AI Industry Landscape

GB200

$10.5

NVL72 superchip · frontier

↓ hyperscalers ease floor

B300 Ultra

$6.00

Blackwell Ultra · allocation-gated

↓ thin, wide dispersion

B200 SXM

$3.80

180GB HBM3e · demand-led

↑ firm, only card tightening

H200 SXM

$2.50

141GB HBM3e · peaked $7.50

↓ eased below $3

MI300X

$1.50

192GB · AMD value alt

↓ ~30%, AMD glut

H100 SXM

$1.70

80GB HBM3 · the anchor

↓ soft, still training king

A100 80G

$0.68

80GB SXM4 · legacy glut

↓ abundant, no scarcity

RTX 5090

$0.40

32GB · new consumer

↓ ~33%, inventory broadens

RTX 4090

$0.32

24GB · budget

↓ flat, abundant

H200 SXM spot $/hr · 3M

Supply Signal: Two-Speed Market Hardens

The "surplus everywhere" story is over — the market has bifurcated. Prior-gen and commodity silicon keep drifting down: H200 has eased below $3 (~$2.50), MI300X fell ~30% to ~$1.50 as AMD floods aggressive spot bundles, A100 sits in an outright legacy glut (~$0.68), and consumer RTX 5090 dropped ~33% to ~$0.40 as inventory broadens. Hyperscalers pile on — AWS cut its own H100/H200/A100 up to 45%. But the newest Blackwell is firm-to-tightening: B200 is the one card that actually rose into late June (~$3.80, demand-led), while GB200 (~$10.5) and B300 (~$6) stay allocation-gated with huge price dispersion — a thin, immature spot market, not real liquidity. H100 is the anchor: soft on spot but still the training workhorse. The structural constraint is now HBM supply, not chip output — which is why Blackwell won't cheapen quickly even as the low end gluts. For Bitcoin miners pivoting to AI: the hotel-model $150-500K/MW assumption keeps compressing, and only Blackwell-class capacity holds pricing; providers are competing on reliability and dev experience, not just $/hr.

Prices are spot market, 8-GPU cluster, $/hr per GPU. Reserved/contract rates are 30-50% lower.

Live data: getdeploying.com · vast.ai · sfcompute.com

SanDisk Corporation (SNDK)

$2,064

+857% YTD · #1 in S&P 500

$306B

Market Cap

~11x

Fwd P/E (Q4 ann.)

78.4%

Q3 Gross Margin

Strong Buy

Consensus (15B / 4H / 0S)

$8.0B

Q4 FY26 Guided Rev.

Quarterly Flash Revenue & Gross Margin — WD Flash segment (est.) → SanDisk standalone

Q1–Q4 '22 through Q4 '24: WD Flash segment estimates (~48% of total WD revenue). Actual flash revenue known for Q2'23 ($1.66B) and Q1'24 ($1.56B).
Q1–Q4 '26: SanDisk standalone (Q1–Q3 actual, Q4 = guided midpoint $8B). FY25 spinoff transition period not shown.

Q2 FY26 — Revenue by Segment

Datacenter $440M · Edge $1,700M · Consumer $907M

SNDK Standalone — Quarterly Progression

Q3 FY26: Datacenter up 233% QoQ. Q4 FY26 guided $7.75–8.25B.

Global NAND Supply — by Company (EB/year)

Dashed = total demand. Colored area ABOVE dashed = oversupply → price crash. Dashed ABOVE colored area = demand exceeds supply → pricing power. 2024–2027: shortage window. 2028+: new fabs tip balance.

Global NAND Demand — by End Market (EB/year)

Dashed = total supply. AI/HPC uses bull-case trajectory (~50%/yr growth from hyperscaler capex). 2024–2027: demand exceeds supply = the shortage narrative. Supply catches up ~2028 → next cycle risk.

Fab Timeline — Kioxia / SanDisk 50/50 JV (Japan)

Under construction

Operational

Ramping

All fabs operated as 50/50 JV — SanDisk receives 50% of output. Yokkaichi (Mie) = original campus since 1993. Kitakami (Iwate) = new AI-era campus, K2 online Sep 2025.

Yokkaichi — Mie Prefecture

Operational

Fab 1–2

Est. 1993/1995. Originally DRAM, converted to NAND 1999. Fab2 rebuilt 2016 for 3D NAND. Mature nodes, cost-optimized.

Operational

Fab 3–4

Opened 2005/2007. First 300mm wafer fabs at Yokkaichi. High output efficiency. Now producing BiCS3/4 3D NAND.

Operational

Fab 5

Phase 1: 2011 (19nm planar NAND). Phase 2: 2014. Converted to 3D BiCS FLASH. Japan government subsidy recipient 2022.

