Monolithic Power Systems

Phase A — Understand the business

Lens 1 · Company Overview

Monolithic Power Systems is a fabless analog and mixed-signal semiconductor company that designs high-performance power-management ICs and modules — the chips that convert, regulate and deliver electrical power inside electronic systems. Founded in 1997 by Michael Hsing (still CEO/Chairman; took it public in 2004), it sells into six end markets: storage & computing, enterprise data, automotive, communications, consumer, industrial.

The model in plain terms: MPS does not own fabs. It designs the IP, owns its proprietary BCD process + 3D packaging + system know-how, and outsources wafer fabrication and assembly to third parties (typical lead times 16–26 weeks). Its differentiation is integration — more functionality on a single monolithic chip (smaller, more efficient, more accurate, fewer external components), which is exactly what matters as power density explodes in AI servers.

Scale (FY2025): revenue $2,790.5M, +26.4% YoY; gross margin 55.2%; operating income $728.6M (26.1% op margin); GAAP net income $621.5M; 4,500+ employees; one reportable segment.

Customers / channel — concentration is real and rising. MPS sells largely through distributors. In 2025, three distributors were 26%, 18% and 10% of total revenue (i.e. ~54% through three intermediaries). Beneath the distributor layer sits the true demand driver, Nvidia, the dominant buyer of power-delivery silicon for AI GPUs (not named as a 10%+ direct customer because it buys through the channel, but it is the swing factor for the Enterprise Data segment — see Lens 13). Contract structure: standard analog distribution with limited stock-rotation rights (small % of trailing-6-month purchases returnable) and price-adjustment/variable-consideration accounting. No take-or-pay revenue; MPS itself takes on take-or-pay/prepay obligations upstream to its wafer suppliers ($442.2M total purchase commitments, $389.8M due within a year).

Geography: ~92% of FY2025 revenue from customers in Asia (94% in 2024, 87% in 2023) — billing primarily in USD, so reported revenue is FX-insulated, but the demand and manufacturing base sit squarely in the China/Taiwan supply orbit.

Lens 2 · Supply Chain

Map: third-party wafer foundries (geometries MPS specifies) → third-party assembly/test → MPS (design IP, proprietary BCD process, 3D packaging) → distributors (3 account for ~54% of revenue) → OEM/ODM board makers → hyperscaler & system buyers (Nvidia GPU boards the marquee node) → end markets.

Named chokepoints and dependencies, grounded in the filing:

• Upstream — wafers. MPS is capacity-constrained at times: "This lack of capacity has at times constrained our product sales and revenue growth… suppliers may require… long-term capacity reservations, prepayments, take-or-pay terms, or tool funding". Suppliers are not named in the filing; industry reporting points to a foundry mix that MPS has been actively diversifying away from China — note the 2023 executive PSU goal that paid out for "securing additional manufacturing capacity outside China" (139,000 shares awarded). This is a tell: management is being explicitly compensated to de-risk the China foundry footprint.
• MPS's own footprint is in the U.S., Europe and Asia, with subsidiaries in China and Taiwan (Renminbi / New Taiwan Dollar functional currencies); it operates manufacturing/R&D facilities in China and notes the risk of keeping them "at full operational status".
• Distribution chokepoint: the same three distributors are a single point of revenue concentration and an accounts-receivable concentration (the 10-Q tracks distributors ≥10% of AR).
• Downstream chokepoint — Nvidia. The single most important link is the design-win socket on Nvidia GPU power boards. When Edgewater flagged in Nov 2024 that "NVDA has canceled half of MPWR's backlog," the stock fell ~25% — the supply chain's most valuable node is also its most fragile.

Names or it didn't happen: foundry partners are not disclosed in the filing, so they are n/a — not disclosed in 10-K; the demand-side names (Nvidia; distributors at 26/18/10%) are sourced.

Lens 3 · Competitive Advantages (moats)

MPS's moat is process + integration + design-cycle lock-in, not scale.

