Conviction: Speculative-to-Low (high-asymmetry, high-volatility) Status: Researching — not held in any account, not currently in any watchlist (coverage gap)

---

Editorial Note

AOSL is the $900M-cap David that beat MPWR to GB300 by being first to OpenVReg. A company smaller than one-seventieth of MPS broke through NVIDIA's GPU power ladder by walking into the room before the standard was finalized — December 2024 OVR4-22 controller, January 2025 OpenVReg16 production, October 2025 official 800VDC partner list, and per Ming-Chi Kuo the primary DrMOS supplier for GB300 NVL72 at ~70% share. That is a sentence that should not be possible. The tension that makes this trade interesting is that NVIDIA's standards diversification opened the door — and physics builds the wall behind it. AOSL got first through the door; whether it can hold a 1,500A current path against MPWR's incumbent firepower in Vera Rubin is the entire bet.

---

The Story Right Now

AOSL closed Friday at $43.08, up 95.7% in 30 days with RSI at 78 — squarely in parabolic territory and 6.3% off its 52-week high of $45.97 set April 24. A year ago this stock was a $17 small-cap power-discrete supplier that nobody on AI Twitter had ever mentioned; the November 2025 close was $17.47. The chart from there to $43 is one of the cleanest "consensus catches up to the news" runs of the cycle, and it ran because of three press releases that nobody who wasn't deep in the GPU power-delivery weeds noticed when they happened: the Dec 2024 world-first OVR4-22, the Jan 2025 industry-first 16-phase OpenVReg16, and the Oct 2025 inclusion on NVIDIA's 14-vendor 800VDC partner list. The market re-rated when Ming-Chi Kuo confirmed AOSL as the primary GB300 DrMOS supplier. Today's price is the post-reveal price, not the pre-reveal price.

The cohort context tells you what's going on. MPWR is at $1,583 RSI 68, VICR $268 RSI 78, POWI $73 RSI 78, NVTS $17 RSI 73, ON $103 RSI 87 — every single VRM/PMIC name on the NuttyCLD AI Power Crisis cohort is in some flavor of late-stage breakout. The whole supply chain just got repriced. AMD RSI 79, ARM 67, MRVL 76, INTC 87 — the AI infrastructure rally has run for a quarter and the delivery layer (everything between the rack and the transistor) is the last cohort the market climbed onto. AOSL ran 96% in 30 days because it was the most asymmetric setup in the cohort: tiny float, no analyst coverage on the AI angle until Q4 2025, and a reference position claim that, if true, justifies a 5-10x multiple expansion versus where it was eight months ago. That's why the move happened. The question now is whether it's done.

The fundamentals don't quite tell you to buy. AOSL's last quarter (Dec 2025) printed $162.3M revenue at -8.2% net margin and -$0.45 EPS — the company is still losing money on a GAAP basis. Q1 FY2026 (Sep 2025) was $182.5M with +37.3% YoY growth in Power IC (the AI exposure), but Q2 guidance came in at $160M ±$10M — sequential decline as PC and consumer power supply softened faster than AI ramped. The thesis isn't "AOSL is a great business right now," it's "the AI mix is shifting fast enough that the business inflects in the next 2-3 quarters." Earnings beat rate of 75% (per leading indicators) is real, but YoY revenue growth is -6.3% and revenue acceleration is -14.5% — those are bearish leading indicators, and the stock running 95% on top of them means the market is paying for future mix, not current performance. Insider activity is mixed: 21 sells against 1 buy by share count (-$4.8M net), but the one buy was Mike Chang, the founding chairman, picking up 33,727 shares on March 9, 2026 — a positive signal from the family that built the company.

Bottom line, this is a coverage gap we just acquired the data on. AOSL is the cleanest expression of the picks-and-shovels via reference position paradigm Nutty's framework suggests. It's also the cleanest expression of "small-cap parabolic on a thesis that hasn't yet shown up in the income statement." Both are true. RSI 78 is not an entry. The question for the desk is whether a pullback to $34-37 (SMA20 area, ~15% down) becomes the entry, or whether the right read here is "we missed the easy money and now the bear case is more interesting than the bull case." Stop is $34.50 (-20%). Target is $56 (+30%) — roughly a return to 2021 cycle highs but well below the all-time $70 from March 2022.

---

Quick Snapshot

---

Action Matrix

---

Price Data

Technical Levels

The technical picture: This is a textbook parabolic blow-off. The stock 2.5x'd in 5 months on a thesis that hasn't yet shown up in the income statement. RSI 78 + 30D return +95.7% is the kind of setup that historically resolves with a 20-30% pullback before the next leg up or a multi-month consolidation. Either way, today is not the entry. SMA20 at $29.50 is the obvious mean-reversion target if the rally cools.

---

Company Overview

One-Liner

Vertically integrated analog/discrete power semiconductor maker — MOSFETs, GaN/SiC FETs, multi-phase controllers, DrMOS, PMICs — that punched above its $1.3B market cap to win co-leader status on NVIDIA GB300 by being first into the OpenVReg standard.