Operational

Fab 6

Opened 2018. Dedicated 3D NAND from day one. BiCS5 (112L) and BiCS6 (162L). Primary volume fab for current generation.

Operational

Fab 7

Announced Oct 2020, operational 2022. CBA (CMOS Bonded to Array) technology. 218-layer BiCS8. Gov. subsidy up to ¥92.9B approved 2022.

Kitakami — Iwate Prefecture

Operational

K1 / Fab 1

Ground broken 2022. Initially paused during 2023 NAND bust. Resumed and operational ~2024. BiCS7/8 technology. New campus, purpose-built for AI era.

Ramping

K2 / Fab 2

Operations began Sep 29, 2025. Produces 218-layer CBA (CMOS Bonded to Array) — most advanced NAND architecture. Gov. subsidy up to ¥150B (combined with Yokkaichi). Ramping in stages; meaningful output H1 2026.

JV Structure

50/50 Flash Forward JV

SanDisk and Kioxia split all output 50/50. Total JV capacity ≈ 30% of global NAND. Kioxia sold out all 2026 production. Neither party benefits from flooding market.

NAND Market Share — 2025/2026 (approx.)

SanDisk = WD/Kioxia JV output × 50%. Combined JV ~29–30% of global NAND bits.

NAND Fab Buildings — Operational vs Under Construction

WD/Kioxia count includes all 9 JV buildings (SanDisk receives 50% of output). Samsung estimate includes V- and P-series NAND lines.

Key NAND Capex — 2026 Announced Spend ($B)

Samsung and SK Hynix capex includes DRAM/HBM. Micron capex is total (majority DRAM). SanDisk/Kioxia JV capex is shared — SanDisk books ~50%.

Analyst Ratings & Price Targets — May 2026

Firm	Analyst	Rating	Price Target	Upside vs $1,479	Date
Citigroup	Asiya Merchant	Strong Buy	$2,025	+37%	May 19 '26
Susquehanna	—	Buy	$2,000	+35%	May 1 '26
Bernstein	Mark Newman	Buy	$1,700	+15%	May 4 '26
Raymond James	—	Buy	$1,470	-1%	May 1 '26
Barclays	Thomas O'Malley	Hold	$1,200	-19%	May 1 '26
RBC Capital	Srini Pajjuri	Hold	$1,000	-32%	May 1 '26
Consensus	19 analysts	Buy	$1,460 avg	-1%	—

High PT: $2,350 · Low PT: $1,000 · 15 Buy/Strong Buy · 4 Hold · 0 Sell. The spread is entirely a cycle-duration call, not a company quality call.

Price Target Distribution

10-Year Earnings Scenario (SemiAnalysis Bull Case)

Assumes upcycle holds through FY27, rollover FY28, trough FY29, next peak FY32–33. Cumulative ~$63B net income vs $219B market cap today.

Data Provenance — What Is Sourced vs Estimated

✓ Sourced Data verified against a primary source (earnings release, TrendForce press release, SEC filing). ~ Estimated Derived or modeled — treat as directional, not factual. ⊘ Paywalled Data exists but requires a TrendForce / IDC subscription.

Revenue & Margins Tab

Data	Status	Source
WD Flash quarterly revenue (FY22–FY24, pre-spinoff)	✓	Western Digital 10-Q / 10-K SEC filings — Flash segment
SNDK quarterly revenue (Q1–Q3 FY26)	✓	SanDisk Corp 8-K earnings press releases (SEC EDGAR, Feb 2025+)
Gross margin figures	✓	Same SEC filings; cross-referenced stockanalysis.com
Q4 FY26 revenue guidance $7.75–8.25B	✓	SNDK Q3 FY26 earnings call (May 2025)

Fabs Tab — Supply & Demand Charts

⚠ The exabyte (EB/year) figures on the supply and demand charts are estimates reconstructed from market share percentages and growth rates. TrendForce publishes the authoritative EB data but it is subscription-only. Treat the EB numbers as illustrative of the shape and direction of the market, not precise figures.