• Proprietary process / packaging IP. Its BCD process and 3D packaging let it deliver higher-integration, higher-efficiency parts than discrete-component rivals — the filing leans on this repeatedly ("more highly integrated, smaller in size, more energy-efficient… more cost-effective"). Gross margin of 55%+ sustained for years is the quantitative proof the differentiation is real.
• Design-win switching costs. Analog power parts get designed into a board; re-qualifying a competitor is costly and slow. This is the durable moat — but it cuts both ways: when you lose the next socket (Blackwell), the installed base doesn't save you.
• Founder/technologist DNA. Hsing is a power-electronics inventor who has compounded the business for ~28 years; the org is built around analog design talent, which is genuinely scarce.

Where the moat is narrower than bulls claim:

• Bargaining power is asymmetric at the top. Nvidia needs power delivery; it does not need MPS specifically. Texas Instruments (massive 300mm scale), Infineon (high-power SiC), and Renesas are credible second sources. Reporting suggests Infineon could take a 60–70% majority of certain Blackwell power sockets, with business shifting to Renesas/Infineon. A moat that a $4T customer can route around at will is a position, not a fortress.
• vs. peers on perceived value: MPS is "best" on integrated-module elegance; TI is "better" on analog scale/capacity; Infineon is "better" on high-power SiC. MPS wins on the cutting edge of density (it pioneered 48V distribution — 4x lower current, 16x lower loss — and is sampling 800VDC for Rubin) but must keep re-winning each GPU generation.

Verdict on the moat: A-grade franchise moat in mainstream analog (sticky, high-margin, diversified), bolted to a B-grade position in the AI-GPU socket that the bull case is paying a fortune for.

Lens 4 · Segments

MPS reports one GAAP segment (the CODM/CEO reviews consolidated net income) — so there is no segment-level operating income or EBITDA disclosure. Segment economics by product line: n/a — single reportable segment, not disclosed. What is disclosed is revenue by end market, and the trend is the entire story.

FY revenue by end market ($000s):

The critical nuance: Enterprise Data (the AI/data-center socket) roughly doubled in 2024 (to $716M), then went flat/down −2% in 2025 — the FY2025 number masks a mid-cycle air-pocket that lines up exactly with the Nov-2024 Blackwell allocation scare. Then it re-accelerated violently into 2026:

Q1 2026 revenue by end market ($000s):

Enterprise Data is back to ~⅓ of revenue and is the marginal driver; Communications (optical modules, AI-adjacent) +55%. The legacy/diversification segments (Storage, Consumer) are flat-to-down. Translation: MPS is, at the margin, an AI-datacenter power-delivery company with a stable analog annuity underneath. Management raised the 2026 Enterprise Data growth outlook to ~85% YoY.

---

Phase B — Measure performance

Lens 5 · Earnings Result (latest print — Q1 2026, reported 2026-04-30)

The print:

• Revenue $804.2M, +26.1% YoY (vs $637.6M); beat the ~$782M consensus.
• Gross margin 55.3% (GAAP), essentially flat vs 55.4% Q1'25 — "decrease driven by higher warranty expense %, offset by lower manufacturing overhead %".
• Operating income $241.2M (30.0% margin) vs $168.8M (26.5%) — +360bps of operating leverage YoY, the real story. R&D fell to 12.4% of revenue (from 14.4%) and SG&A to 12.9% (from 14.5%); a big driver is stock-based comp falling to $41.1M from $52.8M.
• GAAP net income $193.2M; GAAP diluted EPS $3.92 (vs $2.81, +39%). Non-GAAP EPS $5.10, above expectations.
• Effective tax rate 21.8% (vs 22.3%), normal-looking — a relief after the 2024 tax distortions.
• Balance-sheet flags: inventories rose to $619.2M (from $564.6M at YE25) and AR to $302.1M (from $255.6M); operating cash flow $250.3M (roughly flat vs $256.4M). Inventory building ahead of the H2 ramp — consistent with management's bullish backlog commentary, but worth watching (16–26wk lead times mean they must pre-build).
• Guidance: Q2 2026 revenue $890–910M (~$900M midpoint, ~35% YoY), GAAP gross margin 55.1–55.7%. Management flagged potential "strong headwinds" to margins in H2 — the one cautious note in an otherwise euphoric call.
• Market reaction: shares closed +3.7% post-print. Tone shift = decisively more confident on AI/Enterprise Data backlog visibility into H2.