Business Model

Key Segments (FY2025)

Geographic Mix

Company Facts

• Market Cap: $1.29B (per latest fetch — note: the deep-dive context cited "~$900M" was the Q1 figure; stock has run 50%+ since)
• Employees: 2,428
• Founded: 2000; IPO'd 2010 on NASDAQ
• Headquartered: Sunnyvale, CA + Hong Kong (dual hub)
• Manufacturing: Oregon Fab (US, ~6") + Chongqing JV (China, 8" + 12" — 2024 deconsolidation, retained minority stake)
• Last 10-K filed: 2025-08-28
• Exchange: NASDAQ (XNAS)

---

Competitive Analysis

Industry Position

Competitor Map (the NuttyCLD VRM/PMIC cohort)

Competitive Moat

Moat Assessment

Moat Width: Narrow Moat Trend: Widening (on AI sockets) — but fragile (every GPU generation re-opens the socket)

Summary:

AOSL's moat is the OpenVReg reference-position relationship and unusual vertical integration for a $1.3B company. That's narrow but real. The vulnerability is that NVIDIA explicitly publishes a 14-vendor 800VDC partner list precisely because it doesn't want to depend on AOSL or anyone else. Standards diversification opens the door (let AOSL in) but also caps the moat (any of the other 13 can take share next gen). The asymmetry comes from being a $1.3B company holding a socket that an $85B company also wants — even modest revenue capture at NVIDIA scale moves the needle on AOSL's P&L the way it can't move MPWR's.

---

Management Assessment

Leadership

Founder Involvement

Capital Allocation

Red Flags

• Excessive exec compensation — not flagged; comp reasonable for size
• High turnover in key roles — no, founder still active, CEO transition handled cleanly
• Related party transactions — yes, the historical Chongqing JV ownership structure has had related-party complexity (now resolved by 2024 deconsolidation, but worth flagging on China exposure)
• China geopolitical exposure — the SEC 10-K explicitly lists "Geopolitical and economic conflicts between United States and China may adversely affect our business" as a risk factor. Real, not hypothetical.

---

Financials

Key Metrics (TTM ending Dec 2025)

Quality Checks

Revenue Quality

Leading Indicators

---

Valuation

Current Multiples (approximate, May 2026)

Historical Range

Fair Value Estimate

My Fair Value: $45-50 base case (modest upside from here on inflection, with binary tails to $90+ if Vera Rubin socket holds or $25-30 if it doesn't and AI mix stalls)

The peer-multiple comp argument is the bull case, but it requires AOSL to act like a peer — i.e., reach the operating leverage that MPWR/ADI/TXN have at scale. The bear case is that a $1.3B vertical-integrator with 22% gross margins isn't a $90 stock no matter how good the GB300 socket is, because the company never gets to MPWR-style 55% gross margins at this scale.

---

Bull Case

Why This Could Work (5 Reasons)

1. Reference position over scale — the OpenVReg first-mover claim is durable

   • Evidence: World-first OVR4-22 (Dec 2024), industry-first 16-phase OpenVReg16 (Jan 2025), official 14-vendor 800VDC partner (Oct 2025), Ming-Chi Kuo confirmed primary GB300 DrMOS supplier at ~70% share
   • Implication: Standards-defined sockets reward the company that helped define them. AOSL has the relationship-level moat MPWR doesn't on this specific generation.

2. NVIDIA wants multi-source — the 14-vendor 800VDC list is permanent policy, not bridge solution

   • Evidence: NVIDIA explicitly publishes the partner list to advertise non-dependence on any single vendor
   • Implication: AOSL's seat at the table doesn't go away in Vera Rubin or Feynman, even if MPWR retakes lead share. The socket gets fractioned across vendors, not concentrated. AOSL's $900M-cap base means even 10% share of a $5-10B AI power TAM is transformative.

3. Vertical integration is genuine moat at this size

   • Evidence: SiC MOSFETs + GaN FETs + multi-phase controllers + DrMOS + IPMs all under one roof at a $1.3B company. R&D ~14-15% of sales. Owns Oregon fab.
   • Implication: When NVIDIA needs a quick custom design at the standard, AOSL can spin it across the full power stack in-house. Most competitors at this scale source one or two of those layers externally.

4. Power IC mix shift is the inflection — already at 40% of product revenue

   • Evidence: Power IC revenue +37.3% YoY in Q1 FY2026, record quarterly high, ~40% of product mix vs ~25% pre-AI
   • Implication: Mix-shift compounds: Power IC carries higher margins than discrete MOSFETs. Get to 50%+ Power IC and gross margins inflect from 22% toward 28-30% — that's a 50-100% earnings expansion at flat revenue.

