§ 01Executive View
Navitas's competitive position is fragile-but-optionally-real, and the cohort thesis depends on a single thing holding: that "density" is a durable architectural advantage rather than a window that closes the moment Infineon and TI scale 300mm GaN. On the evidence assembled here, that window is real — Navitas demonstrated the first credible single-stage 800V-to-6V GaN-IC reference design and is one of fourteen named NVIDIA 800 VDC silicon partners — but it is closing fast. Three things have already gone the wrong way in twelve months: (i) TSMC announced exit from GaN foundry by July 2027, forcing Navitas to relocate its entire production base mid-cycle to PSMC and GlobalFoundries; (ii) Innoscience took the #1 GaN power-device share at ~30% to Navitas's ~17% on the back of 8-inch in-house manufacturing; (iii) revenue collapsed from $83.3M (2024) to $45.9M (2025) as the consumer-charger franchise rolled off before AI data-center revenue scaled. The bull case is that AI data center pulls Navitas through the trough on the back of NVIDIA-validated single-stage architecture wins. The bear case is that an IDM (Infineon, TI) on 300mm with internal capacity wins the high-volume sockets while Navitas — fabless, mid-transition, sub-scale — takes the architecturally interesting but commercially smaller wins. Conviction is 2 (lean long) because the asymmetry exists but the floor is real.
§ 02Competitive Set
Direct GaN competitors
| Vendor | Wafer cadence | Integration depth | Mix (auto / DC / consumer) | IP / position |
|---|---|---|---|---|
| Infineon (IFX.DE) | 12-inch (300mm) GaN-on-Si customer samples shipped Q4 2025 — first to 300mm; Kulim 200mm capacity | IDM (full vertical); €830M GaN Systems acquisition (Oct 2023) for IP, datacenter design wins, and HEMT process | Datacenter (€2.5B target by 2027), industrial, automotive — datacenter is now lead | GaN Systems' substrate IP plus Infineon's mature 600/650V HEMT portfolio; the deepest patent estate in the West |
| Texas Instruments (TXN) | 200mm production at Dallas + Aizu (4× capacity Oct 2024 expansion); 300mm pilot complete | IDM, vertical integration target 95%+ internal by 2030; full PMIC + GaN + analog stack | Data center (full 800V architecture unveiled Mar 2026 with NVIDIA), industrial, auto | 800V-to-6V bus converter at 97.6% peak, >2000W/in³ — directly competing on Navitas's flagship spec |
| Innoscience (2577.HK) | 8-inch (200mm) at scale — only high-volume 200mm GaN player today; ~80% chip-yield uplift, ~30–40% cost reduction vs 6-inch | IDM (Suzhou + Zhuhai); fabless model rejected | Consumer (mass), data center (sole Chinese partner in NVIDIA 800V), automotive | #1 GaN power-device share at ~30% (Yole 2024); won ITC and PTAB rulings vs Infineon and EPC in 2025 |
| Power Integrations (POWI) | 6-inch internal PowiGaN; only vendor with 1250V and 1700V GaN in volume | Fabless GaN process IP, deep magnetics + controller IP | Power supplies, AI data center 800V (NVIDIA-listed); historically AC/DC charger/SMPS | InnoSwitch franchise (1B+ units shipped) is the deepest reference-design lock-in in low/mid-power AC/DC |
| EPC (private) | 6-inch GaN-on-Si | Fabless | Lidar, motors, low-voltage data-center 48V→1V — discrete focus | Foundational eGaN IP; lost a key '294 patent to Innoscience PTAB invalidation Mar 2025 |
| ROHM (6963.T) | 6/8-inch GaN process; SiC-stronger | IDM | Auto and industrial primarily | Mostly a SiC story; GaN portfolio is real but not differentiated |
| Renesas (6723.T) | 6-inch via Transphorm acquisition (Sep 2024) | IDM | Auto, industrial; "GaN LLC DCX" 800V→48V transition stage | Transphorm IP (cascode 650V); now pushing into the same data-center transition slot Navitas occupies |
| onsemi (ON) | 200mm GaN process inherited via Nexperia/internal; small share | IDM | Auto-led, expanding to data center | GaN is secondary to its SiC franchise; included in NVIDIA 800V partner list |
| STM, AOS, MPS, Richtek | Various (mostly 6-inch, fabless) | Mixed | AC/DC, supply-side | Listed in NVIDIA 800V partner ecosystem but not lead architectural participants |
The honest read: the corpus distillation "Infineon scale, TI vertical, Navitas density" is understated competition by half. The actual GaN power-device top-3 by 2024 share (Yole) is Innoscience > Navitas > EPC, with Infineon at #4 and Power Integrations at #5 (per Yole / 36Kr / Bank of America 2025 data). Innoscience is the structural threat the user's notes don't flag — they have the only mature 200mm in-house line, hold their own in the patent disputes, and are the sole Chinese partner on NVIDIA's 800V list.
Adjacent / Substitute (for the conversion socket NVTS targets)
- Vicor (VICR) — 800V→48V high-density Bus Converter Modules (BCMs); occupies the transition-stage two-stage architecture that Navitas's single-stage is trying to obsolete. If Vicor's two-stage stays good enough at 96–97% efficiency, the single-stage architectural advantage is incremental, not category-defining.
- Monolithic Power Systems (MPWR) — last-mm VRM; currently dominates the 48V/12V→0.7V GPU board socket. NVTS does not displace MPWR — they are downstream from each other in the conversion chain. But if 800V-to-6V single-stage works, MPWR's TAM at the intermediate-rail layer compresses.
- Silicon (Si) FETs in 650V class — still cheaper. The substrate trend is GaN, but at lower power densities silicon wins on $/A.
- Two-stage GaN architectures from Infineon/TI/Power Integrations — every direct competitor sells the two-stage answer (800V→48V→6V) in volume today with proven datacenter PSUs. Navitas's single-stage is the bet that the architecture compresses; if it doesn't, NVTS sells two-stage with the rest of the field at #2-or-lower share.
Emergent threats
- TSMC GaN exit (July 2027) — Navitas's lead foundry partner is leaving the business by end-July 2027. Navitas has signed PSMC (200mm, volume H1 2026) for 100V, then 650V transitioning over 12-24 months; and GlobalFoundries (Burlington, VT — production "later 2026") which licensed TSMC's 80V/650V process. This is not a customer-replacement event; it is a process-portability event. Every yield, reliability, and binning curve has to be re-qualified. Customer confidence in the supply roadmap during this window is the single largest near-term risk to design-win conversion.