Unusual vs its own history: the Q1 operating-leverage jump (+360bps) and the SBC step-down are both favorable inflections; the inventory/AR build is the offsetting yellow flag.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts on disk (transcripts=0) — this lens is ``.

• Trajectory of tone: From the defensive crouch of late-2024/early-2025 (management forced onto a Nov-2024 call to rebut Edgewater, insisting "no performance issues… remain on Nvidia's parts list" ) → to confident re-acceleration in 2026 ("record quarter," Q2 guide to ~$900M, Enterprise Data outlook raised to ~85%).
• Recurring phrases: "balanced capital allocation" (dividends + buybacks + tech investment), "diversified growth across regions and product families," "above-industry growth," "gold standard" integrated modules.
• What they started saying: vertical power delivery, 800VDC, Vera Rubin readiness — the vocabulary moved from generic "enterprise data" to specific next-gen GPU power architecture.
• What they stopped emphasizing: the Blackwell-allocation defensiveness has faded from the narrative as Rubin optimism took over — convenient, and worth pressure-testing (Lens 13).

Lens 7 · Comps

Peer set: large/mid-cap analog & power semis. Multiples are ``, dated; gaps are marked n/a (never fabricated).

The one number that matters: MPWR's NTM EV/EBITDA ~49x vs peer median ~21x — roughly 2.3x the group, and richer than Nvidia (~17x) or Broadcom (~19x). On earnings, ~81x forward (2026E EPS $19.31) / ~112x trailing GAAP. MPS is the single most expensively-valued name in its entire peer complex, justified by bulls on the highest growth + best ROIC (43%) + cleanest AI exposure. The market is not mispricing the quality; it is pricing the growth as certain. (Industry-average analog forward P/E is ~28.6x — MPWR trades at roughly 3x that.)

Lens 8 · Stock-Price Catalysts (>5% moves, ~5yr)

Pattern is dominated by AI/Nvidia-socket headlines and earnings, not macro:

• Nov 11, 2024 / Jan 2025: ~−15% to −24.9% — Edgewater "Blackwell allocation at risk," "Nvidia canceled half of MPWR's backlog." The largest drawdown of the period and the defining bear catalyst.
• 2023–2025: +221% since institutions "bought big in 2023"; +135% off the 52-wk low of $671.18 into the 2026 AI surge.
• Aug 2024: prior 52-wk high ~$947.
• Q1 2026 earnings (Apr 30, 2026): +3.7% on the record print + $900M guide.
• Jun 16, 2026: −9.29% to $1,498.77 — sympathy selloff after Broadcom's soft forward AI-chip guidance (no raise to its AI forecast); MPWR has no company-specific news but trades as a high-beta AI-power proxy.
• Jun 19, 2026: +7.97% — AI-surge snap-back.

What the market actually reacts to: (1) Nvidia design-win/allocation signals (the dominant lever — a single research note moved it 25%), (2) AI-capex sentiment from the broader complex (Broadcom/Nvidia prints drag it ±8% with no MPWR news), (3) its own beat-and-raise cadence. It barely reacts to rates or its diversified/legacy segments. This is a momentum, single-theme stock — the diversification in the P&L is not what's driving the tape.

