Editorial Note
The IP/topology rentier — owns the road design, not the socket. POWI is the only volume supplier of 1,250 V and 1,700 V GaN switches on the planet, and those parts let an 800 V data-center bus drop straight to 12 V (or even 1 V at the LEGO-PoL endgame) without the multi-stage ladder MPWR/AOSL/Vicor are fighting over. The thesis is structural rather than socket-specific: if Nutty's "shorten the path" arrow keeps moving and direct conversion becomes the dominant topology, POWI eats — and sell-side hasn't caught up (consensus PT $52-53 vs. spot $73, after a +40% 30-day rip).
The Story Right Now
POWI is the strangest name in the AI-power-delivery cohort. MPWR sells the multi-phase intersection. AOSL holds the OpenVReg reference position. Vicor sells factorized power modules under the GPU. Delta sells the shelf and busbar. POWI sells the topology — the IC and the architecture together — and then licenses or co-designs that architecture into reference designs that NVIDIA, Delta, and the 14 Kyber partners route around. They're on NVIDIA's published 800 VDC partner list (alongside ADI, AOSL, EPC, Infineon, Innoscience, MPS, Navitas, onsemi, Renesas, Richtek, ROHM, ST, TI), but the bull case isn't "POWI wins a Vera Rubin VRM socket." It's "if direct 800 V→12 V (1,250 V PowiGaN, half-bridge) or 1,000 VDC→12 V (1,700 V InnoMux-2) becomes the way the industry skips the 48 V intermediate stage entirely, POWI's IP defines the lane." That is the most radical of the four topologies in NuttyCLD's Part 3 taxonomy, the same philosophical bucket as the Princeton LEGO-PoL academic demo (48 V→1 V at 780 A in a merged hybrid stage). It's also the bucket Vicor lived in for 25 years before Q4 2025 finally broke the curse.
The tape has woken up. POWI is +40.0% over 30 days, +58.3% over 3 months, RSI 75, sitting -3.8% from 52-week high, and gunning into May 7 Q1 2026 earnings. That is a stock pricing in the AI-data-center narrative materially ahead of the actual revenue capture — total 2025 revenue was $443 M (+6%), with PowiGaN product revenue +40% but coming off a small base, and Q4 down 2% YoY. The structural narrative gap matters: this is a $4.1 B market-cap analog/mixed-signal IC company that just printed Q4 operating margin of 8.5% (down sharply from Q3 2024's 10.0% and well below 5-year averages), 2025 industrial +15%, consumer still under inventory pressure. The market is repricing on optionality, not throughput. Insider activity is screaming that house — net 58k shares sold YTD for ~$14.1 M, including a notable director sell of 10,000 shares around the March drawdown. That's a real "house dispersing into the rip" tell.
The market context makes POWI a sympathy play more than a pure-data-center beneficiary. Mega-cap AI tech is parabolic (INTC RSI 87, GOOGL 82, AMZN 77 per the May 1 market brief), the broader semis cohort is stretched (ON RSI 81 with +145% 1Y), and capital that doesn't want to chase $5 trillion mega-caps is rotating into the picks-and-shovels next-tier AI-power names. POWI fits that template: small-enough cap to move on a Q1 print, narrative-aligned with NuttyCLD's series, optically cheap relative to MPWR's $1,583 share price ($85 B+ cap) and even VICR's $268 (~$13 B). But the 121-145x P/E and ~9x EV/Revenue at depressed earnings means the chase has already priced in a non-trivial portion of the optionality.
Our take: this is a medium-conviction "thesis trade with a hard catalyst" — POWI is the cleanest pure-play on the radical-topology endgame, the IP/topology layer of the AI-power stack, and the 800 V Kyber transition is real. But execution risk is high (Vicor's 25-year history is the cautionary tale), the rip into earnings is mechanically extended, and the consensus PT below spot signals sell-side hasn't validated the new framing yet. Worth a starter on a pullback into RSI <55 / $60-65, not a chase at $73 the week of earnings.