Data	Status	Source
NAND market share % by company (Q1–Q4 2024, Q1–Q4 2025)	✓	TrendForce quarterly press releases (trendforce.com/presscenter)
Quarterly revenue by company (2023–2025, all quarters)	✓	TrendForce quarterly press releases — exact $ figures verified
Total industry revenue: 2023 = $38.7B, 2024 = $65.6B, 2025 = $65.0B	✓	Summed from verified TrendForce quarterly press releases
2026 revenue forecast: $147.3B (+112% YoY)	✓	TrendForce Jan 2026 press release (trendforce.com/presscenter/news/20260122-12893)
Enterprise SSD bits: 7.5 EB (2023) → 30 EB (2024)	✓	TrendForce Apr 2024 (trendforce.com/presscenter/news/20240423-12122)
Total bit demand growth: +16% YoY in 2024	✓	TrendForce Aug 2023; Micron FY2024 earnings guidance ("mid-teens %")
Supply bit demand growth 2026: +20–22% (demand) vs +15–17% (supply)	✓	TrendForce Mar 2026 press release (structural shortage forecast)
2026 industry capex: $22.2B total; Kioxia/SNDK $4.5B (+41%)	✓	TrendForce Nov 2025 (trendforce.com/presscenter/news/20251113-12780)
Total NAND supply in EB/year by company (supply chart)	⊘	TrendForce DataTrack subscription — chart uses estimated values from share % × est. total
Demand segment breakdown in EB (mobile, PC, AI, enterprise)	⊘	Yole Group / IDC subscription — chart uses estimated values based on segment growth rates
Forward supply/demand (2026–2030 EB)	~	Modeled from: TrendForce growth rates + company capex guidance + analyst consensus

vs. Peers Tab & Analyst Tab

Data	Status	Source
Market share % (peers doughnut chart)	✓	TrendForce Q2 2024 press release via StorageNewsletter
Fab building counts per company	~	Estimated from company investor presentations + news coverage
Capex figures ($B per company)	~	Estimated from earnings guidance; includes DRAM for Samsung/Hynix/Micron
Analyst ratings, price targets, dates	✓	Individual firm research notes (May 2025–May 2026); cross-referenced via financial news

Key Free Sources

Source	What It Covers	Access
trendforce.com/presscenter	Quarterly NAND revenue rankings, market share, pricing outlook, capex, forecasts	Free — press releases
blocksandfiles.com	Aggregates TrendForce data with commentary; good for historical quarters	Free
storagenewsletter.com	Quarterly market share tables (cites TrendForce)	Free
SEC EDGAR (sec.gov/cgi-bin/browse-edgar)	SNDK 8-K earnings, WD 10-K/10-Q flash segment, Micron quarterly earnings	Free
news.skhynix.com	SK Hynix IR releases — annual NAND revenue, strategic commentary	Free
investors.micron.com	Micron quarterly NAND bit shipment growth, ASP trends, segment revenue	Free
TrendForce DataTrack	Full quarterly EB by company, segment demand breakdown, historical share tables	Subscription required
IDC SSD Forecast (US53968825)	Segment demand 2025–2029 in volume and revenue	Subscription required
Yole Group NAND Market Monitor	Full supply/demand in EB with segment breakdown	Subscription required
SemiAnalysis "Memory Mania"	Bull case AI storage demand thesis, fab-level analysis	Free preview / Subscription for full

Wolfspeed Corporation (WOLF)

~$47

new equity · post-Ch.11 (Sep '25)

$2.3B

Market Cap

−27% GAAP

Gross Margin Q3 FY26 (−21% non-GAAP)

$150M

Q3 FY26 Revenue

+30%

AI DC QoQ (Q3 FY26)

$1.35B

CHIPS Act Grant

Quarterly Revenue & Gross Margin — FY2023–FY2026

⚠ Wolfspeed filed Chapter 11 and emerged September 29, 2025. Successor period (Q2 FY26+) uses fresh-start accounting — asset values reset, not comparable to Predecessor. GAAP GM: −27% Q3 FY26; includes $46M underutilization charge which is NOT excluded from non-GAAP (non-GAAP is −21%, adjusting only SBC and restructuring). Ex-underutilization, implied GM is ~+4%.

The Underutilization Problem

Wolfspeed built Mohawk Valley — the world's largest SiC fab — at a cost of ~$5 billion. Fixed costs are enormous regardless of output.

When a factory runs at 20-30% utilization, the fixed overhead per chip is brutal. The path to positive gross margins requires hitting ~40-50% utilization.

Current utilization (est.)

~28%

Breakeven (~40%)

40%

Target margin (~50%+)

55%

Source: Wolfspeed earnings calls; utilization % is management estimate, not disclosed precisely.

Q3 FY26 — Why Stock +7% on a Miss

Item	Value	Signal
Revenue	$150.2M	Miss vs ~$210M est.
EPS	−$3.26	Deep loss
GAAP gross margin	−27%	GAAP loss: −$40M
Non-GAAP gross margin	−21%	Adj. for SBC + restructuring
Underutilization charge (in COGS)	$46M	Not excluded from non-GAAP
GM ex-underutilization (implied)	~+4% GAAP	Underlying biz is near breakeven
AI DC revenue	+30% QoQ	Accelerating
Debt cut (senior secured)	−43%	Post-Ch.11 restructuring
Cash raised (private)	$476M	Runway extended
Annual interest saved	~$62M	Structural improvement

The $46M underutilization charge is the real story — it's embedded in COGS but not excluded from non-GAAP. Without it, GAAP GM would be roughly +4%. The charge stays until Mohawk Valley runs at meaningful utilization. Durham 150mm device production is now wound down; it's all a Mohawk Valley story.

