Editorial Note
The 25-year supporting actor stepped into the lead role and the market is finally watching. MPWR is the "scarred and back with pricing power" story of the AI cycle — lost a quarter on Blackwell to Infineon's Vertical Power Delivery, recovered ~50% B300 HGX share by Q3 2025, and is now projected at ~70% Vera Rubin VRM socket share at +60% ASP. The catch: at $1,583 with RSI 75, +42% in 30 days, and a forward P/E of 52x vs a 5y average of 45x, the recovery is priced. KeyBanc's $1,250 PT (now $1,250, raised from $1,050 in Sep) is already through. The trade is no longer "is the comeback real" — it's "what's left after consensus has caught up."
The Story Right Now
The arithmetic of the last few centimeters is brutal: Blackwell B200 needs ~1,500A on the core rail, Vera Rubin pushes ~2,875A, and to hold 1,500A within 20mV requires a path resistance of ~13 micro-ohms. That's not a taste question, it's physics. NuttyCLD's AI Power Crisis series (six parts, March 30 → May 1) frames MPWR as the protagonist of a 25-year arc that just turned cinematic — the multi-phase VRM, the boring 12V → 0.8V step that nobody cared about for two decades, is now the choke point determining whether NVIDIA's most expensive silicon can actually run at spec. Nutty's intersection metaphor is the cleanest portfolio frame in the whole series: MPS sells the entire traffic-control system at the intersection. Not a single component, not just the controller — the integrated power module, the controller, the DrMOS, the qualification, the supply discipline. That's a different income statement than AOSL ("first into the lane") or Infineon+Delta ("underground entrance").
The lead-change arc is what makes this trade interesting and dangerous at the same time. Hopper had MPS comfortable and dominant. Then GB200 hit 700W+ and the PMIC thermal issues that already existed in Hopper became blocking — Infineon's Vertical Power Delivery (paired with Delta's system integration) won an estimated 60-70% of GB200 PMIC sockets in 2024, with Renesas filling more of the gap. MPS Blackwell exposure cratered, the sell-side repriced, and the stock spent most of 2024 in the doghouse. Then GB300 happened — KeyBanc's September 2025 channel checks said MPS had recovered ~50% B300 HGX share after fixing the thermal architecture, with $900M-cap AOSL emerging as co-leader via OpenVReg reference position. The PT went from $750-ish to $1,050 in one analyst note. By March 2026, KeyBanc raised again to $1,250 on supply-chain work suggesting ~70% Vera Rubin socket share at ASPs ~60% higher than Blackwell — annualizing into 2027 at roughly $420M of incremental revenue and material EPS lift.
Q1 2026 (reported April 30) confirmed the floor. EPS $5.10 vs $4.90 expected, revenue $804.2M vs $782M expected (+26% YoY), and management raised the enterprise data segment growth floor from 50% to 85% YoY — which is the data-center exposure ramping. Q2 guide implies sequential +12% to ~$900M. That print is the first hard read on the Vera Rubin socket-share thesis materializing in the income statement, and it landed clean.
The bottom-line take: the comeback is real, the pricing power is real, the architecture-rotation is real — and so is the 75 RSI. This is not an entry zone, it's a confirmation tape. The deep value moment was Sept 2025 at $750-ish when KeyBanc first raised. The momentum moment was October 2025 when MPS hit a new 52-week high on the KeyCorp upgrade. We're now in the third moment — "AI Power Architect at All-Time Highs" — where the Vera Rubin number is consensus and the asymmetry has compressed. The interesting trades from here are (a) sized starter on a pullback to RSI <55, (b) the more asymmetric peers further down the chain (AOSL, VICR, NVTS), and (c) the Murata MLCC adjacency that nobody talks about yet.