---

Phase C — Judge people & books

Lens 9 · Management

• Track record (elite). Michael Hsing, 66, founder-CEO since 1997, took MPS public in 2004 and compounded revenue at ~23.7% over 10 years / ~25.9% over 5 years, with a 5yr average ROIC of ~43% (3yr ~37%, TTM ~32%) — top-decile for the entire semiconductor industry. He is a power-electronics inventor running an analog-design org — exactly the archetype you want at this stage.
• Capital allocation (disciplined, shareholder-friendly). "Balanced": dividends + buybacks + reinvestment. Dividend grown ~26%/yr over 10 years with no cuts; in Feb 2026 the Board raised the quarterly dividend $1.56 → $2.00. A new $500M buyback was authorized Feb 2025 — but only ~$6.6M (≈8,000 shares) was actually repurchased in 2025 ($493.4M still available). Read: management did not buy back stock at $700–1,500/share — arguably valuation-disciplined (good), or simply preferring dividends; either way they are not torching cash on buybacks at 100x.
• Tenure & skin in the game. Affiliates (officers/directors/5%+ holders) held 14,146,000 shares excluded from the public float calc at Jun-2025 — meaningful founder/insider ownership. But: reporting notes continued insider selling with no reported purchases — normal for a founder diversifying at all-time highs, but a soft negative during a restatement/litigation window.
• Red flags (material, see Lens 10): (1) An unremediated material weakness in ICFR and an adverse auditor opinion on internal controls as of both Dec-2024 and Dec-2025 — happening on this team's watch. (2) CFO transition mid-crisis: Bernie Blegen (CFO since 2016) ceases as CFO immediately upon the 10-K filing, with Corporate Controller Rob Dean as interim CFO — losing your veteran CFO while a material weakness is open and a securities class action is live is a governance yellow-to-amber flag. (3) Eye-watering PSU economics: 2023 executive PSUs cost $156.2M in SBC; 2026 executive PSUs were struck off a $1,164.83 grant-date price with a $300/share purchase feature waived if the stock rose $300 — generous, performance-linked (revenue-growth-vs-SIA and TSR-vs-PHLX hurdles), but a large ongoing dilution/expense.
• Founder vs professional manager: Founder-technologist. Implication: deep product conviction and long-termism (good for a multi-GPU-generation R&D race), but key-man risk (the filing flags dependence on Hsing) and a control culture that just produced a controls failure.

Lens 10 · Forensic Red Flags

Acting as a forensic analyst. This is the lens where MPWR earns its skepticism — the accounting events here are unusually serious for a "quality compounder."

1. Restatement + material weakness + ADVERSE ICFR opinion (the headline).

• On Feb 26, 2026, the Audit Committee concluded the FY2024 10-K and all four 2025 10-Qs "should no longer be relied upon" due to an error in deferred-income-tax accounting tied to a one-time foreign tax incentive. The restatement decreased income-tax benefit and net deferred income taxes by $194.6M for FY2024.
• EY issued an ADVERSE opinion on internal control over financial reporting: MPS "has not maintained effective internal control… as of December 31, 2025," with a material weakness "in controls related to the Company's review of deferred income taxes" — and ICFR was also not effective at Dec-31-2024. The CEO and CFO concluded disclosure controls were not effective as of Dec-31-2025. The weakness is unremediated.
• Mitigant: it is a non-cash deferred-tax error, the FY2025 financials themselves carry an unqualified (clean) opinion, and management asserts the statements "fairly present" results. This is a controls/tax-accounting failure, not (on current evidence) revenue or cash fraud. But "the one part of the financials that blew up is the most complex, judgment-heavy area, run by a team that just lost its CFO" is exactly the setup a short-seller wants.

2. The $1.1B deferred-tax benefit that made 2024 look unreal.

• FY2024 GAAP net income of $1,592.1M (72.1% net margin!) was not operating performance — it was a $1.0B income-tax benefit (−177.9% effective rate) driven by recognizing a $1.1B deferred tax asset for a 10-year foreign tax incentive granted in 2024. Strip it out and 2024 pretax income was $572.9M. Any model using 2024 GAAP net income or EPS ($32.76 basic) as a base is corrupted — use operating income or non-GAAP. This is the single most important "don't get fooled" item in the file.
• Related and staggering: in Dec-2024 an intercompany transaction created a $23.2 BILLION step-up in tax basis of intangibles, against which MPS recorded a full $23.2B valuation allowance (it does not expect to realize the DTA for GAAP). Net P&L effect ≈ zero because of the full allowance — but a $23.2B intangible step-up at a company with $2.8B revenue is an aggressive tax-structuring posture that (a) invites scrutiny and (b) is precisely the kind of complexity that produced the restatement. EY flagged the realizability of the foreign tax incentive DTA as the sole Critical Audit Matter.