The Three-Facility Chain

DURHAM, NC — LEGACY

150mm Crystal + Device

Grows small 150mm SiC crystals. Old 150mm device fab winding down. Still sells some substrates externally (Materials segment ~$50M/qtr). Ignore for growth thesis.

→

SILER CITY, NC — THE JP

200mm Crystal Factory

World's largest SiC crystal facility. Grows 200mm boules at ~2700°C over weeks, slices into substrate discs. Feeds Mohawk Valley. Also sells substrates to competitors who build 200mm device fabs. Started production readiness late 2025. Built in phases — each phase unlocks more Mohawk capacity.

Investment: ~$2–3B · Status: Ramping

→

MARCY, NY — MOHAWK VALLEY

200mm Device Fab

Takes JP's 200mm substrates. Etches SiC power chips (MOSFETs, diodes). Only US-scale 200mm SiC device fab. Opened 2023. Ran at 10–20% until JP started supplying at volume. Now ~28%. Revenue potential: >$2B/yr at full utilization.

Investment: ~$2–3B · CHIPS Act: $1.35B · Status: 28% utilized

JP and Mohawk are designed as a matched pair — JP's capacity phases were built to unlock corresponding Mohawk utilization steps. Low utilization 2023–2025 was partly JP being under construction (substrate supply constraint) and partly soft EV demand (demand constraint). Both hitting at once.

Mohawk Valley — Key Stats

Total capex (JP + Mohawk)

~$5B

CHIPS Act grant

$1.35B

Wafer size

200mm SiC

Current utilization

~28%

Revenue at 28%

~$400M/yr

Revenue at full capacity

>$2.6B/yr

The math: At 28% utilization, Wolf earns ~$400M/yr from devices. Each 1% of utilization added is worth ~$14M/yr. Getting to 40% (GM breakeven) means adding ~$170M/yr in revenue — entirely dependent on more EV orders or early datacenter wins.

The risk: Both constraint types are easing slowly. JP is now producing. EV demand is recovering but not booming. Wolf must stay solvent through 2028 to reach the Phase 3–4 inflection where their 10kV moat matters.

Funding Timeline

Cumulative capex vs. government support. ~$5B total across JP + Mohawk Valley; $1.35B CHIPS Act grant offsets a portion. Post-bankruptcy debt: ~$1.7B (down from ~$6.4B). Annual interest reduced ~$62M/yr.

Competitive Moat Assessment — Where Does Wolf Stand by 2029?

Capability	Wolf Today	STM by 2029	Infineon by 2029	Onsemi by 2029	Moat durability
200mm crystal growing	✓✓ 35yr expertise, JP at scale	✗ buys substrates	✗ buys substrates	~ (via GTAT acq.)	Strong — JP yield curves take years
200mm device fab	✓ 6yr head start, Mohawk	~ Catania 200mm ramping	~ Dresden 200mm planned	~ Hudson NH transition	Narrowing — competitors catching up on 1200V
1200V SiC devices (EV)	✓ competitive today	✓ strong by 2029	✓ strong by 2029	✓ strong by 2029	Weak by 2029 — commoditising
3300V SiC (rectifiers)	✓ available now	✗ not announced	~ limited	✗	Good — specialized market
10kV SiC (SST/Phase 4)	✓✓ only commercial bare die	✗	✗	✗	Very strong — nobody close

⚠ Estimated based on public capex announcements and technology roadmaps. Competitors building 200mm device fabs still need to BUY 200mm substrates — from Wolf (JP), Coherent, or SiCrystal. Wolf benefits as substrate supplier even when competing on devices.

800VDC Transition — 4 Phases & Wolf's Role (Source: SemiAnalysis, May 2026)

PHASE 1 · 2026–2027

Sidecar Retrofit

Power rack added alongside existing AC. Uses 650V GaN + 400V capacitors — EV supply chain, not Wolf's voltage class.

Wolf: ✗ minimal

PHASE 2 · 2027–2028

800V-Native Compute

Servers built 800V-native, UPS eliminated. Still GaN FETs + some 1200V SiC. Wolf competes on a small slice.

Wolf: ~ small

PHASE 3 · 2028–2029

Facility-Level DC

1200–3300V SiC centralized rectifiers replace per-rack gear. DC distributed across entire hall. Wolf competitive here.

Wolf: ✓ growing

PHASE 4 · 2029+

Solid State Transformers

Direct MV→800VDC. Only 10kV SiC MOSFETs qualify. Wolf shipped bare die March 2026 — only commercial source. ~$13B TAM.

Wolf: ✓✓ moat

Bottom line: Through 2028, Mohawk Valley utilization ramp is almost entirely EV-dependent. The 800VDC tailwind is structurally a Phase 3–4 story arriving 2029+. No datacenter SST has UL certification as of May 2026.