- Innoscience 8-inch in-house — already in production; Navitas's PSMC 200mm line starts H1 2026; GlobalFoundries production "later in 2026." Innoscience has a 12-24 month head start on the wafer-economics step Navitas needs to match.
- NVIDIA partner-list dilution — NVIDIA's published 800 VDC silicon partner list contains fourteen vendors: AOS, ADI, EPC, Infineon, Innoscience, MPS, Navitas, onsemi, Power Integrations, Renesas, Richtek, ROHM, STMicroelectronics, TI. "NVIDIA partner" is not a moat — it is a participation badge. The competitive question is which of the fourteen wins meaningful socket share, not which is on the list.
§ 03Moat Assessment
| Moat type | Score | Why |
|---|---|---|
| Cost advantage | 1 / 5 | Fabless, mid-transition between TSMC (exiting) and PSMC/GlobalFoundries (ramping). Innoscience has 8-inch in-house already; Infineon shipped 300mm samples Q4 2025; TI has 200mm at Aizu+Dallas (4× expansion) plus 300mm pilot complete. Navitas is structurally the highest-cost of the four scale players for the 2026-2027 window. |
| Switching costs | 2 / 5 | GaN-IC integration (driver + FET co-packaged) means a NVTS reference design is genuinely sticky once locked in — the customer doesn't have a discrete-FET drop-in replacement. But for greenfield 800V data-center designs, the IC is being chosen now, so this is forward-protective only after design wins land. |
| Network effects | 1 / 5 | None. Power-semi is spec-driven, multi-source policy from Tier-1 customers. No two-sided network. |
| Intangibles (IP, brand) | 3 / 5 | Genuine GaN-IC integration patents (driver-on-die GaN HEMT) — the "Navitas density" thesis lives here. But Innoscience won an ITC ruling against Infineon in Dec 2025 and a PTAB invalidation against EPC's '294 patent in Mar 2025; the GaN patent landscape is being actively reshuffled. Navitas's specific IP has not yet been ruled invalid by a competitor challenge. |
| Efficient scale | 1 / 5 | The opposite of efficient scale. The market explicitly favors larger players because customers want vertically-integrated supply security at hyperscaler scale. |
Verdict: narrow · Trend: stable-to-eroding.
The moat is real but narrow and concentrated entirely in the GaN-IC architectural-integration vector. The trend is not eroding on the technology axis — Navitas's single-stage 800V-to-6V demo is genuinely first and the IP estate is intact — but it is eroding on the supply-chain axis: TSMC exit forces a 12-24 month process-portability transition during the exact window Infineon and TI are scaling 300mm. The cohort thesis ("density beats scale") survives a one-year Infineon scale lead. It does not survive both a 300mm Infineon volume ramp and a delayed PSMC/GlobalFoundries transition. Navitas needs the PSMC 100V product live in H1 2026, GlobalFoundries 650V in production by late 2026, and at least one Kyber-class 800V reference-design lock-in before Rubin Ultra ships in 2027. That is three executions in eighteen months for a $46M revenue company.
The deeper read: this is not a "moat" investment — it is a "the architecture wins" investment. The bet is that single-stage 800V-to-6V becomes the de facto rack PSU pattern, that NVIDIA's reference design pulls the rest of the market with it, and that customer reference-design lock-in is established faster than Infineon/TI can match the architecture on their own silicon. That is a reasonable bet. It is not a moat.
§ 04Share Trajectory
GaN power device share, 2024 (Yole Group / Bank of America / TechInvestments compilation):
- Innoscience: ~30% (up from low single-digits in 2021)
- Navitas: ~17% (down from ~16% in 2023; Power Integrations narrowly behind)
- Power Integrations: ~17%
- EPC: ~12-15%
- Infineon: ~10% (post-GaN Systems integration)
For NVTS specifically, "share" is misleading because the historical share is dominated by a low-power consumer/charger franchise that is intentionally shrinking — mobile fell to <25% of revenue in Q4 2025 from majority a year prior. The relevant share metric for the cohort thesis is AI data-center 800V design-win share among the 14 NVIDIA-listed silicon partners — and that data does not yet exist in any reliable source. The OCP 2026 / Computex 2026 / GTC 2026 design announcements through H1 2026 will be the first real read.
What is observable directly:
- Revenue collapsed from $83.3M (FY24) to $45.9M (FY25), a -45% decline as the consumer franchise rolled off ahead of the AI data-center ramp.
- Q1 2026 revenue grew +58% YoY (per Power GaN IC fabless industry report) on early high-power orders.
- Q4 2025 was the first quarter in company history where high-power markets were the majority of revenue — an inflection on mix, not yet on absolute scale.
Sources: Yole Power GaN 2025 report; Bank of America 2025 GaN device share commentary cited by Bamboo Works; Navitas Q4 2025 earnings release; TechInvestments substack compilation.
§ 05Pricing Power
- Has the company raised prices in the last 24 months? No clean evidence either way — Navitas has not disclosed ASP trends. The product mix-shift from consumer to high-power data center is mathematically a positive ASP move (a 100V GaN FET for an AI rack PSU is materially more dollar-content per die than a charger eGaN). But a mix-shift is not pricing power; it is portfolio rotation.
- Did volumes hold or grow at the new price? No. Total revenue fell -45% in 2025. Mobile/consumer volumes collapsed; high-power volumes are early-ramp. The volume curve is a U-bottom in 2025, with the inflection slope through 2026 entirely dependent on data-center design-win conversion.
- Customer concentration leverage: Navitas does not break out customer concentration in detail, but management commentary makes clear that NVIDIA's 800 VDC reference architecture is the single anchor customer-architecture for the high-power pivot. NVIDIA has thirteen other named silicon partners on the same architecture. NVIDIA holds all the negotiating leverage. This is the same pattern that compressed Wolfspeed's pricing power: a fragmented supplier base in front of a concentrated customer base.
Pricing-power score: 1 / 5. There is no demonstrated pricing power; what looks like ASP improvement is mix shift. NVIDIA holds the architecture and the partner shortlist; Navitas competes for inclusion, not for price.
§ 06Bull Points
- Architectural first-mover on single-stage 800V-to-6V. The corpus is right that this is "where the industry wants to go" (per "The AI Power Crisis — Part 2"). Navitas demonstrated the GaNFast 800V-to-6V power delivery board ahead of competitors, with the explicit value proposition of replacing the 48V intermediate bus. If the architecture wins, the reference-design lock-in is real for the design-cycle generation it ships in.
