is chemically inert and mechanically hard.

Wolfspeed is vertically integrated across all of it. Building 10 in Durham feeds 200mm substrates and epi to Mohawk Valley, disclosed on Jun 24, 2024 at ~20% wafer-start utilisation and targeting ~25% by year-end — roughly where it still sits. A 200mm fab at a fraction of design load is the entire operating problem.

Product and business overview

Two segments. Power Products — SiC MOSFETs (600V, 1200V, 1700V+), Schottky diodes and integrated modules for EV traction inverters, on-board chargers, DC fast-charging, industrial drives, solar inverters and AI-datacentre PSUs; Q2 FY2026 Power revenue $118.3M, AI-datacentre revenue up ~50% sequentially. Materials Products — merchant SiC bare and epitaxial wafers, historically the highest-margin piece and what Renesas’s 2023 take-or-pay was buying; Q2 FY2026 Materials revenue $50.2M and FY2025 fell 12.3% to $343.6M, a direct read on the wafer price war. Also ships GaN-on-SiC RF devices for defence — small but one of the few areas where a US-only supply chain is a moat.

Business model and pricing

Device pricing is under NDA. The shape is a merchant fab plus long-cycle design-in: wafer contracts are long-dated take-or-pay with prepayments (Renesas’s $2B was the extreme case), and device volumes are won at automotive OEMs where each traction-inverter socket implies six to seven years of revenue. That funded Mohawk Valley — build ahead of demand, sell future capacity for cash, expect the EV curve on time. When Mercedes, Lucid, JLR and GM all pulled EV programs in 2024-2025 (Mercedes withdrew guidance; JLR pushed its electric Land Rover to late summer 2026 per EVXL and Torque News) the fab was left with fixed cost, empty capacity, and a $2B liability to a customer no longer needing wafers.

Traction over time

Sources: FY2025 10-K (Sep 2025); Q2 FY2026 8-K (Feb 4, 2026); Semiconductor Today (Jun 25, 2024). A 200mm SiC fab at 20-25% capacity is structurally loss-making at any realistic device price.

Market analysis

The SiC power-device market grew from ~$2.68B in 2024 to ~$3.1B in 2025 per Yole; Yole’s Dec 2025 outlook projects ~$10B by 2030 at ~20% CAGR, automotive-and-mobility retaining ~70% of demand. Yole also describes a correction: the 2019-2024 capex boom created upstream overcapacity, and by 2025 utilisation had fallen to ~50% on wafers and ~70% on device lines. The average hides a bimodal split — Chinese capacity priced to fill, Western capacity priced to earn a return on the last five years of capex. The 2024-2025 EV pause hit the automotive socket Wolfspeed’s 2023 model rested on; AI-datacentre is not yet large enough to fill Mohawk Valley alone.

Competitive intel

Infineon Technologies (~EUR 15B) shipped first 200mm SiC devices from Kulim, Malaysia in Feb 2025 and has committed up to EUR 5B to make Kulim the world’s largest 200mm SiC fab, supplementing captive substrate with Chinese wafers from SICC and TankeBlue. STMicroelectronics runs Catania and a Sanan JV in Chongqing targeting Q4 2025 mass production. onsemi is winning EliteSiC design-ins at VW and BMW. ROHM Semiconductor owns the Japanese-automotive channel via Denso and Mitsubishi Electric. Coherent’s Silicon Carbide LLC — the ex-II-VI substrate business — took $1B from Denso and Mitsubishi Electric on Dec 4, 2023 for 25% and long-term offtake. Chinese cohort SICC, TankeBlue and Sanan Optoelectronics together took ~40% of the SiC substrate market by 2025 (up from ~10% in 2021) and drove 6-inch wafers below $500 (Semimedia) — at or below Wolfspeed’s US cash cost. Navitas Semiconductor and other GaN players are eating the sub-1200V AI-PSU segment.