Quick Snapshot
| Signal |
Reading |
| Overall |
🟡 Confirmation tape, not entry — comeback priced, RSI 75, recovery thesis fully intact |
| Moat |
Wide (proprietary BCD process, 25-year design-win incumbency, NVIDIA reference position) |
| Key insight |
Sells the whole intersection — controller + DrMOS + module + qualification — not a discrete component. That's why ASP is +60% on Rubin. |
Action Matrix
| Action |
Level |
Why |
| Current |
🔍 Research / Watch |
Comeback story playing out as scripted; entry tape gone; wait for pullback |
| Entry Zone |
$1,505 - $1,580 |
Current -5%, near 20D SMA at $1,153 = full reset zone for size |
| Stop-Loss |
$1,267 (-20%) |
Below SMA50 ($961); breaks both the parabolic move and the recovery thesis |
| Target |
$2,058 (+30%) |
Above KeyBanc $1,250 (already in price); structural Rubin ramp scenario |
Price Data
| Stock |
Price |
7D |
30D |
3M |
1Y |
52wkHi |
RSI |
Status |
Action |
| MPWR |
$1,583.48 |
-0.3% |
+41.6% |
+35.2% |
+150% |
-4.7% |
75 ↑ |
🟢 Strong-up parabolic |
🔍 Watch — pullback only |
Legend
- RSI 75 = Overbought; FOMO warning
- 52wkHi = -4.7% from peak ($1,661.79 hit intraday April 2026)
- vs SMA20 = +37%, vs SMA50 = +65%, vs SMA200 = ~+90%+ — golden cross intact
Company Overview
One-Liner
Fabless analog/mixed-signal chipmaker that designs the power management ICs and integrated voltage regulator modules feeding everything from GPUs and EVs to consumer electronics and data-center AI sockets.
Business Model
| Question |
Answer |
| What they sell |
Power management ICs, multi-phase VRMs, DrMOS, integrated power modules — proprietary BCD process |
| Who pays |
Hyperscalers (via NVIDIA reference designs), OEMs, automotive Tier 1s, industrial, comms, consumer |
| Revenue model |
Component + module sales; ASP scales with integration depth and qualification difficulty |
| How sticky |
Very — design-wins last GPU generations; qualification is 12-18 months; switching mid-program is rare |
Key Segments (FY2025)
| Segment |
Revenue % |
Growth |
Notes |
| Enterprise Data (data center / AI) |
~30%+ |
50% → 85% guide |
The Vera Rubin engine; growth floor raised on Q1 call |
| Storage & Computing |
~20% |
Mid-teens |
Notebook, server, SSD power |
| Automotive |
~15% |
Slowing |
EV PMIC + ADAS — same 800V supply chain that AI inherited |
| Industrial / Comms / Consumer |
Balance |
Mixed |
Consumer cyclical, industrial firming |
Geographic Mix
- HQ: Kirkland, Washington; design centers in Asia, Europe, US
- ~4,500 employees worldwide; revenue heavily international (Taiwan/Korea/China customer base)
Competitive Analysis
Industry Position
| Question |
Answer |
| Market position |
Lead VRM vendor across multiple GPU generations; ~70% projected Rubin socket share |
| TAM (AI VRM/PMIC) |
Morgan Stanley: Kyber rack power-solution value rises >10x vs GB200 |
| Growth rate |
Enterprise data 85% YoY guide; total Q2 guide +12% sequential |
| Key competitors |
Infineon (IFNNY), AOSL, Vicor (VICR), Renesas (RNECY), Power Integrations (POWI), Navitas (NVTS), ADI, TI |
| Position |
Leader by socket share; protagonist of the multi-phase VRM lineage |
Competitive Moat
| Moat Type |
Present? |
Evidence |
| Proprietary process (BCD) |
🟢 Wide |
Hosted at third-party fabs; design IP locked; differentiated integration |
| Switching costs |
🟢 |
Each NVIDIA generation = 12-18mo qual cycle; mid-program switches are catastrophic |
| Reference design position |
🟢 |
Listed on NVIDIA's official 14-vendor 800VDC Kyber partner list AND holds the Rubin socket |
| R&D intensity |
🟢 |
17-19% of revenue on R&D — nearly 2x industry average |
| Scale advantage |
🟡 |
$79B mkt cap dwarfs AOSL ($900M); allows high-end qualification effort small competitors can't match |
Competitor Map (the five-company intersection)
| Company |
Role |
Risk to MPWR |
| IFNNY (Infineon) |
VPD via Delta partnership; won early Blackwell |
The architectural alternative — if VPD wins long-term, MPS multi-phase becomes legacy |
| AOSL |
OpenVReg co-leader on GB300; $900M cap |
Proves NVIDIA wants multi-source — caps MPS share at ~70%, not 100% |
| VICR (Vicor) |
Factorized architecture originator |
Backlog inflection Q4 2025; long-term IP overhang on the topology |
| POWI |
1,250V/1,700V PowiGaN direct conversion |
Topology rentier; could disintermediate if direct conversion ships at scale |
| NVTS (Navitas) |
800V→6V GaNFast at 96.5% efficiency |
Aggressive intermediate-bus path; threatens 12V incumbency |
| RNECY (Renesas) |
Filled the Blackwell gap |
Stable second source — eats some share each generation |
| ADI |
Maxim-acquired analog portfolio |
Broad incumbent; less concentrated AI exposure but defensive |
Moat Assessment
Moat Width: Wide
Moat Trend: Widening (Rubin socket share gain at higher ASP) but architecture-cyclical — the moat resets every NVIDIA generation. Nutty's framing is exactly right: this is not yet a market with a final winner. It is a market where the definition of winning keeps changing from one generation to the next.