- NVIDIA 800V silicon partner status is confirmed. Navitas was named as one of fourteen NVIDIA 800 VDC silicon partners (Oct 2025 OCP Summit), with both GaN (GaNFast) and SiC (GeneSiC) products integrated into the Kyber rack-class architecture targeting Rubin Ultra (2027 ship). The validation is real even if it is not exclusive.
- GeneSiC adds optionality. The 5th-gen GeneSiC TAP architecture (35% efficiency improvement, AI-server targeted) gives Navitas SiC content in the same 800V architecture where its GaN content sits. SiC datacenter revenue at competitor onsemi was $250M+ in 2025 — the SiC datacenter TAM is real and Navitas has a credible-if-small participation. Pro-forma 2025 SiC contribution: ~$145M (per industry reports), an order of magnitude larger than Navitas's standalone 2025 revenue.
- Balance sheet repaired. $236.9M cash at end-Q4 2025 (vs $150.6M at Q3 end), with $95.6M private placement Nov 2025. This buys roughly 8-12 quarters of runway at current burn even if data-center revenue ramp is slower than guided. Bankruptcy risk is off the table for the cohort window.
- GlobalFoundries Burlington partnership = US-domiciled supply. With CHIPS Act and onshore-manufacturing politics, a US-fabricated GaN supply line (vs Chinese Innoscience or Taiwan PSMC alone) is a real differentiator for hyperscaler procurement.
§ 07Bear Points
- Innoscience is taking share, not Navitas. #1 in GaN power-device share at ~30% with the only at-scale 200mm in-house line and a winning ITC patent posture. The user's "three-way race" frame (Infineon/TI/Navitas) silently omits the actual top-share Chinese player.
- TSMC GaN exit forces a process-portability transition during the data-center pull. PSMC 100V production starts H1 2026 (months from now); 650V transitions over 12-24 months. GlobalFoundries production starts "later 2026." Customer qualification cycles for new fabs are 6-18 months. Navitas could be qualifying the supply chain for NVIDIA Kyber while Infineon is shipping 300mm parts.
- The 300mm GaN cost-curve scenario from synthesis is real, and Navitas is on the wrong side of it. Infineon shipped first 300mm GaN customer samples in Q4 2025; TI's 300mm pilot is complete; Intel demonstrated 12-inch GaN-on-TRSOI Q4 2024; IMEC has launched a 300mm GaN program. Navitas's announced roadmap is 200mm at PSMC/GF, not 300mm. If 300mm volume economics arrive at Infineon by 2027-2028, the 30-50% wafer cost-out becomes a structural moat advantage for IDMs that fabless 200mm players cannot match.
- Customer concentration risk is the inverse of what NVTS press releases imply. "NVIDIA-validated" looks like a moat. With 14 silicon partners on the same NVIDIA list, validation is the entry condition, not the differentiator. NVIDIA wields all the leverage in the partnership economics.
- Revenue collapse before inflection. -45% revenue YoY in 2025 with negative GAAP gross margin (-17.2% Q4'25). Non-GAAP operating loss $12.1M Q4. The "U-bottom" is plausible but the right side of the U requires design-win conversion in a competitive field with all 14 NVIDIA partners chasing the same sockets.
§ 08Conviction (1–5)
2 / 5 (lean long). Higher than 1 because the architectural first-mover claim is real and validated by NVIDIA 800V participation, the balance sheet is fixed, and the SiC optionality is genuine. Lower than 3 because the supply-chain transition risk is severe (TSMC exit is a category-1 disruption, not a footnote), the share leadership is held by Innoscience not Navitas, and the cohort's load-bearing thesis ("density beats scale") explicitly depends on a 300mm cost-collapse scenario going slowly — which is not the way Infineon and TI are guiding.
This is materially below the Wolfspeed -2 conviction (5/5 short on competitive grounds) by symmetry: the WOLF case is overwhelmingly clean evidence in one direction; the NVTS case has real evidence in both directions. The single most important fact for the PM: the corpus's three-way race framing materially understates competition. The actual top-share GaN player is Innoscience, and the actual binding near-term risk is the TSMC-PSMC-GlobalFoundries supply transition through 2026-2027.
§ 09Key Risks to This Read
- What would invalidate the long lean: (a) Innoscience taking the high-volume NVIDIA Kyber sockets in 2026-2027 on the strength of its 8-inch in-house cost curve before Navitas's PSMC/GF lines hit volume; (b) Infineon's 300mm GaN ramp arriving 12-18 months earlier than guided and being adopted at hyperscaler scale via the existing Infineon datacenter customer base; (c) any single PSMC or GF qualification slip pushing Navitas's GaN supply scarcity into the Rubin Ultra design-in window.
- What would invalidate the short lean: (a) NVIDIA explicitly anchoring single-stage 800V-to-6V architecture as the reference Kyber path with Navitas as the named-architecture lead; (b) a 2026 data-center revenue ramp that gets Navitas to a $50M+ AI-DC quarterly run-rate without Innoscience or Infineon disclosure of equivalent share gain; (c) a strategic acquirer (Infineon, TI, ON, or a hyperscaler) at a premium — the GeneSiC SiC business plus the GaN-IC IP estate is plausible M&A.
- My implicit assumptions: (i) the corpus's distillation that "the architecture wants to go single-stage" is correct, but (ii) the architecture-first-mover advantage compresses to 12-24 months once IDMs catch up, and (iii) the supply-chain transition (TSMC→PSMC/GF) is the binding near-term constraint, not the architectural debate. If (i) is wrong (single-stage stays niche), the whole NVTS thesis is incremental. If (ii) is wrong (IC integration is genuinely defensible long-term), NVTS could re-rate sharply. If (iii) is wrong (TSMC stays in long enough or PSMC ramps cleaner than expected), the bear case weakens.