5. Founder buy at the breakout is the strongest signal in the table

   • Evidence: Mike Chang (Exec Chairman, 10% owner) bought 33,727 shares on March 9, 2026 — AT the breakout, not before
   • Implication: Insiders typically sell the breakout. A founder buying into RSI 70 with $1.5M+ of his own money says "I think we're going higher from here, even at this level." That's the highest-quality insider signal in semis right now.

Upside Scenario

---

Bear Case

What Could Go Wrong (5 Risks)

1. MPWR retakes share in Vera Rubin — the 70% projected MPWR socket share leaves little room for AOSL

   • How it plays out: KeyBanc + Morgan Stanley sell-side already model MPWR ~70% Vera Rubin VRM at +60% ASP. If that's right, AOSL's GB300 share doesn't transfer. Reference position evaporates next generation.
   • Probability: Medium-high — this is the consensus view among the analysts who matter.

2. Standards diversification is a ceiling, not a floor

   • How it plays out: The same NVIDIA policy that opened the door for AOSL on GB300 means AOSL has 13 other approved competitors. Pricing power compresses; AOSL becomes a commodity supplier behind the leader.
   • Probability: Medium — happens slowly across multiple generations.

3. China supply-chain exposure becomes acute

   • How it plays out: ~50%+ revenue exposure to China + Chongqing JV history + escalating US-China semi restrictions. SEC 10-K explicitly flags this. A new Section 301 / chip export rule that targets China-fabbed power semis hits AOSL specifically.
   • Probability: Medium — depends on policy direction; not zero.

4. Acquired before getting to scale — could be the bear case dressed as a bull case

   • How it plays out: A $1.3B company with a hot AI socket gets bought at $60 by ON or Renesas before the multi-quarter inflection plays out. Shareholders win short-term but lose the multi-bagger optionality.
   • Probability: Medium — the math is plausible at strategic premiums of 30-50%.

5. PC/consumer segment drag persists — current numbers don't support the multiple

   • How it plays out: Q2 FY2026 guide of $160M (-12% sequential) shows the non-AI book is still soft. If AI ramp doesn't outpace consumer/PC decline through 2026, revenue stays flat YoY and the parabolic move retraces hard.
   • Probability: Medium-high — leading indicators (revenue accel -14.5%, YoY -6.3%) say this is happening NOW.

Thesis Killers

• Vera Rubin socket disclosure (Q3-Q4 FY2026 calls) confirms MPWR took >65% share with AOSL <10%
• Q3 FY2026 prints below $155M (vs $160M guide) AND Power IC growth decelerates below +20% YoY
• Loss of OpenVReg-related design wins at Quantum-X / Spectrum-X / Feynman platforms
• US-China semiconductor escalation that specifically targets China-fabbed power components

Downside Scenario

---

Market-Moving News

Historical (What Moved This Stock Before)

Recent News

News Patterns

Pre-AI, AOSL was a pure semis-cycle name — moves with PC/consumer demand. The OpenVReg sequence is the structural story that decoupled it from the PC cycle. Watch for: (1) every NVIDIA architecture reveal with named partners, (2) Ming-Chi Kuo / supply-chain analyst notes, (3) quarterly Power IC growth disclosures. The market historically underreacts to the structural news (took 6+ months for the OVR4-22 win to price in) and overreacts to PC/consumer slowdowns. That's the pattern.

---

Related Markets & ETFs

Sector Context

The whole VRM/PMIC cohort is moving (MPWR, VICR, POWI, NVTS, ON all RSI 68+). AOSL is the most parabolic of the group on a percentage basis because it had the smallest float and the biggest positive structural surprise. This is sector-led plus stock-specific. Distinguishing the two: if SMH cools 10%, AOSL likely drops 20-25%; if AOSL keeps the GB300 socket through Vera Rubin, it can decouple to the upside even on a sector pullback.

---

Cross-References (Where This Appears in Our System)

---

Related Tickers to Monitor

---

Catalysts & Timing

Upcoming Catalysts

Key Dates

---

Entry Strategy

Position Sizing

My conviction: Speculative-to-Low — high-asymmetry but technical setup (RSI 78 parabolic) is wrong for entry today Target allocation: 0-1% pre-pullback, 1-2% on a $34-37 entry, 3% only if Q3 prints clean

Entry Approach

Technical Levels

---

Sources

---

Research Log

---

The Gold

Key Discoveries

Open Questions

• Does Vera Rubin socket disclosure (likely Q3-Q4 2026) confirm AOSL retains material share, or does MPWR take ~70% as sell-side projects?
• What's AOSL's gross margin trajectory if Power IC reaches 50%+ of product mix? (The DCF answer hinges on this.)
• How exposed is AOSL specifically to escalating US-China semiconductor restrictions vs. peers?
• Is the 30K MLCC count per GB300 (Murata story in adjacent perspective) a tailwind or competitive constraint for AOSL's discrete + IC bundle pitch?
• Why was the Mike Chang March 9 buy timed AT the breakout instead of months earlier? (Possible: he was in a blackout window earlier; possible: conviction crystallized on the Q1 print.)

Mistakes (If Applicable)