The summary in two sentences: MPS has the deepest 25-year supply chain in multi-phase VRMs and the strongest customer relationship at NVIDIA — that's the moat. But every architectural transition (Hopper → Blackwell → Rubin → Feynman) is a new audition, and 2024 proved the moat doesn't make you immune to losing a generation.
Management Assessment
Leadership
| Role |
Name |
Since |
Background |
Notes |
| Chairman, President & CEO |
Michael R. Hsing |
Aug 1997 (founder) |
EE BS, UC Berkeley; senior silicon technology developer at multiple analog IC firms before founding MPS |
28-year tenure, 2.05% direct ownership (~$707M), known for "innovation-first" 17-19% R&D budget vs industry ~10% |
| CFO |
Bernie Blegen |
(long-tenured) |
Multi-cycle finance leader |
Stable; not a turnover signal |
Founder Involvement
| Question |
Answer |
| Founder-led? |
Yes — Hsing has been CEO since IPO and founding |
| Founder ownership |
2.05% direct (~$707M at $1,583) |
| Skin in the game |
High; rare combination of founder-led + S&P 500 + 28-year tenure |
Capital Allocation
| Metric |
Track Record |
| M&A discipline |
Conservative — organic growth model, no big-ticket acquisitions |
| Buyback timing |
Modest, dividends growing (28% Q4 2025 div hike) |
| R&D investment |
Very High (17-19% of revenue, ~2x industry) |
| Debt management |
Conservative — zero long-term debt, $3.5B equity, debt/equity = 0 |
Red Flags
- Excessive exec compensation — not flagged
- High turnover in key roles — no
- Insider activity bearish — leading indicator shows -$751.9M net insider value over trailing window (mostly automated 10b5-1 sales into the parabolic move; not a panic signal but worth flagging)
- Aggressive accounting — no
Financials
Key Metrics
| Metric |
Q1 2026 |
Q1 2025 |
Q1 2024 |
Trend |
| Revenue |
$804.2M (Q1 actual) |
$637.5M |
$457.9M |
📈 |
| Revenue Growth YoY |
+26% |
+39% |
— |
Reaccelerating into Rubin |
| Gross Margin |
~55% |
55.4% |
55.1% |
📈 Stable mid-50s — exceptional for analog |
| Operating Margin |
~26% |
26.5% |
20.9% |
📈 |
| Net Margin |
~22% |
21.2% |
20.2% |
📈 |
| FCF Margin |
12.4% |
— |
— |
🟡 Compressed (capex into Rubin ramp) |
Quality Checks
| Check |
Status |
Notes |
| FCF positive? |
🟢 |
$93M Q1; full-year normalizes higher |
| Profitable? |
🟢 |
22%+ net margin |
| Debt manageable? |
🟢🟢 |
Zero long-term debt |
| Cash runway |
n/a |
Net cash position; quarterly $2/share dividend funded organically |
Leading Indicators (current cached read)
| Indicator |
Value |
Signal |
| Revenue Acceleration |
-9.0% |
🔴 (sequential noise; the QoQ in 2025 was lumpy — the trailing read overweights the early-2025 rip) |
| YoY Growth |
+20.8% (trailing) → +26% Q1 actual |
🟢 |
| Earnings Beat Rate |
100% |
🟢🟢 |
| Insider Trading |
-$751.9M net |
🔴 Bearish (10b5-1 cluster on the parabolic) |
| Composite |
-0.2 |
🟡 Neutral — earnings + revenue overcome the insider tape |
Valuation
Current Multiples (consensus)
| Metric |
Current |
5Y Avg |
Industry Avg |
| P/E (TTM) |
~87x |
~50x |
~46x |
| Forward P/E (2026) |
~52x |
~45x |
~30x |
| 2026E EPS |
$21.94 (mid) |
— |
— |
| Implied 2026 P/E at $1,583 |
~72x |
— |
— |
| 2027E EPS (with Rubin ramp) |
step function higher |
— |
— |
Analyst Targets (May 2026)
| Source |
PT |
Spread to current |
| KeyBanc (Sep 2025) |
$1,050 |
already through |
| KeyBanc (latest) |
$1,250 |
-21% |
| Wall St avg (12 analysts) |
$1,294 |
-18% |
| Wall St high |
$1,500 |
-5% |
| Rosenblatt |
$1,575 |
flat |
| Wall St low |
$1,000 |
-37% |
Fair Value Take
The dispersion is the tell. Consensus PT of $1,294 is below current price. The bull case ($1,500-1,575) is roughly flat. The bear case ($1,000) is the "Q1 was the peak, repricing follows" scenario. Today's price is already pricing the Rubin upside — KeyBanc's $420M of incremental 2027 revenue is in the multiple.