Works cited
- Mark Lapedus substack — 5 Reasons Why TSMC Is Exiting the GaN Market
- Independent analysis of TSMC GaN exit rationale
- TSMC reallocating GaN capacity to higher-margin AI logic
- NVIDIA Developer Blog — Building the 800 VDC Ecosystem for Efficient, Scalable AI Factories (silicon-partner list)
- NVIDIA 800 VDC silicon-partner list contains 14 vendors: AOS, ADI, EPC, Infineon, Innoscience, MPS, Navitas, onsemi, Power Integrations, Renesas, Richtek, ROHM, ST, TI
- Demonstrates that 'NVIDIA 800V partner' status is participation, not exclusivity — undercuts the implicit moat framing in NVTS press releases
- Lists box-builder partners (ABB, Eaton, GE Vernova, Heron, Hitachi, Mitsubishi, Schneider, Siemens, Vertiv) and component partners (Bizlink, Delta, Flex, Lead Wealth, LITEON, Megmeet)
- Power Electronics News — APEC 2025 GaN vs SiC competitive boundary
- 650V is the GaN/SiC overlap zone — both technologies contestable; relevant for NVTS's GaN-IC vs GeneSiC SiC product-line strategy
- GaN encroaches up from low/mid voltage; SiC retains 1200V+ — places NVTS density bet specifically in mid-voltage AI-DC zone where Power Integrations and Innoscience also compete
- Cross-referenced from WOLF/sources.json id=19
- Semiconductor Today — Yole Power GaN device market 42% CAGR to $3bn by 2030 (Oct 2025)
- GaN revenue of $920M expected in 2026, up 58% on 2025 — calibrates the 2026 ramp slope
- Confirms 42% CAGR baseline and reinforces Yole 2025 sizing as the cohort's anchor TAM source
- TrendForce 2024 GaN power-device market share data (via 36kr summary, 2025)
- 2024 global GaN power-device share: Innoscience 29.9%, Navitas 16.5%, EPC 12.4%, Infineon 10.3%, Power Integrations 9.8%
- Top-5 concentration ~85% — basis for HHI calculation (~1,490, moderately concentrated)
- Top-3 share ~57% — used in market-structure table
- + 1 more
- TrendForce — GlobalFoundries Moves on GaN: TSMC and Navitas Ties Position U.S. as New GaN Production Hub (Nov 27, 2025)
- Navitas held roughly half of TSMC's GaN wafer output prior to the exit announcement — primary independent corroboration of NVTS-TSMC concentration
- TrendForce confirms Navitas multi-foundry strategy: TSMC (legacy), PSMC (200mm Taiwan), GlobalFoundries (200mm Vermont)
- Article does not name X-Fab, suggesting X-Fab is SiC-only (not part of GaN strategy)
- TrendForce — NVIDIA Picks Innoscience as Sole Chinese Supplier for 800 VDC Power
- Innoscience confirmed as sole Chinese partner in NVIDIA 800V silicon list
- Innoscience runs 8-inch (200mm) in-house GaN at scale — has 12-24 month wafer-economics head start over Navitas's PSMC ramp
- Yole Group — 'Power GaN 2025' / 'From chargers to data centers' press release (Oct 2025)
- Power GaN device market $355M (2024) growing to ~$3B (2030) at 42% CAGR — primary external TAM anchor
- Application split by 2030: consumer/mobile >50%, automotive 73% CAGR, telecom/datacenter 53% CAGR to ~$380M
- First commercial 800V HVDC rollouts anticipated 2027 — corroborates NVTS's stated 2027 inflection
- + 1 more
- Yole Group — Power SiC overcapacity downturn until 2027–2028; $10bn by 2030 (Dec 2025)
- Power SiC device market projected to reach $10.3B by 2030 at ~20% CAGR
- SiC market in correction cycle through 2027–2028 due to upstream overcapacity + automotive softness — directly relevant to NVTS GeneSiC near-term cycle position
- Data center cumulative SiC opportunity ~$200M over next 5 years
- + 1 more
- Bamboo Works — Innoscience makes gains in patent dispute, as growing competition remains bigger threat
- Bank of America 2025 GaN device share commentary: Innoscience ~30%, Navitas ~17% (#2)
- Confirms NVTS is not the share leader — corpus three-way race framing is incomplete without Innoscience as #1
- Digitimes — TSMC GaN exit prompts Navitas to shift orders to PSMC for Nvidia AI servers
- Navitas-PSMC 200mm partnership; 100V volume H1 2026; 650V transitioning over 12-24 months
- Direct evidence of supply-chain transition timing risk vs Rubin Ultra 2027 ship
- Electronic Design — GlobalFoundries Partners with Navitas to Ramp Up Next Generation of GaN Power ICs
- GF licensed TSMC's 80V/650V GaN process; development work starts early 2026, production later 2026 in Burlington, VT
- US-domiciled supply optionality differentiator vs Innoscience and Taiwan-only PSMC
- Infineon — Completes acquisition of GaN Systems
- Infineon acquired GaN Systems for $830M, closed Oct 2023
- Inherited GaN Systems' IP estate, datacenter design wins, and 600/650V HEMT portfolio
- Establishes Infineon as the deepest GaN patent holder among Western IDMs
- Semiconductor Today — Power Integrations 1250V/1700V PowiGaN for 800VDC AI datacenters
- GaN now extending to 1250V and 1700V (formerly SiC-only territory) — direct adjacency disruption into NVTS's high-voltage GaN-IC roadmap
- AI datacenter 800V slot increasingly contested by competing GaN suppliers, not just NVTS density bet
- Cross-referenced from WOLF/sources.json id=20
- Semiconductor Today — TSMC to cease GaN foundry production by end-July 2027
- TSMC GaN foundry exit confirmed end-July 2027 due to Chinese pricing pressure
- Forces Navitas mid-cycle process-portability transition during AI datacenter pull window
- Most material near-term competitive risk to NVTS roadmap on the supply-chain axis
- Semiconductor Today — TSMC to cease GaN foundry production by end-July 2027 (Jul 3, 2025)
- TSMC GaN foundry exit by end-July 2027 confirmed via independent industry trade press
- Cited rationale: pricing pressure from Chinese GaN rivals (Innoscience and others)
- Most-cited supply-chain risk for NVTS in deep-dive analyses across cohort
- Stocktitan / Power Semiconductors Weekly — Navitas Q4 2025 strategic shift summary
- Confirms FY25 SiC pro-forma revenue contribution ~$145M (combining GeneSiC line)
- GaN/SiC mix reshape underway — supports the 'right tier of bifurcating market' market-structure stance
- NVTS positioned as one of NVIDIA-named 800V partners for 2027 production
- TI — Quadruples internal GaN manufacturing capacity (Aizu + Dallas)
- TI 4× internal GaN capacity expansion (Oct 2024)
- 200mm production at both Dallas and Aizu Japan; 300mm pilot complete
- Demonstrates TI vertical-integration depth materially exceeds Navitas fabless model