SiC TAM by Lane 2025–2030E ($B)

⚠ Estimates. EV from Yole/consensus. "Sidecar lane" = Phase 1–2 GaN + low-V SiC (Navitas/Infineon territory). "Wolf's lane" = Phase 3–4 high-V SST SiC. Dashed orange = Wolf's total addressable revenue across all segments.

Mohawk Valley Utilization — Base Case Forecast

⚠ Estimated base case. EV recovery drives 2026–2028; SST/Phase 3–4 ramp drives inflection 2029+. Assumes Wolfspeed remains operational post-Chapter 11. GAAP gross margin breakeven at ~40%.

Who Wins Each Phase

Phase	Timeline	Technology	Primary Beneficiaries	Wolf's Cut
EV / Auto (core)	Now	1200V SiC MOSFETs	Wolf, STM, Infineon, Onsemi, Rohm	~15–20% share — current revenue base
Phase 1 — Sidecar	2026–2027	650V GaN, 400V caps	Navitas, Delta, Vertiv, Eaton	Minimal — wrong voltage class
Phase 2 — Native 800V	2027–2028	GaN FETs, some 1200V SiC	Navitas, GaN Systems, STM, Infineon	Small — 1200V SiC niche
Phase 3 — DC Facility	2028–2029	1200–3300V SiC rectifiers	Wolfspeed, Infineon, Onsemi	Growing — higher-voltage SiC competitive
Phase 4 — SST	2029+	10kV SiC MOSFETs	Wolfspeed (only commercial bare die)	Dominant — unique capability, $13B TAM

Data Provenance

✓ Sourced Verified against earnings release or SEC filing. ~ Estimated Directional / modeled. ⊘ Paywalled Exists but requires subscription.

Data	Status	Source
Q3 FY26: Revenue $150.2M, GAAP GM −27%, EPS −$3.26	✓	Wolfspeed Q3 FY2026 press release (BusinessWire May 5, 2026) + income statement reconciliation table
Non-GAAP GM −21% (adjustments: SBC $2.8M + restructuring $6.2M)	✓	Official non-GAAP reconciliation table in same press release
Underutilization charge: $46M (in COGS, NOT excluded from non-GAAP)	✓	CFO statement on Q3 FY26 earnings call (Motley Fool transcript, May 5 2026)
Implied GM ex-underutilization: ~+4% GAAP, ~+10% non-GAAP	~	Calculated: ($150.2M revenue − $40M GAAP loss + $46M utilization) / $150.2M
Q3 FY26 segment split: Power $100.1M (90% from Mohawk Valley), Materials $50.1M	✓	Q3 FY26 earnings press release segment table
Q3 FY25 Predecessor GAAP GM: −12%	✓	Comparator column in Q3 FY26 earnings release (Predecessor, not comparable)
Q2 FY26 GAAP GM: ~−40% (Successor)	~	Derived from earnings call comment: Q2→Q3 improved "double-digits" from ~−40%
Chapter 11 bankruptcy + emergence Sep 29, 2025; fresh-start accounting adopted	✓	Wolfspeed 8-K (Sep 2025) + Q3 FY26 press release accounting note
Private placement $476M; senior secured debt −43%; saves ~$62M/yr interest	✓	Q3 FY26 earnings press release + 8-K (SEC EDGAR)
CHIPS Act grant: $1.35B	✓	US Dept of Commerce announcement + Wolfspeed press releases (2024)
Mohawk Valley total capex: ~$5B; 200mm wafer size	~	Analyst consensus / Wolfspeed investor presentations ($4.5–5.5B range cited)
Fab utilization ~28%	~	Derived — management did not disclose precise %; stated utilization improvement is "key lever"
SiC TAM forecasts ($B by year)	⊘	Yole Group SiC Market Monitor (subscription) — chart uses consensus range from free analyst summaries
Navitas: Q1 2026 revenue $8.6M, AI +50% QoQ, $125M equity raise	✓	Navitas Q1 2026 earnings press release + 8-K (May 2026)
800VDC 4-phase transition, sidecar TAM ~$11B peak 2028, SST TAM ~$13B by 2030, Wolf 10kV SiC bare die availability, Phase 1–2 uses GaN not Wolf	✓	SemiAnalysis "Inside the 800VDC Revolution — Part 1" (May 26, 2026) — Nicolas Bontigui et al.
Utilization forecast & Wolf addressable revenue by phase	~	Estimated/modeled based on SemiAnalysis phase timeline + Yole EV SiC consensus

NVTS Navitas Semiconductor Pure-play GaN · Fabless · 800VDC Phase 1–2

MKT CAP
~$4.24B

REVENUE RUN RATE
~$35M/yr

EV/REVENUE
~200×

GROSS MARGIN
~45–50%

MODEL
Fabless

STOCK (52W)
$5.44 → $34.17 · now ~$17

Annual Revenue — The Pivot Story ($M)

2021–2024: consumer GaN era (phone chargers, laptops). 2025: deliberate pivot — exited low-margin consumer, revenue halved. 2026E: AI/datacenter ramp begins. The market is not paying for current revenue.