Quick Snapshot
| Signal |
Reading |
| Overall |
🟡 Hot tape, structural thesis intact, near-term execution unproven |
| Moat |
Narrow (turning Wide if direct-conversion wins) |
| Key insight |
Only volume supplier of 1,250 V / 1,700 V GaN; 175 M+ GaN switches deployed; sells the topology not just the part |
Action Matrix
| Action |
Level |
Why |
| Current |
🔍 Research / Watch |
RSI 75, +40% 30D, into May 7 earnings — extended |
| Entry Zone |
$60-$70 |
Pullback to ~SMA20 region or RSI <55; preferred entry below $65 |
| Stop-Loss |
$58.65 (-20%) |
Standard skill default; matches break of pre-rip consolidation |
| Target |
$95.30 (+30%) |
Re-rate to ~$5.3 B cap if Q1 confirms PowiGaN data-center traction |
Price Data
| Stock |
Price |
1D |
7D |
30D |
3M |
52wkHi |
RSI |
Status |
Action |
| POWI |
$73.32 |
— |
+0.4% |
+40.0% |
+58.3% |
-3.8% |
75 ↑ |
🟡 Stretched |
🔍 Research |
Legend
- RSI 75 = Overbought zone, FOMO warning
- 52wkHi -3.8% = essentially at the high; no reversion margin
Company Overview
One-Liner
POWI designs analog/mixed-signal high-voltage power-conversion ICs (AC-DC, DC-DC, gate drivers, motor drivers) and is the only volume supplier of 1,250 V and 1,700 V GaN switches — the building blocks for direct 800 VDC AI data-center power architectures.
Business Model
| Question |
Answer |
| What they sell |
High-voltage AC-DC ICs, LED drivers, BLDC motor drivers, IGBT/SiC MOSFET gate drivers, 1,250 V/1,700 V PowiGaN switches |
| Who pays |
OEMs across appliances, mobile, industrial, EV, lighting, data center; named NVIDIA Kyber 800 VDC partner |
| Revenue model |
Component IC sales + reference-design licensing; ~877 employees, fabless, founder-era IP-heavy |
| How sticky |
High — power-conversion IC designs are validated into customer products; topology IP is multi-year design-in cycle |
Key Segments (FY2025)
| Segment |
Revenue % |
Growth |
Notes |
| Industrial (high-power gate drivers, metering, tools, automotive, broad industrial) |
~40% |
+15% |
Record sales in high-power gate-driver business; key growth engine |
| Consumer (appliances, lighting) |
~30% |
weak |
Still under inventory pressure; improvement signal but not growth |
| Computer (incl. AI data center via PowiGaN) |
~15-20% |
accelerating |
The AI optionality bucket; small base, +40% PowiGaN product growth FY25 |
| Communications (mobile chargers) |
~10-15% |
mixed |
Mature; commoditizing fast-charge market |
Geographic Mix
| Region |
Revenue % |
Notes |
| China / Asia |
~75% |
Heavy concentration; tariff and trade-policy risk per 10-K |
| US / Europe |
~25% |
Industrial + automotive |
Competitive Analysis
Industry Position
| Question |
Answer |
| Market share |
Dominant (only volume supplier) in 1,250 V / 1,700 V GaN switches; minority share in lower-voltage GaN/Si AC-DC |
| Market size (TAM) |
AI 800 VDC power solutions: Morgan Stanley estimates Kyber rack power-solution value rises >10x vs. GB200 |
| Growth rate |
PowiGaN +40% FY25; broader high-voltage GaN market early innings |
| Key competitors |
Navitas (NVTS), Infineon (IFNNY), Texas Instruments (TXN), onsemi (ON), Wolfspeed, Innoscience, EPC |
| Position |
Niche leader in radical-topology IP; topology rentier vs. socket vendor |
Competitive Moat
| Moat Type |
Present? |
Evidence |
| Network effects |
🔴 |
Limited; semiconductor design wins don't network |
| Switching costs |
🟢 |
Power-conversion designs validated and locked into product BOMs for years |
| Cost advantages |
🟡 |
Fabless model + GaN process IP; not lowest-cost but highest-voltage |
| Intangible assets |
🟢🟢 |
1,250 V / 1,700 V PowiGaN process IP; deep patent moat in HV power-conversion topologies dating to 1988 |
| Efficient scale |
🟡 |
Small TAM in radical 800 V direct-conversion — efficient scale possible if topology dominates |
Moat Assessment
Moat Width: Narrow today, Widening if direct-conversion wins
Moat Trend: Widening (NVIDIA Kyber 800 VDC partner inclusion + only-volume-supplier status)
Summary:
POWI's moat is process IP and topology IP in the highest-voltage GaN regime. They can do something nobody else has demonstrated in volume — handle 800 V input directly. If the industry collapses the 48 V→12 V→0.8 V ladder into a 800 V→12 V→0.8 V (or even 800 V→1 V) flow, POWI's topologies are the reference. The risk is that the radical path stays radical for another generation, the way Vicor's factorized architecture did from 1981 to 2025.