History and evolution

• 1987 — Cree Research incorporated by six NC State alumni; IPO 1993.
• Sep 2017 — Gregg Lowe (ex-TI, ex-Freescale) becomes CEO.
• Mar 2021 — Cree LED sold to SMART Global for up to $300M.
• Oct 4, 2021 — Cree renames to Wolfspeed (WOLF).
• Apr 2022 — Mohawk Valley Fab opens; Siler City announced Sep 2022.
• Nov 13, 2022 — Palmour dies; Siler City renamed in his honour.
• Jun-Jul 2023 — $1.25B Apollo senior secured term loan; Renesas prepays $2B on a 10-year wafer supply.
• Sep-Oct 2024 — 150mm Durham device fab flagged for closure; preliminary $750M CHIPS award and $750M financing from Apollo, Baupost, Fidelity, Capital Group.
• Nov 18, 2024 — Lowe fired.
• May 2025 — Feurle appointed CEO; WOLF drops 23% as the “in-court solution” is disclosed.
• Jun 30, 2025 — Chapter 11 filed (case 25-90163).
• Sep 30, 2025 — Emergence: $4.6B of debt cancelled; ~95% of new equity to creditors and Renesas.
• Feb 4, 2026 — Q2 FY2026 misses: $168.5M vs $199.4M; EPS -$6.11 vs -$0.63.

What people say

The case for. Post-emergence bulls (Castellano; Financial Content’s Sep 30, 2025 “Phoenix Rising”) rest on four points: $1.3B of cash plus 70% less debt gives ~2 years of runway even at current burn; 200mm SiC device technology is years ahead of the Chinese cohort and comparable to Infineon; Mohawk Valley is a strategic US asset with $750M of CHIPS support committed; AI-datacentre SiC grew 50% sequentially in Q2 FY2026, opening a vector independent of the EV inflection.

The complaints. Glassdoor themes: layoffs “every quarter”, voluntary reductions that gutted institutional knowledge, morale visibly low pre-Ch11, only 42% would recommend, “chronically understaffed” and “highly bureaucratic”. The Renesas write-down (~$1.7B per Renesas’s Jun 2025 disclosure) is the loudest customer complaint a semiconductor company can generate. Sell-side has cut: BofA to $15 from $25 with Underperform in early 2026; Morgan Stanley previously to $8; Canaccord to $10. Short interest at 46% of shares on loan with a 900%+ borrow fee (Options Trading Report, Aug 2026) captures the market view. The short thesis on Barchart and Foreign Policy Journal: shrinking assets, persistent losses, structural cash burn, a fab at 20-25%, -46% GAAP gross margin, and a Chinese cohort that re-priced the industry below Wolfspeed’s US cost base. The refinancing did not fix any of that.

Outlook: well positioned or at risk?

Wolfspeed is at-risk — a $6B+ capacity bet made three years too early into a market re-priced from below by Chinese wafer supply and out-scaled from above by Infineon’s Kulim commitment, and the balance-sheet fix did not fix underutilisation, unit economics or competitive position. Post-emergence Wolfspeed has ~$1.3B of cash, Mohawk Valley at ~20-25% utilisation, negative gross margins, a CEO in place less than a year, and an automotive base whose 2025-2026 orders collapsed. AI-datacentre SiC is a real second vector but too small to fill Mohawk Valley alone. The bull case requires simultaneously: EV volume snapping back with Wolfspeed winning share; AI SiC scaling faster than GaN substitution; the Chinese price war stalling on 8-inch; and CHIPS disbursement surviving milestone conditions under a new administration. Any two failing sinks the plan. The most probable good outcome for reorganised equity is a Japanese or Korean strategic acquiring at a premium — not standalone profitability.

How to attack it

The direct attack is not a startup problem. Building competitive 200mm SiC substrates and Trench MOSFETs is $2-5B of capex, ten years of process learning and an automotive-qualification cycle. Every startup that has tried it — Ascatron, GeneSiC — has ended as an acqui-hire. The sharp wedges are three. First, GaN substitution in the sub-1200V AI-datacentre socket — Wolfspeed’s fastest-growing product vector is also the one where GaN (Navitas, EPC, Innoscience) is structurally more efficient at high frequency; a well-funded GaN power-module attacker with hyperscaler design-ins can pick off Wolfspeed’s AI-PSU wins before that socket becomes material. Second, SiC module integration and thermal packaging — Wolfspeed sells devices; the real value is a fully-integrated traction-inverter module a la Vitesco, Nidec or ZF; a design house owning module-level thermal, gate-drive and packaging IP turns Wolfspeed’s devices into a bill-of-materials line item. Third, fabless-plus-Chinese-foundry — a US design house that fabs at SICC or Sanan and sells in Europe and India can undercut Wolfspeed automotive pricing by 30-40% while keeping US IP and export-control compliance.