- TI — Unveils Complete 800 VDC Power Architecture for AI Data Centers with NVIDIA (March 2026)
- TI 800V-to-6V bus converter at 97.6% peak efficiency, >2000W/in³ density
- TI directly competing on Navitas's flagship single-stage spec with vertical-integration cost structure
- 30 kW 800V AC/DC PSU and 800V capacitor bank product extensions
- Tom's Hardware / DCD on GB200 NVL72 rack pricing and configuration
- GB200 NVL36 ~$1.8M, NVL72 ~$3M list per HSBC estimates — baseline for content-per-rack triangulation
- GB200 NVL72 ~120 kW power consumption — current-state rack power baseline
- Used to anchor the 'today vs 2027–2028 rack content' bands in voltage-stack section (sell-side ranges, not primary tear-down — flagged as directional)
- TrendForce — GlobalFoundries-TSMC-Navitas GaN production hub (Nov 2025)
- Navitas's foundry-light model uses TSMC and now GlobalFoundries — confirms asset-light differentiation
- Foundry-fabbed GaN broadens NVTS's potential capacity without IDM capex — relevant to barrier-to-entry trend (declining for fabless tier, rising for IC integration)
- Cross-references Innoscience IDM scale advantage discussion
- TrendForce — Innoscience Scores Major Patent Win Against Infineon as ITC Rules No Infringement
- Innoscience won ITC ruling vs Infineon Dec 2025 (no infringement of two remaining patents)
- PTAB invalidation of EPC '294 patent claims (Mar 2025)
- Demonstrates GaN patent landscape is fluid; Navitas's IP estate has not yet been challenged but is not assured of indefinite protection
- NVIDIA Developer Blog — 800 V HVDC Architecture for AI Factories (May 2025)
- Infineon named as lead partner in NVIDIA 800V HVDC ecosystem
- Navitas named as ecosystem partner (lower tier than lead) — relevant for 'reference-design durability' assessment
- Full production with Kyber rack-scale 2027 — calibrates the catalyst window
- + 1 more
- Bloomberg Tax — Navitas Semi Warns of Material Weakness in Internal Controls (2024)
- Material weakness disclosed in 2024 over stock compensation and license-agreement accounting
- Q1 2024 10-Q delayed; 10-K/A amended; no restatement of prior financials required
- Establishes baseline disclosure-quality risk for FY2025/FY2026 controls assessment
- Cohort companies.json — NVTS entry (id 12)
- NVTS sentiment +1, mentionCount 4 — thinnest deep-dive corpus support after ETN
- supportingQuotes from 'The AI Power Crisis — Part 2': single-stage 800V-to-6V demos 'especially revealing... too early to say whether that approach becomes mainstream'
- Risks captured: smaller scale vs Infineon/TI, no comparable 300mm anchor, density advantage erosion if 300mm cost curve dominates, gallium supply concentration
- Cohort companies.json — NVTS entry (id=12) and EV/end-market ecosystem entries
- NVTS catalysts: single-stage 800V-to-6V design wins, OCP / Computex 2026 reference designs, Kyber/Rubin Ultra 2027 timeline
- NVTS risks: smaller scale vs IFX/TI, no comparable 300mm anchor, single-stage not yet proven mainstream, gallium critical-mineral concentration
- EV ecosystem entries (Porsche, BYD, Hyundai, Tesla, Mercedes, BMW, VW) as end-market geography proxy for NVTS GeneSiC EV revenue
- + 1 more
- Cohort corpus — 'Building a Datacenter Part II' (Crucible Capital, April 2025)
- NVTS cited in footnotes only — no main-text BOM-level partnership with Vertiv / Schneider / Delta confirmed
- Schneider 800V real revenue impact framed as 2028-2030 — primary source for the calendar-mismatch risk for NVTS
- SuperMicro DCBBS framing relevant as a future Navitas reference-design opportunity but not currently confirmed
- Cohort corpus — 'The AI Power Crisis — Part 2' (Crucible / NuttyCLD, May 2025)
- Primary source for 'Infineon scale, TI vertical integration, Navitas density' competitive structure
- 'Navitas's direct 800V-to-low-voltage demonstrations are especially revealing. It is still too early to say whether that approach becomes mainstream. But it clearly shows where the industry wants to go.'
- Vertiv / Eaton / Schneider / Delta named as the four box-builder partners at the 800V transition layer — none publicly named with Navitas BOM-level reference design in this corpus
- + 1 more
- Cohort cross-reference AVGO/customer.md
- Comparison template for what a 'real' multi-year hyperscaler customer relationship looks like — NRE + per-unit royalty + 24-36 month tape-out cycles + foundry-pass-through
- Used to calibrate that the NVIDIA-Navitas 'collaboration' has none of those contractual features yet
- Demonstrates that concentration on dramatically stickier customers is the goal NVTS is aspiring to but has not achieved
- Cohort cross-reference NVDA/customer.md
- Hyperscaler customer frame — top-2 NVDA customers each >10% revenue; 'Customer A/B/C/D' alphabet-coded disclosure
- Rack-as-product BOM structure — board to Supermicro/Quanta/Foxconn to hyperscaler — relevant for understanding where NVTS could be designed in (BOM level vs. ecosystem partner level)
- Pull-through vs channel-fill / pre-buy framework
- Cohort cross-reference WOLF/customer.md
- EV-OEM auto qualification 12-24 month cycle benchmark applicable to NVTS GeneSiC entry
- Wolfspeed LTSA roster (GM, Mercedes, JLR, Lucid, Toyota OBC, BorgWarner) is the comparison set NVTS GeneSiC has not matched
- '$5.8B design-in pipeline' framing flagged as suspect-RPO template — directly applicable to NVTS '$2B+ design-in pipeline' management framing
- + 1 more
- Cohort cross-reference — TSM/regulatory.md (Section 232 derivative-product expansion, Taiwan-tail)
- Section 232 scope-expansion framework directly applicable to GaN/SiC fabbed at TSMC and Powerchip
- Cross-strait kinetic event framing as macro-owned tail risk
- FDPR creep to power-semi adjacency framework
- Cohort cross-reference — TSM/supply-chain.md (foundry capacity frame, tier-2 chokepoints, Taiwan-tail)
- Aixtron/Veeco MOCVD lock-in framework adapted for GaN epi side
- Taiwan-tail risk frame (~92% of advanced wafer capacity inside one strait) — adapted to PSMC concentration risk for NVTS
- Pass-through power inversion logic — TSMC pricing power compares to Navitas's squeezed-middle position
- + 1 more
- Cohort cross-reference — WOLF/regulatory.md (CHIPS direct grant template, §48D mechanics)