Management Assessment
Leadership
| Role |
Name |
Since |
Background |
Notes |
| Executive Chairman (interim CEO transition) |
Balu Balakrishnan |
CEO since 2002, Exec Chair 2025 |
National Semi product line manager; joined POWI 1989 (founding-era) |
Age 70; announced retirement Feb 2025; staying through transition |
| CEO |
TBD |
search ongoing |
External + internal candidates |
Search firm retained Q1 2025; not yet named as of May 2026 |
| CFO |
Nancy Erba |
Nov 2025 |
Joined as new CFO |
Fresh hire; first full year as CFO |
| Chief People & Transformation |
Julie Currie |
Nov 2025 |
New role |
Signals organizational reshaping |
Founder Involvement
| Question |
Answer |
| Founder-led? |
Founding-era CEO (Balu since 1989, CEO since 2002) — transitioning out |
| Founder ownership |
Modest; insider net selling YTD = -58k shares, -$14.1 M |
| Skin in the game |
Diminishing — leadership transition + insider sells = "house dispersing" tell |
Capital Allocation
| Metric |
Track Record |
| M&A discipline |
Conservative — no major M&A; organic R&D-driven |
| Buyback timing |
Steady programmatic buybacks; not aggressive on dips |
| R&D investment |
High — $24-26 M/quarter R&D vs. $103-119 M revenue (~22-23% of revenue) |
| Debt management |
Conservative — total liabilities $99 M vs. $772 M assets; net cash positive |
Red Flags
- Leadership transition during AI inflection — uncomfortable timing
- Insider selling into the rip (-$14.1 M YTD net)
- Excessive exec compensation
- High turnover (just two new C-suite hires; not turnover, but instability signal)
- Aggressive accounting
Financials
Key Metrics
| Metric |
Q4 2025 |
Q4 2024 |
Q4 2022 |
Trend |
| Revenue (Q) |
$103.2 M |
$105.3 M |
$107.0 M (est) |
➡️ flat |
| Full-year Revenue |
$443 M |
$419 M |
$651 M (peak 2022) |
📉 below peak |
| Revenue Growth YoY (Q4) |
-2% |
— |
— |
📉 |
| Gross Margin |
52.9% |
54.4% |
~52% |
➡️ |
| Operating Margin |
8.5% |
3.7% |
~25%+ |
📉 vs. cycle peak |
| Net Margin |
12.9% |
8.7% |
— |
📈 |
| FCF Margin (FY2025) |
19.7% |
— |
— |
🟢 — $111.5M OCF less $24.4M filed capex on $443M revenue |
Quality Checks
| Check |
Status |
Notes |
| FCF positive? |
🟢 |
19.7% FY2025 FCF margin; corrected from a single-quarter OCF+investing-CF figure mislabeled TTM (D038/D040) |
| Profitable? |
🟢 |
$13.3 M net income Q4; long-time profitable |
| Debt manageable? |
🟢🟢 |
Total liabilities $99 M; net cash positive |
| Cash runway |
Indefinite |
Conservative balance sheet, dividend-paying |
| Dividend |
🟢 |
$0.215 quarterly, ex-date Feb 27 2026 |
Revenue Quality
| Factor |
Assessment |
| Recurring % |
Low (component shipments) but design-locked |
| Customer concentration |
Diversified across appliances, mobile, industrial, EV |
| Contract length |
Short PO cycles; long design-cycle stickiness |
| China exposure |
~75% — material trade-policy risk |
Leading Indicators
| Indicator |
Value |
Signal |
| Revenue Acceleration |
-15.86% |
🔴 Bearish — Q4 -13% sequential |
| YoY Growth |
-1.94% |
🔴 |
| Earnings Beat Rate |
100% |
🟢🟢 STRONG (4-of-4) |
| Insider Trading |
-$14.1 M net |
🔴 Bearish (0 buys / 58 sells) |
| Composite |
-0.2 |
🟡 Neutral |
Valuation
Current Multiples
| Metric |
Current |
5Y Avg |
Industry Avg |
| P/E (TTM) |
121-146x |
~30x |
40x semis |
| P/S |
~9x |
~7x |
~6x |
| EV/Revenue |
~9x |
~7x |
~6x |
| Market Cap |
$4.1 B |
— |
— |
Historical Range
| Metric |
Current |
5Y High |
5Y Low |