3. Income-statement / working-capital integrity (clean-ish).

• Revenue recognition is straightforward analog distribution with variable consideration (stock-rotation, price adjustments) — no long-term percentage-of-completion games.
• Receivables/inventory vs revenue: AR up to $302M and inventory up to $619M in Q1'26, both rising faster than the sequential revenue — a pre-build ahead of the H2 ramp, defensible given 16–26wk lead times, but the classic spot to watch if the AI order book softens (channel-stuffing risk lives in the distributor model). MPS recorded no AR write-offs and no allowance for credit losses — aggressive-looking zero, mitigated by standby LCs/advance payments on some customers.
• SBC flattering non-GAAP: FY2025 SBC was $227.5M (8.2% of revenue); the ~$5.10 Q1 non-GAAP EPS vs $3.92 GAAP gap is largely SBC. Standard for the sector, but the non-GAAP/GAAP wedge is wide — and the consensus "$19.31 2026 EPS" the bull case quotes is the non-GAAP figure; GAAP will be materially lower.
• Goodwill/intangibles: tiny ($25.9M goodwill, $8.8M acquisition intangibles) — no impairment risk there. The balance sheet's giant line is the $1,182.9M net deferred tax asset (28% of total assets) — the very thing that was misstated.

Regulatory findings (required sub-section).

• SEC EDGAR (LR + AAER): "No LR found" and "No AAER found" for Monolithic Power Systems since 2021-06-21. No SEC enforcement action to date.
• Item 3 (Legal Proceedings), FY2025 10-K: Beyond ordinary-course IP/employment matters, the material item is a securities class action — Waterford Twp. Gen. Emps. Ret. Sys. v. Monolithic Power Systems, Inc., et al., No. 25-cv-220 (W.D. Wash.), filed Feb 4, 2025, alleging §10(b)/§20(a) and Rule 10b-5 violations (material misstatements/omissions). Two related shareholder derivative suits were consolidated as Miller v. Hsing, et al., No. 25-cv-527 (W.D. Wash.), filed Mar 26, 2025, alleging breaches of fiduciary duty (the Derivative Litigation is stayed pending the Securities Action). MPS calls the suit "meritless" and intends to defend vigorously; management does not believe a material loss from known matters is probable as of Dec-31-2025.
• Non-SEC enforcement (web): No FTC/DOJ/FDA/CFPB consent decree, settlement, fine, or penalty surfaced in search. A plaintiff law firm announced an investigation into potential fiduciary breaches by the board following the restatement — investigation, not a charge.
• FCPA exposure (disclosed risk): MPS explicitly flags elevated FCPA/anti-corruption risk given "significant operations in Asia" and frequent contact with potential "foreign officials" — a standing risk, no known action.
• Summary: No SEC LR/AAER and no non-SEC enforcement to date — but an open securities class action + derivative suits, a board-fiduciary investigation, and an unremediated material weakness with an adverse ICFR opinion. For Lens 10 purposes this is a materially elevated forensic/governance risk profile, even though no fraud has been alleged or proven and no enforcement charge exists.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (FY2026 / FY2027 / FY2028 EPS)

Built bottom-up from the latest actuals + guidance; outputs are with arithmetic shown; consensus anchors are. No forecast.ts create logged (watchlist breadth rule).

Anchors: FY2025 revenue $2,790.5M (+26.4%); Q1'26 $804.2M (+26.1%); Q2'26 guide $900M (+35% YoY); Enterprise Data outlook ~+85% YoY '26; FY2025 op margin 26.1%, Q1'26 op margin 30.0%; FY2025 SBC 8.2% of rev; FY2025 non-GAAP-ish tax rate ~19–22%; diluted shares ~48.3M ('25) drifting up ~1–2%/yr on SBC. Consensus: 2026E non-GAAP EPS ~$19.31, 2027E ~$23.88.

FY2026 (base) — revenue: H1 ≈ Q1 $804M + Q2 ~$900M = $1,704M; if H2 grows modestly seq on AI ramp but with management's flagged margin headwinds, full-year ≈ $3.65–3.75B (~+31–34%). Operating margin holding ~28–29% on leverage; non-GAAP net margin ~30%. Base FY2026 non-GAAP EPS ≈ $19.0–19.5, in line with the $19.31 consensus → I adopt ~$19.30.

FY2027 (base): Enterprise Data decelerates from ~85% to ~30–40% as Rubin ramps but laps a huge 2026; total revenue $4.6–4.8B (+27%); modest further leverage → non-GAAP EPS ~$23.5–24.5, bracketing the $23.88 consensus → ~$24.

FY2028 (base): Normalizing to "great analog compounder + AI annuity," revenue $5.6–5.9B (+22%), EPS ~$28–30.