The Valuation Problem

Scenario	Revenue (2030E)	Multiple	Implied mkt cap	Return
Bear	$200M	8×	$1.6B	−77%
Base	$500M	12×	$6.0B	−13%
Bull	$800M	18×	$14.4B	+109%
Mega-bull	$1.2B	22×	$26.4B	+283%

The core problem: Base case at $500M revenue delivers a loss vs today. You need the bull scenario AND a rich multiple simultaneously — but by 2028-2029 when revenue peaks, the sidecar market is rolling over, which compresses multiples rather than expanding them.

What could change this: A second act — GaN in Phase 3–4 (higher voltage GaN beyond 650V), or acquisition premium (Infineon paid $830M for GaN Systems at far smaller scale).

Market Cap vs Revenue — The Dislocation

Revenue peaked in 2024 ($83M) and fell sharply on the consumer pivot. Market cap moved in the opposite direction in 2026 on the 800VDC narrative. The gap is the bet the market is making.

Why Navitas Wins Phase 1–2 (and Only Phase 1–2)

THE VOLTAGE FIT

800VDC bus → 48V requires conversion at 400–650V. GaN excels here — faster switching, lower on-resistance, smaller magnetics than SiC. SiC is overkill at these voltages; silicon breaks down. GaN is the only right answer for Phase 1–2 sidecar racks.

GANFAST ADVANTAGE

Competitors sell bare GaN transistors — engineers must design their own gate driver. Navitas integrates the gate driver into the GaN IC (GaNFast). Dramatically easier to design in, fewer components, faster qualification. This is why hyperscaler power teams prefer Navitas — they're time-poor, not cost-poor.

EV SUPPLY CHAIN BRIDGE

The OCP Diablo spec explicitly uses EV-grade 650V GaN FETs. Navitas already supplies those GaN chips for EV on-board chargers. The same chip crosses over to the datacenter with zero incremental R&D. Google stated they're using EV-grade components in their 800VDC designs.

Phase 1–2 Sidecar GaN Opportunity ($B)

⚠ Estimated. Sidecar GaN content = ~8–10% of power rack BOM. Peak ~$1.5B in 2028 then declines as Phase 3–4 SSTs displace sidecar racks. NVTS potential share ~25–35% if design wins materialise. The peak-and-decline shape is the key risk.

The Information Cascade — Who Knew When

2023 — INSIDERS

OCP publishes Diablo spec with 650V GaN. Navitas engineers in the working group. Hyperscaler architects commit to 800VDC internally. NVTS ~$8.

2024 — TIER 1 INSTITUTIONAL

Goldman/MS semi analysts doing primary research. Navitas winning design-ins for Kyber-class power racks. Smart hedge funds start building. NVTS $5–9.

EARLY 2026 — CONFIRMED

Nvidia GTC: 800VDC on stage. Vertiv/Eaton announce H2 2026 commercial products. TI reference design with Nvidia published. NVTS $5 → $15.

MAY 2026 — SEMIANALYSIS PUBLISHES

"Inside the 800VDC Revolution Pt 1." Narrative reaches broader institutional + retail. NVTS $15 → $29. +200% last month. Early money distributes into strength.

H2 2026 — REVENUE MOMENT OF TRUTH

Kyber racks ship. If Navitas chips are inside, revenue inflects. If timeline slips, multiple collapses. This is what the stock is pricing.

The Bull Case

Design wins compound. Semiconductor design-ins take 12–18 months to show in revenue but lock in 4–5 years of supply. If Navitas is already qualified in Kyber-class sidecar designs (likely, given Google OCP statements), revenue starts Q3–Q4 2026 and grows for years regardless of competition.

GaNFast is a real moat. Infineon/TI sell raw GaN FETs. Navitas sells an integrated IC with gate driver. For power supply designers under time pressure, GaNFast dramatically reduces design time. Stickier than a commodity transistor.

200mm GaN (with PSMC). Navitas announced 200mm GaN production with PSMC in 2025. Same story as Wolf going 200mm SiC — 78% more chips per wafer run. At scale, structural cost advantage over competitors still on 150mm.

Acquisition floor. Infineon paid $830M for GaN Systems (tiny, pre-revenue). Navitas has a brand, customer relationships, and design-in momentum. Strategic value to TI, Infineon, or even a hyperscaler wanting vertical integration is well above current price.