| Stock Price |
$73.32 |
$108 (2024) |
~$48 (2025) |
| % from 52wk high |
-3.8% |
— |
— |
Fair Value Estimate
| Method |
Fair Value |
Upside/Downside |
| Sell-side consensus PT |
$52.50 |
-28% |
| Tickernerd-style range |
$45-$56 |
-23% to -39% |
| Bull-case (data-center re-rate) |
$95-$110 |
+30% to +50% |
| Comps (vs. NVTS, VICR enthusiasm multiples) |
$80-$95 |
+9% to +30% |
My Fair Value: $65-$75 base case (essentially fair-valued at spot if PowiGaN AI ramp delivers); $95+ requires Q1 print to confirm topology adoption.
Critical valuation note: Wall Street consensus PT of $52-53 is below the current $73 price by ~28%. The stock has run ahead of analyst models, which haven't yet incorporated the 800 VDC narrative shift. Either the analysts catch up (re-rate higher), or the rip mean-reverts. May 7 Q1 print is the binary.
Bull Case
Why This Could Work
Radical topology is the long-run endgame.
- Evidence: Princeton LEGO-PoL 48 V→1 V at 780 A in a merged hybrid stage. POWI 1,250 V PowiGaN does 800 V→12 V in a half-bridge. 1,700 V InnoMux-2 supports 1,000 VDC input with >90.3% efficiency to 12 V system in liquid-cooled fan-less 800 VDC.
- Implication: Direct conversion is no longer theoretical — it's NVIDIA Kyber's stated direction. POWI is the only volume supplier of the parts that enable it.
PowiGaN +40% FY25 is the leading edge of an inflection.
- Evidence: 175 M+ GaN switches deployed across fast chargers, data centers, EVs. PowiGaN product revenue +40% FY25 off a small base. Industrial segment +15% (high-power gate drivers, EV charging, metering).
- Implication: The base is small but the growth slope is steep. If data center adds even $50 M to the AI-power-delivery line by FY27, that's 10%+ of total revenue at high incremental margin.
Established business gives the radical thesis a floor.
- Evidence: $443 M FY25 revenue, 19.7% filed FY25 FCF margin, dividend-paying ($0.215/Q), $99 M total liabilities vs. $772 M assets, 100% earnings beat rate. Long history of profitability through cycles.
- Implication: Unlike NVTS (-$ losses, micro-cap), POWI has a real business under the optionality. You're not paying for hope alone.
Topology IP rents are a different revenue model.
- Evidence: NVIDIA Kyber 14-partner list inclusion. POWI publishes reference designs (InnoMux-2, 800 VDC architecture white paper) used by ODMs. Some revenue captured at the reference-design / co-engineering layer rather than pure socket competition.
- Implication: POWI doesn't have to win every Vera Rubin VRM socket. They can lose Vera Rubin entirely and still win the next-gen Kyber-derived architecture if direct-conversion takes hold in 2027-2028.
Sell-side mispricing.
- Evidence: Consensus PT $52-53 vs. spot $73. Strong Buy 67% / Buy 33% rating mix despite below-spot targets — analysts haven't updated models for the AI-power-delivery framing.
- Implication: A solid Q1 print on May 7 with PowiGaN data-center commentary forces a wave of PT upgrades. Re-rate optionality is large.
Upside Scenario
| If This Happens |
Stock Could |
| Q1 2026 PowiGaN +50%+, AI data-center mention in CC |
$90-$100 (+25-35%) |
| Vera Rubin or Feynman 800 VDC reference design uses 1,700 V InnoMux-2 |
$110+ (+50%) |
| Industry adoption of direct 800 V→1 V (LEGO-PoL-style) |
$150+ (re-rates as topology rentier) |
Bear Case
What Could Go Wrong
Radical topology doesn't win commercially — the Vicor curse.