The exploitable weaknesses are visible from outside. Cost structure: US-only footprint at a moment when the marginal SiC wafer costs under $500 in China (Semimedia, 2025). Execution: Mohawk Valley is still nowhere near 50% utilisation four years after opening (Semiconductor Today, Sep 2024). Culture: Glassdoor’s “layoffs every quarter” theme plus the loss of Palmour’s technical judgment thin the institutional knowledge base. Channel: Wolfspeed lacks Infineon or ROHM’s tier-1 automotive distribution, and the Renesas write-down proved the merchant-wafer take-or-pay model does not survive a customer cycle. Capital: a >4-year consensus path to positive gross margins is longer than the CHIPS milestone clock and longer than the GaN substitution window. Any attacker offering an on-time module at 30% lower cost this year — rather than a 2028 promise — wins.

Adjacent-segment play

The Wolfspeed capability that generalises best is not the fab. It is SiC crystal-growth and 200mm substrate know-how — the piece Palmour and the founding cohort actually invented, and the piece Chinese entrants have not fully replicated at 200mm. Three viable adjacencies outside merchant power devices. First, quantum computing — SiC hosts colour-centre qubits (silicon-vacancy, divacancy) among the most promising solid-state quantum platforms; a startup packaging high-purity semi-insulating SiC as a substrate-plus-integrated-photonics offering to PsiQuantum, Quantinuum, Xanadu and DoE labs would find a demand curve uncorrelated with EV cycles. Second, RF and defence — Wolfspeed already ships GaN-on-SiC for defence and 5G, and the export-controlled US supply chain is a moat; spun out from the merchant power arm it would trade on a much higher multiple. Third, high-temperature power electronics for aerospace and geothermal — SiC’s temperature ceiling makes it enabling for downhole electronics and aircraft electrification; a specialist selling qualified modules into oil-and-gas and eVTOL commands custom-part pricing a merchant EV inverter never will.

The adjacent-segment company already doing pieces of this is Coherent’s Silicon Carbide LLC — co-owned by Denso and Mitsubishi Electric since December 2023 — which stayed narrow on substrates plus long-term automotive offtake and, by staying narrow, avoided the fab-utilisation trap Wolfspeed fell into. The moat inside Wolfspeed is the substrate and epitaxy, not the device fab. A reorganised Wolfspeed that spun off merchant devices and doubled down on substrate plus defence plus quantum would be smaller, more expensive and structurally more defensible than the vertically integrated one that just cost creditors $4.6B.

Sources and further reading

• Wolfspeed Successfully Completes Financial Restructuring (Wolfspeed, September 30, 2025)
• Wolfspeed: 91-Day Prepack Cuts $4.6B in Chapter 11 (Elevenflo, October 2025)
• Wolfspeed Q2 FY26 Earnings Release (Wolfspeed IR, February 4, 2026)
• Earnings call transcript: Wolfspeed’s Q2 2026 results disappoint (Investing.com, February 2026)
• Wolfspeed FY2025 10-K summary (Public, September 2025)
• Renesas Announces Expected Loss on Wolfspeed RSA (Renesas, June 2025)
• Mohawk Valley 200mm SiC fab reaches 20% utilization (Semiconductor Today, June 25, 2024)
• Preliminary Terms with Wolfspeed under CHIPS (US Commerce, October 15, 2024)
• Power SiC faces overcapacity downturn until 2027-2028 (Yole via Semiconductor Today, December 18, 2025)
• China’s 150mm SiC substrate prices plunge (SemiMedia, 2025)
• Infineon vs Wolfspeed for world’s largest SiC fab (eeNews Europe, 2025)
• Wolfspeed Shares Plummet 23% on Slowing EV Demand (EVXL, May 9, 2025)
• WOLF: 47% Short Interest, 932% Borrow Fee (Options Trading Report, August 2026)
• Wolfspeed Ahead: An NC State Spinoff (NC State News, October 2022)