- Lutnick Investment Accelerator template (Intel equity-conversion model) as realistic CHIPS direct-grant outcome path
- Confirms power-semi BIS exposure framing — civil industrial uses outside Section 232 scope at issuance
- Section 301/232 directional asymmetry framework adapted to NVTS GaN context
- Cohort cross-reference — WOLF/supply-chain.md (SiC substrate market dynamics, Aixtron lock-in)
- Confirms SiC substrate ASP collapse mid-teens 2024 — tailwind for GeneSiC unit economics as non-IDM consumer
- Wolfspeed share fall 60%+ → ~34% (2021-2024); Coherent (II-VI), SK Siltron CSS, ROHM, TanKeBlue, SICC as the merchant alternatives
- Aixtron G10-SiC / Planetary lock-in case study — directly transposable to GaN epi tools at PSMC and GF Burlington
- + 1 more
- Cohort sibling — NVDA/macro.md (Taiwan-tail probability and AI-capex framing reused)
- Taiwan kinetic-event probability ~2-4%/yr, blockade ~5-8%/yr over 2026-2030 — applied to NVTS production base
- Hyperscaler operating-cash-flow funding model for AI capex (insulating from credit cycle) — applies to NVTS AI-DC revenue ramp
- USD/TWD wafer-cost dynamic — applies to NVTS TSMC + PSMC GaN cost base
- + 1 more
- Cohort sibling — NVTS/competitor.md (competitor-analyst, 2026-05-04)
- Innoscience #1 GaN power-device share ~30% Yole 2024 vs Navitas ~17% — frames China-domestic-build headwind
- TSMC GaN foundry exit by July 2027; PSMC + GlobalFoundries transition — frames Taiwan-tail concentration timeline
- Innoscience as sole Chinese partner on NVIDIA 800V list — frames asymmetric AI-DC competitive risk
- + 1 more
- Cohort sibling — NVTS/customer.md (customer-analyst, 2026-05-04)
- End-market mix FY25 estimates: mobile 60-70%, AI-DC 5-8%, EV 10-15%, solar 10-12%, industrial 5-8% — basis for blended cycle position table
- Two distributors >10% of revenue, ~17% top customer concentration — frames revenue-side FX exposure via Asian distributors
- NVIDIA collaboration is non-binding May 2025 — caveat on AI-DC ramp timing
- + 1 more
- Cohort sibling — NVTS/financial.md (financial-analyst, 2026-05-04)
- 83x EV/Sales valuation framing as anchor for terminal-multiple sensitivity to rates
- FY25 revenue $46M (down 45%); $237M cash post Nov-2025 PIPE; ~7-yr runway at $8M/quarter burn
- $202M of common stock issuance FY25 — recursive equity-funding-cost channel of rate sensitivity
- + 2 more
- Cohort sibling — NVTS/market.md (market-positioning-analyst, 2026-05-04)
- Yole Power GaN 2025: $355M (2024) → ~$3B (2030) at 42% CAGR; data-center slice ~$380M by 2030 at 53% CAGR
- SiC market correction through 2027-2028 (Yole) — frames NVTS GeneSiC headwind in 2026 cycle table
- Datacenter GaN at Y0 of volume ramp — frames calendar-mismatch between 2026-tactical-loser and 2027+-structural-winner
- + 1 more
- Cohort sibling — WOLF/competitor.md (SiC competitive frame Navitas inherits via GeneSiC)
- SiC device share 2024 (TrendForce): STM ~33%, onsemi ~25%, Infineon ~15%, Wolfspeed ~11%, Rohm + others balance
- Navitas/GeneSiC sits in 'others' — no top-5 SiC device share in any 2024 data; SiC is supplemental optionality not a primary moat
- Wolfspeed substrate share collapse 60%+→34% (2021-2024) frames how fast structural transitions can punish single-product power-semi pure-plays
- Cohort sibling — WOLF/financial.md (cautionary-tale benchmark)
- Capex-light (NVTS $1.5M) vs capex-heavy (WOLF $1.27B FY25) contrast — NVTS structurally less catastrophic
- But: WOLF at 3.5x EV/Sales vs NVTS at 82.7x — multiple risk worse at NVTS
- Both share negative ROIC ~30%, SBC discipline issues, dependence on hyperscaler design-win timeline
- Cohort sibling — WOLF/macro.md (rate / EV-cycle / Taiwan-tail framing reused)
- Auto-loan rate environment 2023-2026 suppressing US/EU EV unit demand — applies to NVTS GeneSiC EV revenue
- Residential solar payback math rate-sensitivity — applies to NVTS Enphase / solar microinverter revenue
- SiC substrate ASP deflation under Chinese competition — applies to NVTS GeneSiC SiC product line pass-through power
- + 2 more
- Cohort synthesis (semiconductor-industry) — macro lens for NVTS
- GaN three-way competitive structure: 'Infineon scale, TI vertical, Navitas density' as anchor framing for NVTS strategic position
- Section 6 contested claim #2 — single-stage 800V-to-low-voltage 'too early to say whether mainstream' is the binding macro-uncertainty for NVTS
- Open Question #2 — GaN three-way race timing; 300mm cost curve dominance risk before density advantage matters
- + 3 more
- Cohort synthesis.md — semiconductor-industry chip-to-grid value chain (2026-05-04 refresh)
- L8b value-chain placement: 'high-density GaN conversion; the density-bet pure-play in the 800V transition layer'
- Theme #3.3 framing: 'Infineon fights with scale, TI with vertical integration, and Navitas with density'
- Theme #3.2 chip-to-grid pass-through: Vertiv Q4'25 organic orders +152% YoY, $15B backlog, 2.9x book-to-bill; Eaton Q4'25 DC orders +3x YoY
- + 4 more
- Compound Semiconductor — Innoscience files lawsuit against Infineon (Suzhou, Jan 2025)
- Counter-suit in Suzhou Intermediate People's Court (patents 202311774650.7 and 202211387983.X)
- Signals enforceability of Chinese GaN patents against foreign players — read-through to NVTS IP enforcement strategy in China
- DigiTimes — Infineon's GaN patent wall forces global firms to rethink China ties (Dec 5, 2025)
- Innoscience 30% global GaN share (2024) vs Navitas 17% — direct market-share comparison
- Patent ecosystem framed as defining factor in China-coupled supply chain decisions
- GlobalFoundries press release — GF and Navitas Partner to Accelerate U.S. GaN (Nov 20, 2025)
- GF Burlington Vermont GaN production for NVTS — development early 2026, production late 2026
- National-security framing in announcement language
- Not a CHIPS Act §9902 direct grant; sits indirectly under GF's $1.5B Malta PMT
- GlobalFoundries — Long-term Strategic Partnership with Navitas for U.S. GaN (Nov 20, 2025)
- GF licensed TSMC's 80V and 650V GaN power-semi processes
- Development with Navitas scheduled for early 2026
- Production expected later 2026 at GF Burlington, Vermont fab
- + 2 more
- Infineon press release — Infineon sues Innoscience for Patent Infringement (March 2024)
- Four GaN patent infringement claims filed at ITC and N.D. Cal.