- How it plays out: 1,250 V / 1,700 V PowiGaN technically capable, but MPS multi-phase + Infineon/Delta VPD remain "good enough" through Vera Rubin and Feynman. Direct conversion stays in white papers and academic demos.
- Probability: High — base rate for radical power architectures is bad. Vicor took 25 years to monetize a similar philosophy. POWI's 1,250 V PowiGaN was launched in 2020-2022; data-center adoption is still single-digit % of revenue.
Data center revenue is much smaller than the headline.
- How it plays out: PowiGaN +40% off a small base means ~$15-25 M of data-center revenue in 2025 — meaningful at the gross-margin level, immaterial to total revenue. Q4 2025 industrial outshone consumer; AI was not the headline driver.
- Probability: High — confirmed by current numbers. The thesis requires 2027-2028 ramp, not 2026.
Insider selling into the rip is signaling something.
- How it plays out: $14.1 M net insider sales YTD with zero buys, including 10k-share director sale during the March drawdown. Leadership transition (CEO retiring, new CFO/CPO 6 months in) at the moment of an AI-narrative spike.
- Probability: Medium — could be normal scheduled selling and CEO transition optics, but the cluster is real.
Tariffs and China exposure (~75% revenue).
- How it plays out: Trade-policy escalation hits consumer (mobile, appliances) volume; China customers pull design-ins; competitor Innoscience (China) takes share in lower-voltage GaN.
- Probability: Medium — explicitly called out in 10-K risk factors.
Valuation has run ahead.
- How it plays out: Spot $73, sell-side PT $52-53, P/E 121-146x at depressed earnings. Mean reversion alone is -28% to PT consensus. RSI 75 into earnings is the textbook setup for sell-the-news.
- Probability: Medium-High for short-term mean-reversion; lower for thesis-killing decline.
Thesis Killers
- Q1 2026 print confirms PowiGaN data-center is <5% of revenue with no acceleration commentary
- NVIDIA Vera Rubin reference design selects Navitas 800 V→6 V GaNFast or TI 800 V→6 V over POWI 1,250 V direct
- New CEO appointment de-emphasizes data-center / radical-topology strategy
- China tariff escalation cuts FY26 revenue >10%
Downside Scenario
| If This Happens |
Stock Could |
| Q1 in-line, no AI data-center upgrade |
$58-$65 (-12% to -21%) |
| Q1 miss + soft Q2 guide |
$50-$55 (-25% to -32%) |
| Topology loss in Vera Rubin reference design |
$42-$48 (-35% to -43%) |
Market-Moving News
Historical (What Moved POWI Before)
| Date |
Event |
Impact |
Market Reaction |
Lesson |
| 2022 |
Peak revenue cycle ($651 M) |
+200% trough-to-peak |
Justified |
POWI is highly cyclical |
| 2023-2024 |
Inventory correction |
-50% drawdown |
Overreaction |
Cycle bottom was buyable |
| Feb 2025 |
Balu retirement announcement |
mild |
Neutral |
Founder transitions get absorbed |
| Oct 2025 |
OCP Global Summit white paper on 1,250 V / 1,700 V PowiGaN for 800 VDC |
+20% over weeks |
Justified |
Industry validation matters |
| Mar-Apr 2026 |
NuttyCLD AI Power Crisis Part 3 + 4 |
+40% over 30 days |
Narrative-driven |
The "discovery" by financial Twitter is the bigger move than the underlying news |
Recent News
| Date |
Headline |
Source |
Impact |
Relevance |
| 2026-03-09 |
Director sells 10,000 shares |
Motley Fool |
🟡 |
Insider distribution into drawdown |
| 2025-11-24 |
Julie Currie joins as Chief People & Transformation Officer |
Benzinga |
🟢 |
New role — organizational reshaping |
| 2025-11-18 |
Nancy Erba named CFO |
Benzinga |
🟢 |
Fresh CFO during AI inflection |
| 2025-10-13 |
1,250 V / 1,700 V PowiGaN white paper at OCP Summit |
BusinessWire |
🟢🟢 |
The trigger event for the thesis |
| 2025-02-06 |
Balu Balakrishnan announces retirement |
BusinessWire |
🟡 |
Multi-year transition begins |
News Patterns
POWI moves on (1) cycle inflection prints, (2) NVIDIA partner-list inclusions and OCP Summit announcements, (3) inventory correction signals from peers (MPS, Renesas). The market has never rewarded radical-topology IP claims directly — it requires socket wins or revenue. The current run-up is unusual in that it's pricing optionality without a confirmed socket. Sell-the-news risk is real on May 7.