Scenario bands (FY2027 non-GAAP EPS):

• Bull ~$28+: MPS holds ~70% of the Vera Rubin power-stage socket, Enterprise Data compounds 50%+, op margin pushes 31–32%.
• Base ~$24: as above — wins Rubin share but shares the socket; H2'26 margin headwinds real but contained.
• Bear ~$15–17: Infineon/Renesas take the 60–70% Blackwell/Rubin majority that reporting warns of, Enterprise Data growth halves to ~30% and decelerates hard in '27, legacy stays flat → revenue ~$4.0B, op margin slips toward 26%. At a de-rated ~30x that is a sub-$500 stock.

The forecast that matters (binary, scoreable): "MPWR Enterprise Data revenue grows ≥60% YoY for FY2026" — p ≈ 0.70. (Logged conceptually as the calibration question; no forecast.ts write in watchlist mode.)

Lens 12 · Bull vs Bear

Bull case. MPS is the integration-and-efficiency layer of the AI buildout — it pioneered 48V (4x less current, 16x less loss) and is sampling 800VDC for Nvidia's Vera Rubin, where reporting pegs it as the primary power-stage winner with potential ~70% share. Underneath the AI rocket sits a genuinely elite analog annuity: 55%+ gross margin, 43% 5yr ROIC, ~24–26% revenue CAGR for a decade, +26%/yr dividend, founder-CEO with massive skin in the game, fabless capital efficiency. Enterprise Data nearly doubling (+97.7% in Q1'26) with the outlook raised to ~85% and Q2 guided to ~+35% says the AI socket is re-won, not lost. Operating leverage is inflecting (+360bps YoY in Q1). Earnings surprise potential is real because backlog visibility into H2 keeps improving. This is a secular compounder you hold for a decade, and quality this rare always looks expensive.

Bear case (2–3 permanent-impairment risks).

1. Nvidia socket loss is existential to the growth thesis. Enterprise Data is ~⅓ of revenue and the entire marginal story; it routes through a single end-customer (Nvidia) that can — and per Edgewater did in 2024 — cancel half the backlog overnight, with Infineon projected at 60–70% of certain Blackwell sockets. If MPS is a second source on Rubin rather than the primary, the bull EPS path breaks.
2. Valuation is the risk. ~49x NTM EV/EBITDA (vs ~21x peers, ~17x Nvidia), ~81x fwd / ~112x trailing earnings. The price requires ~30%+ growth to persist; a single deceleration print de-rates the multiple violently (the stock already round-tripped ±9% on Broadcom's guidance with no MPWR news).
3. Governance/controls overhang. Unremediated material weakness, adverse ICFR opinion, restatement, securities class action + derivative suits, CFO departing mid-crisis — any of which could force a confidence-cracking headline at a 100x multiple.

Pre-mortem (it's Dec-2027 and the thesis broke): Nvidia dual-sourced Rubin power to Infineon/Renesas; MPWR's Enterprise Data growth collapsed from 85% to ~15%; a second material weakness or an adverse class-action development hit during the deceleration; the multiple compressed from ~49x to ~22x EV/EBITDA. The stock is down 55–65% from $1,565 despite revenue still growing — a de-rating, not a blow-up.

Are multiples too high? Yes on any absolute or relative basis — MPWR is the most expensive name in its peer complex by ~2.3x. The quality justifies a premium; it does not justify this premium without flawless execution.

Contrarian view (what the market refuses to see): The market is treating the Vera Rubin "70% share" as a settled win while simultaneously knowing Nvidia already cut MPWR's Blackwell backlog and that Infineon is favored on the high-power sockets. The bull narrative quietly swapped "we kept Blackwell" for "we'll win Rubin" without the market re-pricing the demonstrated fragility. The contrarian read: MPS is a superb business whose AI-socket share is structurally contested every generation, and the stock is priced as if the socket is owned.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case.