- Patents in scope: US8686562B2, US9899481B2, US8264003B2, US9070755B2
- Establishes baseline for ITC 337-TA-1407 docket
- Infineon Technologies AG Annual Report FY2025
- IFX FY25 adj. gross margin ~43%; Power & Sensor Systems segment-result margin ~14.9%
- FY26 adj. gross margin guidance low-40s
- Anchor for segment-level (not consolidated) comp on the GaN/power axis
- Macro background — rates, FX, AI-capex, end-market cycle baselines
- US 10y ~4.0-4.5% / real rates 1.5-2.0% / Fed funds 3.75-4.25% / DXY mid-100s as current-regime baseline
- Auto-loan rate environment 2023-2026 suppressing US/EU EV unit demand 5-10% per 100bp
- Residential solar volume rate-sensitivity 10-15% per 100bp over 12-month lag
- + 4 more
- Navitas and NVIDIA Collaborate on 800 V HVDC Power Architecture (Computex 2025)
- Strategic collaboration with NVIDIA on 800V HVDC architecture announced May 21, 2025 at Computex
- Navitas positioned as ecosystem partner in NVIDIA 800V HVDC AI factory architecture
- No volume commitment, no LTA, no preferential supply terms disclosed — collaboration only
- Navitas Investor Day Presentation (September 2025)
- Five-end-market framework: mobile/consumer, AI data center, EV, solar/energy storage, motor drive/industrial
- Reference-design partner roster includes NVIDIA, Murata, Enphase publicly named
- GeneSiC SiC product-line roadmap — auto OBC and DC fast-charging design-in references
- + 1 more
- Navitas press release — 200mm GaN production with PSMC (July 1, 2025)
- Powerchip Fab 8B (Zhunan Science Park, Taiwan) qualified for NVTS 100V–650V GaN-on-Si
- 100V family targets 1H26 production; 650V transitions over 12–24 months
- Concentrates supply in Taiwan jurisdiction — Section 232 derivative-tariff exposure
- Navitas Q4 2025 / Full Year 2025 earnings press release
- FY2025 revenue $45.9M vs FY24 $83.3M (-45% YoY)
- GAAP net loss $117.0M; non-GAAP net loss $41.6M
- Q4'25 revenue $7.3M; high-power markets first majority of revenue
- + 2 more
- Navitas Q4 2025 earnings call transcript (Feb 24, 2026)
- Mobile dropped from majority Q3'25 to <25% of Q4'25 revenue; 'insignificant' by end-2026
- Q4 framed as the bottom; sequential revenue growth expected through 2026
- Operating expenses guided flat at ~$15M/quarter through 2026
- + 1 more
- Navitas Semiconductor 10-K filed Feb 27, 2026 — supply-chain disclosures via Stocktitan summary
- Risk language: 'We have historically relied on a single third-party manufacturer (wafer foundry) to fabricate our GaN products, and on a separate, single wafer foundry to fabricate our SiC products.'
- Single/limited source language extends to key materials and components
- TSMC GaN production cease by July 2027 disclosed; mitigation via Powerchip and GlobalFoundries with buffer-inventory build
- + 4 more
- Navitas Semiconductor FY2024 10-K (filed March 2025)
- Customer concentration: 2 distributors >10% of revenue in FY24 (down from 3 in FY23)
- Top distributor ~17% of revenue (down from ~22% FY23) — concentration easing reflects mobile demand decline rather than diversification
- No long-term agreements, take-or-pay, or volume guarantees disclosed in filing
- + 1 more
- Navitas Semiconductor FY2025 Form 10-K (filed Feb 27, 2026)
- Item 9A controls assessment baseline for any new or unremediated material weakness
- Item 1A risk factors — TSMC GaN exit by July 2027; PSMC and GF Burlington qualification
- Item 3 legal proceedings — confirms no active securities class action as of February 2026
- Navitas Semiconductor Q1/Q2/Q3 FY25 10-Q filings and earnings calls
- Distributor inventory normalization called out in Q1/Q2 FY25 — channel-fill reversal evidence
- AI-DC revenue framed as 'small but fastest-growing' — implies <10% of FY25 run-rate
- Mobile/consumer fast-charge ~60-70% of revenue [inferred from management commentary, not directly disclosed by percentage]
- + 2 more
- Navitas Semiconductor — GeneSiC Acquisition Close (Aug 2022)
- Acquisition closed Aug 2022 — basis for Navitas's SiC product line
- Provides EV traction-inverter, on-board-charging, DC-fast-charge SiC product entry
- Acquisition consideration ~$244M; primary footprint of post-deal SiC revenue contribution
- Navitas Semiconductor — Q4 and Full Year 2025 Financial Results (Feb 24, 2026)
- FY25 revenue $45.9M vs FY24 $83.3M — concrete revenue decline tied to mobile exit
- Q4'25 trough revenue ~$7M; designated as expected bottom
- Mobile <25% of Q4'25 revenue — confirms strategic pivot is in execution, not aspiration
- + 2 more
- Navitas — 800 VDC Power Architecture for NVIDIA AI Factory press release (Oct 2025)
- Navitas spans first stage (grid to 800VDC SST), second stage (800VDC to 54V/12V), and third stage (POL)
- 100V GaN FETs with dual-sided cooled packages mentioned for GPU power boards
- 650V GaN FET portfolio + GaNSafe ICs (integrated control/drive/sensing/protection) referenced
- + 2 more
- Navitas — AI Data Center Opportunity investor presentation (August 2025)
- $2.4B lifetime design-win pipeline disclosure — concrete near-term opportunity anchor (lifetime cumulative, not annual)
- NVIDIA 800VDC partner status confirmed — basis for reference-design momentum claim
- End-market framework: AI data center, grid/energy infrastructure, performance computing, industrial electrification, mobile/consumer/appliance (the latter being de-prioritized)