The Bear Case

Peak market, peak multiple — simultaneously. The sidecar market peaks in 2028 and declines as Phase 3–4 takes over. By the time Navitas hits $500M revenue (2028), growth is decelerating. Markets give 20× revenue only to fast-growing companies. Deceleration → multiple compression → flat or down stock despite revenue growth.

GaN commoditises fast. TI, Infineon, Onsemi, STM all have GaN products. At 650V, this is a known process. Price erosion starts the moment hyperscaler procurement gets competitive bids. Navitas's GaNFast premium compresses over time.

Revenue actually fell last year. $83M → $46M in 2025. The pivot story is real but it means they have almost no current revenue. One quarter of design-win delay and the narrative breaks.

This is a trade, not an investment. Even the bull case requires selling before 2029 when the sidecar market rolls over. Getting the exit right matters as much as the entry. The 200% move last month means the trade may already be crowded.

NVTS vs WOLF — Two Very Different Bets

	Navitas (NVTS)	Wolfspeed (WOLF)
Market cap	~$6.9B	~$3B
Revenue (TTM)	~$35M/yr	~$600M/yr
EV/Revenue	~200×	~5×
Revenue required to justify mkt cap	$700M+ at 10×	Already at 5× — EV alone partially justifies
Market window	Peaks 2028, declines after	SST moat grows through 2030+
Multiple needed	20× at peak revenue	6–8× at full capacity
Main risk	Timing the exit perfectly	Survival to 2028–2029
Verdict	Trade with narrow window	Investment if they survive

Data Provenance

✓ Sourced Verified. ~ Estimated Modelled / directional. ⊘ Paywalled Exists but requires subscription.

Data	Status	Source
2024 revenue $83.3M; 2025 revenue $45.9M (−44.9%)	✓	Navitas Q4 FY2025 earnings press release + 8-K
Q1 FY2026 revenue $8.6M; AI segment +50% QoQ; $125M equity raise	✓	Navitas Q1 FY2026 press release + 8-K (May 2026)
Market cap ~$6.9B; 52-week range $4.22–$29.50	✓	Yahoo Finance / StockAnalysis.com (May 2026)
Market cap ~$1.2B (May 2024 estimate)	~	Derived from historical stock price data (~$5–7 range, ~200M shares)
OCP Diablo 400 spec: 650V GaN FETs + 400V capacitors	✓	OCP (Open Compute Project) Diablo 400 specification, publicly available
Sidecar TAM ~$11B peak 2028; Phase 1 H2 2026; GaN = downstream stage	✓	SemiAnalysis "Inside the 800VDC Revolution — Part 1" (May 26, 2026)
Navitas 200mm GaN production plan with PSMC	✓	Semiconductor Today "Navitas plans 200mm GaN production with PSMC" (Jul 2025)
GaN TAM / sidecar GaN content estimates	~	Modelled: Yole Group power GaN market monitor (paywalled) + SA sidecar TAM × estimated BOM share
Infineon acquired GaN Systems for $830M	✓	Infineon press release (Aug 2023)
Information cascade timeline (who knew when)	~	Synthesised from OCP publication dates, GTC presentations, SA publish date, NVTS price history

₿ Bitcoin Miners → AI Compute Power is the scarcest AI asset. Miners already have it.

AI DEALS SIGNED (SECTOR)
>$40B

SECTOR POWER (TOP 7)
~6+ GW contracted

AI REVENUE PREMIUM
3–25× vs mining/MW

BTC HASHRATE Q1 2026
↓ First decline since 2020

All Major Miners — May 2026

Company	Mkt Cap	Hashrate	Power	Power Cost	AI Pivot	AI Deal Size	₿ Conviction
IREN AU/US	$16.7B	~10 EH/s	4.5 GW pipeline	~$0.04	✓ Done	$9.7B (Microsoft)	Pragmatic
Hut 8 HUT	$11.5B	~8 EH/s	8.5 GW pipeline	~$0.05	✓ Done	$7B (Google/Fluidstack)	Exiting BTC
TeraWulf WULF	$11.3B	~10 EH/s	~500 MW (nuclear)	~$0.03 ☢	✓ Done	$12.8B (Core42/Fluidstack)	Opportunistic
Riot RIOT	$9.3B	38.5 EH/s	1+ GW (TX)	~$0.03	↗ In Progress	AMD lease (Rockdale)	BTC-leaning
Core Scientific CORZ	$8B	20+ EH/s	1.2 GW infra	~$0.04	✓ Done	$10.2B (CoreWeave)	Pragmatic
CleanSpark CLSK	$4.3B	50 EH/s	1+ GW (multi-state)	~$0.04	✗ Pure Miner	None	Maximalist
MARA Holdings MARA	$3.6B	66.4 EH/s ①	1.1 GW	$0.04	✗ Pure Miner	None (hodl strategy)	Maximalist

① MARA has the world's largest hashrate but a fraction of the market cap of AI-pivoted peers. ₿ Conviction = 5 squares = full BTC maximalist, 0 = actively exiting mining.