| Symbol |
Name |
Relationship |
Notes |
| SOXX |
iShares Semiconductor ETF |
Holds POWI |
Sector tailwind |
| SMH |
VanEck Semiconductor |
Holds POWI |
Mega-cap-heavy |
| PSI |
Invesco Dynamic Semis |
Holds POWI |
More mid-cap weighted |
| GANS / SiC ETFs |
Various |
Indirect |
No clean GaN-pure ETF |
| FNGD |
MicroSectors FANG bear |
Indirect inverse |
Risk-off sympathy |
Sector Context
The whole AI-power-delivery cohort is moving together. ON +146% 1Y RSI 81, MPWR $1,583 RSI 67, VICR $268 RSI 69, NVTS $17.45 RSI 67 +31% above SMA20. POWI is mid-cohort in extension. The trade is "AI power delivery" as a basket, not POWI as an idiosyncratic alpha pick.
Cross-References (Where This Appears in Our System)
| Location |
File |
Context |
| Perspectives |
AI Power Delivery |
The IP/topology rentier — "collects tolls on the road design itself" |
| Perspectives |
AI Power Bottleneck |
Parent — generation-side perspective |
| Inputs |
inputs/articles/2026-04-19-nuttycld-ai-power-crisis-part-3-last-few-centimeters.md |
Original radical-topology framing |
| Inputs |
inputs/articles/2026-04-25-nuttycld-ai-power-crisis-part-4-last-micrometer.md |
Series synthesis |
| Journal |
the May 1 full-scan market brief |
Coverage gap surfaced |
| Watchlists |
none yet |
Add to ai-power (or split into ai-power-delivery) |
| Ticker |
Why Related |
Suggested Placement |
Priority |
| NVTS |
Similar radical-density GaN thesis; smaller and more speculative |
ai-power |
🟢 |
| IFNNY (IFX) |
Infineon + Delta VPD combo (Path 3 incumbent); PowerVia partner |
ai-power, semis |
🟢 |
| MPWR |
The dominant multi-phase incumbent (Path 1) vs. POWI's radical Path 4 |
ai-power, semis |
🟢 |
| AOSL |
OpenVReg co-leader on GB300; $900M cap asymmetry |
ai-power, semis |
🟢 |
| VICR |
The philosophical originator that finally got paid in Q4 2025 |
ai-power |
🟢 |
| TXN |
800 V → 6 V → sub-1V reference architecture (Path 2 competitor) |
semis |
🟡 |
| ON |
SiC MOSFET ramp; broader power semis read |
semis |
🟡 |
| EPC |
Pure-play GaN; on Kyber 14-partner list |
ai-power |
🟡 |
Catalysts & Timing
Upcoming Catalysts
| Date |
Event |
Impact |
Watch For |
| 2026-05-07 |
Q1 2026 earnings (after close) |
🟢🔴 binary |
PowiGaN revenue %, AI data-center commentary, Q2 guide |
| 2026-Q3 |
OCP Global Summit (annual) |
🟢 |
New 800 VDC reference designs, partner expansion |
| 2026-2027 |
NVIDIA Kyber rack production ramp |
🟢 |
First real 800 VDC volumes |
| TBD |
New CEO announcement |
🟢🟡 |
Strategic emphasis: AI vs. consumer/industrial |
| 2027 |
TSMC A16 + Super Power Rail volume |
🟢 indirect |
Pulls direct-conversion topology adoption |
| 2028 |
NVIDIA Feynman launch |
🟢 |
First AI accelerator on TSMC A16 + SPR; potential topology inflection |
Key Dates
| Event |
Frequency |
Next Date |
| Earnings |
Quarterly |
2026-05-07 |
| Dividend ex-date |
Quarterly |
2026-05 (next) |
Entry Strategy
Position Sizing
| Conviction |
Allocation |
| Speculative |
0% (research only) |
| Low |
1-2% |
| Medium |
3-5% ← my conviction |
| High |
5-10% |
My conviction: Medium
Target allocation: 3% (basket allocation alongside MPWR, AOSL, VICR, NVTS)
Entry Approach
| Strategy |
Details |
| Entry zone |
$60-$70 (preferred <$65) |
| Starter position |
1% of portfolio at $65-$70 |
| Add on |