- Navitas — Plans for 200mm GaN Production with PSMC (Powerchip) press release (Jul 1, 2025)
- PSMC partnership for 200mm GaN-on-Si in Fab 8B, Zhunan Science Park, Taiwan
- 180nm CMOS-class process, voltage range 100V to 650V
- Initial qualification Q4 2025; 100V family volume production at PSMC 1H 2026
- + 2 more
- Navitas — Redefining Data Center Power: GaN and SiC for 800 VDC Infrastructure white paper (Oct 2025)
- 10 kW DC-DC platform at 98.5% peak efficiency, 1 MHz switching — anchors the density narrative
- 12 kW platform with GeneSiC + GaNSafe + Intelliweave for 500 kW rack power — positions NVTS at the IC tier of the rack power chain
- GaN wins at 800V→6V and intermediate bus stages (P2/P3 of NVIDIA HVDC architecture) — used in voltage-stack content-per-rack section
- Navitas — TSMC & Amkor Manufacturing Partnerships press release (Oct 17, 2017)
- Amkor named as packaging, test and logistics partner — 2017 baseline relationship
- Amkor provides 'high-volume and low-cost QFN packaging platform' for Navitas GaN
- TSMC named as wafer foundry — 2017 baseline GaN-on-Si relationship
- + 1 more
- Navitas-GeneSiC acquisition press release (Aug 2022)
- $100M cash + 24.9M shares + earn-out structure — basis for SiC product line acquisition
- GeneSiC TAM-statement '$20B+ aggregate market opportunity by 2026' explicitly flagged as inflated round-number sizing — used Yole instead
- Acceleration into EV / solar / energy storage markets by 2-3 years — context for SiC mix reshape
- NIST CHIPS for America program portal
- NVTS not listed among CHIPS Act Section 9902 direct funding recipients
- GlobalFoundries Malta NY $1.5B PMT (Feb 2024) covers Burlington VT GaN partnership site indirectly
- TXN $1.6B and Wolfspeed $750M PMTs cited as cohort comparators
- Semiconductor Today — TSMC to cease GaN foundry production by end-July 2027 (July 3, 2025)
- TSMC publicly announces GaN foundry exit by end of July 2027 due to Chinese price pressure
- Hard-dated supply-chain cliff for NVTS GaN volume
- PSMC and GF Burlington qualification must complete before this date
- Stanford Securities Class Action Clearinghouse — NVTS docket search (no class action filed as of May 2026)
- No 10b-5 class action against Navitas Semiconductor disclosed through May 2026
- Notable for a small-cap with prior internal-controls material weakness
- StockAnalysis.com — MPWR Statistics (peer comp)
- MPWR EV/Sales 25.85x, EV/EBITDA 89.39x
- MPWR gross margin 55.18%, operating margin 27.08%
- Closest fabless power-management high-multiple comp; even MPWR is 3x cheaper EV/Sales than NVTS despite 8x higher GM
- StockAnalysis.com — NVTS Balance Sheet
- YE'25 cash $236.86M; total debt $6.47M (net cash ~$230M)
- Goodwill $163.22M (entirely from GeneSiC, unimpaired since Aug 2022 close)
- Intangible assets $53.26M, amortizing
- + 1 more
- StockAnalysis.com — NVTS Cash Flow Statement
- FY22-FY25 OCF: -$44.5M / -$41.4M / -$58.8M / -$42.9M
- FY22-FY25 capex: -$4.6M / -$4.8M / -$6.8M / -$1.5M (fab-lite)
- FY22-FY25 FCF: -$49.1M / -$46.2M / -$65.6M / -$44.4M
- + 2 more
- StockAnalysis.com — NVTS Income Statement (annual)
- FY22 revenue $37.94M; FY23 $79.46M; FY24 $83.30M; FY25 $45.92M
- Gross margins: FY22 31.5%, FY23 39.1%, FY24 34.0%, FY25 31.0%
- Operating losses: FY22 ($123.6M), FY23 ($118.1M), FY24 ($130.7M), FY25 ($107.8M)
- + 1 more
- StockAnalysis.com — NVTS Statistics & Valuation
- Market cap $3.87B; Enterprise value $3.80B; price $16.77 (May 4, 2026)
- Shares outstanding 230.79M (up 12.7% YoY)
- EV/Sales (TTM) 82.69x; P/S 87.71x
- + 3 more
- StockAnalysis.com — ON Statistics (peer comp)
- ON EV/Sales 6.94x, EV/EBITDA 23.70x
- ON gross margin 38.32%, operating margin 17.84%, FCF margin 23.66%
- StockAnalysis.com — TXN Statistics (peer comp)
- TXN EV/Sales 14.36x, EV/EBITDA 30.55x, P/E 48.03x
- TXN gross margin 57.32%, operating margin 35.96%, FCF margin 20.18%
- TrendForce — Innoscience Scores Major Patent Win Against Infineon as ITC Rules No Infringement (Dec 5, 2025)
- ALJ initial determination favors Innoscience on remaining two patents (US 9,070,755 and US 9,899,481)
- ITC Commission final determination scheduled April 2, 2026
- Sets precedent for US patent enforceability of GaN-Systems-era patents — read-through to NVTS IP moat
- USTR Section 301 four-year review and excess-capacity investigation (March 2026)
- March 11, 2026 USTR investigation covers excess capacity in 16 economies including China
- Semiconductors and EVs explicitly within investigation scope
- Possible HTS extension to Chinese-origin GaN/SiC discrete devices — potential positive asymmetry for NVTS
- White & Case — President Trump orders narrowly targeted 25% Section 232 tariff (January 2026)
- Confirms narrow scope at issuance — power semiconductors not covered
- Identifies derivative-product expansion mechanism Commerce can use
- Frame for assessing NVTS Taiwan-fabbed import exposure
- White House Proclamation — Section 232 25% tariff on advanced semiconductors (Jan 14, 2026)
- Section 232 25% tariff effective January 15, 2026; scope narrow (H200/MI325X-class)
- Proclamation contemplates future scope expansion by Commerce — derivative-product expansion risk for GaN/SiC
- Power semiconductors outside scope at issuance; civil industrial uses exempt