• Where revenue is concentrated / what breaks it: ~33% of revenue (Enterprise Data) + a slug of Communications depends on the Nvidia GPU power socket, sold through 3 distributors = ~54% of total revenue. Two single points of failure stacked. If Nvidia shifts allocation (it has the leverage and the alternates — Infineon, Renesas, TI), the growth engine stalls. Edgewater's Nov-2024 note that "Nvidia canceled half of MPWR's backlog" is the proof of concept that this can and did happen.
• Why the moat is weaker than bulls think: It's a socket-by-socket design-win moat, not a franchise lock. You must re-win every GPU generation (Hopper→Blackwell→Rubin). Bulls underwrite a fortress; it's a title you defend annually against scaled competitors. Infineon at 60–70% of certain next-gen sockets is the bull's nightmare made concrete.
• Most dangerous competitor bulls underestimate: Infineon (high-power SiC + scale) for the highest-wattage GPU rails, and TI (300mm capacity, willing to price) for everything mainstream. Renesas is the dark horse Nvidia is reportedly steering business toward.
• Worst capital-allocation / accounting: A $23.2B intangible-basis step-up via intercompany transaction (fully reserved) is aggressive tax engineering; the $194.6M deferred-tax restatement + adverse ICFR opinion is the bill for that complexity. CFO exit during an open material weakness and a securities class action is the governance red flag short-sellers live for. Insider selling with no buys at all-time highs.
• Assumptions that must hold for today's price: (1) MPS is the primary Rubin power-stage supplier at ~70% share; (2) Enterprise Data compounds 50%+ multi-year; (3) gross margin holds ~55% despite management's own H2 headwind warning; (4) no further controls failures or adverse litigation; (5) the multiple stays ~49x EV/EBITDA. All five must hold.
• What a 20–30% growth disappointment does: On a ~49x multiple, a guide-down that resets growth from ~30% to ~10–15% plausibly halves the multiple and the estimates — a −50% to −65% move (cf. the −24.9% on a rumor in 2024).
• Single permanent-impairment scenario & plausibility: Nvidia designs MPWR out of the dominant Rubin power sockets in favor of Infineon/Renesas. Plausibility: moderate and demonstrated — not tail risk. This is the crux of the short.

Lens 14 · Management Questions (ordered by information value)

1. On the Vera Rubin power-delivery platform, what is your confirmed design-win share by socket (B-class vs GB-class vs Rubin), and is MPS the primary or a second source on the highest-wattage rails? (Answer most changes the thesis.)
2. After the Nov-2024 Blackwell allocation cut, how much of MPWR's Enterprise Data backlog today is confirmed POs vs unconfirmed/forecast, and what is the cancellation experience since?
3. What share of Enterprise Data revenue is Nvidia-attributable (directly + through distributors), and how concentrated is it within the top distributor (26% of total revenue)?
4. Material weakness: what specific remediation steps are underway on deferred-tax controls, and when do you expect to conclude ICFR is effective again?
5. Why is Bernie Blegen departing as CFO now, and what is the timeline + profile for a permanent CFO given the open weakness and litigation?
6. Defend the $23.2B intangible-basis step-up and the 10-year foreign tax incentive: what jurisdictions, what audit/BEPS-Pillar-Two exposure, and could it unwind?
7. Management flagged "strong headwinds" to gross margin in H2 2026 — quantify: mix, warranty, pricing, or tariffs? What's the trough?
8. With Infineon reportedly favored on 60–70% of certain high-power sockets, what is MPS's defensible technical edge at 800VDC that they can't replicate?
9. Capital allocation: you authorized $500M of buyback but used $6.6M in 2025 — is that valuation discipline (no buybacks above $X) or a deliberate dividend-over-buyback policy?
10. China foundry/manufacturing de-risking (the 2023 PSU goal): what % of wafers/assembly is now outside China, and what is the target?
11. Distributor concentration (3 = ~54% of revenue): what happens to revenue recognition and channel inventory if the top distributor's terms change?
12. Inventory rose to $619M and AR to $302M ahead of revenue — how much is committed/POed vs speculative pre-build for the H2 AI ramp?
13. Automotive (+43% in '25, EV-linked PSU goals): how exposed is the 2026 plan to the EV-demand slowdown, and what's the design-win pipeline?
14. Key-man risk: what is the succession plan for Michael Hsing, and how is the analog-design bench being deepened?
15. At ~112x trailing / ~49x EV/EBITDA, what growth durability do you believe the market is underwriting, and where do you think consensus is wrong?

---