RSI <55 + post-earnings flush, or breakout above $80 with Q1 confirmation |
| Full position at |
Confirmed AI data-center revenue acceleration on Q1 print |
Technical Levels
| Level |
Price |
Notes |
| Resistance |
$76 |
52-week high zone |
| Support 1 |
$65 |
Pre-rip consolidation |
| Support 2 |
$58 |
-20% stop trigger |
| Support 3 |
$48 |
Q3 2025 low |
| 52-week high |
$76 (~) |
Just below current |
| 52-week low |
$48 (~) |
March-April 2025 |
Research Checklist
Fundamentals
- Read most recent 10-K (filed 2026-02-06)
- Listen to Q1 2026 earnings call (May 7)
- Review investor presentations (OCP white paper Oct 2025)
- Understand revenue drivers (industrial, consumer, computer, comm)
Competitive
- Map competitive landscape (NVTS, IFX, MPWR, TXN, ON, AOSL, VICR, EPC)
- Assess moat durability (process IP narrow, topology IP strong)
- Check customer reviews / NVIDIA Kyber design-in confirmation
Management
- Research CEO/CFO background (Balu retiring, Nancy Erba new CFO)
- Check insider ownership (-$14.1 M YTD net, 0 buys / 58 sells)
- Review capital allocation history (conservative, R&D-heavy, dividend)
Valuation
- Review historical multiples (P/E 121-146x at depressed earnings)
- Compare to peers (vs. MPWR P/E ~50x, NVTS unprofitable)
- Identify entry/exit zones
Risk
- Write bear case
- Identify thesis-killers
- Set stop-loss level ($58.65)
Sources
Research Log
| Date |
Update |
| 2026-05-01 |
Initial deep dive created (per .claude/skills/deep-dive/SKILL.md). Catalyst: NuttyCLD Part 3 + Part 4 capture, May 1 full-scan brief surfaced AI-power-delivery coverage gap. |
The Gold
Key Discoveries
| Discovery |
Implication |
| POWI is the only volume supplier of 1,250 V / 1,700 V GaN switches (175 M+ deployed) |
Genuine moat — not commoditizable in the near term |
| Sell-side consensus PT ($52-53) is below spot ($73) |
Market is ahead of analysts; either re-rate or mean-revert on May 7 |
| PowiGaN product revenue +40% FY25, but off a small base |
Optionality is real; throughput is not yet |
| CEO retirement during AI inflection + insider selling into rip |
Coincidence or signal? The cluster matters |
| Industrial segment (+15%) is the actual growth driver, not data center |
Today's revenue story ≠ today's stock-price story |
| POWI is the most radical of NuttyCLD's 4 topology buckets |
Highest beta to "shorten the path" arrow continuing |
Open Questions
- Will Q1 2026 (May 7) confirm PowiGaN data-center traction in segment or product disclosure?
- Does NVIDIA Vera Rubin reference design select POWI 1,250 V over Navitas 800 V→6 V or TI 800 V→6 V?
- Who replaces Balu and what is their stated AI strategy?
- Is the China revenue (~75%) at risk under 2026 trade policy?
- Should we track this as part of
ai-power (delivery side) or open a new ai-power-delivery watchlist alongside ai-power-bottleneck (generation)?
Decision
Watch into Q1 2026 print on May 7. Do not chase at RSI 75 / $73 the week of earnings. Add to ai-power watchlist (or split watchlist into delivery/generation per the perspective's open question). Re-evaluate after the May 7 binary. Preferred entry: $60-$65 on any post-earnings pullback or general semis basket digestion. Consider as part of an "AI power delivery basket" alongside MPWR, AOSL, VICR, NVTS rather than as a standalone alpha bet.