Market Cap vs Hashrate — The AI Premium

MARA has 6.6× IREN's hashrate but 4.6× lower market cap. The market has completely repriced miners as AI infrastructure companies. Pure BTC conviction is being valued less than power + land + pivot willingness.

Revenue per MW — Mining vs AI Hosting ($K/MW/yr)

At $100K BTC, mining earns ~$200K/MW/yr. Basic AI hosting (hotel model, no GPU ownership) earns ~$600K-$1M. If you own the GPUs, $2-4M/MW at H100 rack rates. The gap is why miners are selling their power capacity to CoreWeave/Microsoft.

The Hotel Model vs Own-GPU Model

HOTEL MODEL (CORZ/HUT approach)

Partner like CoreWeave brings GPUs + customers. Miner provides power, cooling, physical security. Revenue: ~$600K–1M/MW/yr on long-term contract. Zero GPU capex risk. No customer acquisition needed. Lower upside but very predictable. 12-15 year contracts.

OWN-GPU MODEL (IREN approach)

Miner owns the H100/B200 clusters AND the building. Revenue: $2–4M/MW/yr at spot GPU rates. Massive GPU capex ($25K/H100). 10,000 H100s = $250M before a single rack is lit. Plus InfiniBand fabric ($50-100M). Plus liquid cooling retrofit. High risk, high upside.

WHAT TRANSFERS FROM MINING

✓ Power (most valuable — 3-5yr queue for new connections)
✓ Land + permitting (already done)
✓ Operations teams (partial overlap)
✗ ASICs (worthless for AI)
✗ Cooling (air → liquid retrofit needed, $5-15M/MW)
✗ Networking (InfiniBand = zero mining equivalent)
✗ Customers + software (completely different)

Difficulty to Switch: What You Keep vs What You Rebuild

WHAT TRANSFERS — HIGH VALUE

Power contracts — 3-5 year queue for new grid connections. $50-200M to build a substation. Miners have it. This alone is worth hundreds of millions.

Land + zoning — Already industrial-permitted. Often near transmission lines. Cheaper power states (TX, WY, KY).

Ops teams — Electrical engineers, facility ops, 24/7 NOC. Partial overlap with AI data center operations.

WHAT MUST BE REBUILT — EXPENSIVE

Hardware — all of it — ASICs are worthless for AI. Buy H100/H200/B200 GPUs from scratch. $25K/chip. 10,000 chips = $250M before one job runs.

Cooling — full retrofit — Bitcoin mining = air cooled. B200 at full performance = must be liquid cooled. $5-15M/MW to retrofit buildings.

Networking — zero overlap — Bitcoin miners need only internet access. AI clusters need InfiniBand 200-400Gb/s between every GPU. $50-100M for a serious fabric.

Customers + software — Mining is transactional. AI cloud requires sales to ML labs, enterprises, hyperscalers. Different relationships, different sales cycle.

Verdict: The smart move for most miners is the hotel model — sell power + space to CoreWeave/Microsoft/Fluidstack who bring the GPUs and customers. This captures 30-40% of the AI economics with near-zero incremental capex risk. The miners who try to own the GPUs themselves (IREN) have higher upside but are taking on a completely different business.

The ₿ Conviction Factor — Who Will Actually Switch?

Miner	Why They Might Stay	Why They Might Switch	Verdict
MARA	CEO Fred Thiel is a public Bitcoin maximalist. "Hodl all BTC" is the explicit strategy. Expects BTC to $500K+.	Largest hashrate in the world but trading at a deep discount to AI-pivoted peers. Board pressure possible.	Won't switch. Bet on BTC upside.
CleanSpark	Identity is "the most efficient Bitcoin miner." Every quarter they talk about J/TH. It's their brand.	Valuation gap vs peers growing. 50 EH/s at $4.3B vs IREN's 10 EH/s at $16.7B is glaring.	Unlikely near-term. Possible if BTC disappoints.
Riot	Jason Les is BTC-committed. Corsicana was built as the world's largest mining site.	Already doing AMD lease at Rockdale. Calling itself "data center builder." Has 1GW+ that needs a business model.	Gradual pivot likely. Won't abandon mining but AI gets bigger.
Core Scientific	Still mines BTC. Some self-mining revenue continues.	Sold BTC to fund AI conversion. 39% revenue already AI. CoreWeave deal is the playbook.	Already switched. Template for others.
Hut 8	Some residual mining. Keeps small BTC treasury.	Literally said "Bitcoin not a long-term strategic focus." Done.	Fully switched. Fastest pivot in sector.