My fair value zone: $1,200-$1,400 for a starter ("price reflects 70% Rubin / +60% ASP, no further upside"); $1,600-$2,000 if Rubin ASP exceeds 60% premium AND Feynman socket position holds (the latter not knowable until 2027).
Bull Case
Rubin socket-share dominance is incumbency, not luck — once you're in spec at 1,500A+ within 13 micro-ohms, switching costs the customer a year of qualification. The ~70% socket projection is a long-cycle asset.
- Evidence: NVIDIA listed MPS on the 14-vendor 800VDC Kyber partner list; KeyBanc supply-chain work; Q1 enterprise data segment +85% YoY guide.
- Implication: Rubin revenue contribution is multi-quarter, multi-year — not a one-quarter pop.
ASP expansion >> unit growth — the +60% ASP on Rubin VRM is structural, not promotional. More integration (controller + DrMOS + module), more current per socket, more validation cost embedded.
- Evidence: Morgan Stanley estimates Kyber rack power-solution value rises >10x vs GB200.
- Implication: revenue and gross-margin trajectory both improve; analog companies with widening GM are rare.
Cross-cycle exposure (EV + AI) — the 800V SiC + GaN supply chain that EV built (Porsche/BYD/Hyundai) is now AI's second demand curve. MPS plays both ends.
- Evidence: BYD 2.11M EVs 1H 2025; auto segment provides ballast even when AI sentiment cools.
- Implication: Wolfspeed/SiC restructuring story has a second floor MPS sits on top of.
Fortress balance sheet, founder-led, decades-long compounder DNA — 17-19% R&D, zero LTD, founder-CEO 28 years in, 2.05% direct ownership.
- Implication: durable through the next architectural cycle whether MPS wins Feynman socket share or not.
Upside Scenario
| If This Happens |
Stock Could |
| Rubin ramps 2H 2026 with confirmed ASP +60% |
$1,800-$2,000 (+15-25%) |
| Feynman 2028 socket position locked + A16 BSPDN compatibility |
$2,200+ (multi-year) |
| Auto SiC re-acceleration on EV demand recovery |
adds $200 to fair value |
Bear Case
Priced for perfection at RSI 75 / forward P/E 52x — Q1 2026 confirmed the thesis, but the Rubin number is consensus. A miss on Q2 guide ($900M) or any softness on enterprise-data ramp triggers repricing on a stock that's +150% in 12 months.
- How it plays: Q3 2026 earnings disappoints OR competitive update from AOSL/Infineon shifts socket-share narrative; stock retraces to $1,000-1,200 (-25-35%).
- Probability: Medium
Rubin slips to 2027 from late 2026 — production timelines on advanced platforms have slipped repeatedly. NVIDIA Feynman is 2028; Rubin "delayed" is a sentence the market would hate.
Architectural disintermediation (VPD or direct conversion) — Infineon+Delta VPD won the previous cycle. Power Integrations 1,250V/1,700V PowiGaN at 800V input direct-to-1V is the long-term threat. If direct conversion ships at scale on Feynman, MPS becomes the next legacy story.
- Probability: Low near-term, Medium long-term
Insider selling cluster (-$751.9M net) signals top — automated 10b5-1 plans don't usually mean panic, but the cluster size at the parabolic is the pattern that historically precedes 20-30% drawdowns in single names.
- Probability: Medium for a tactical drawdown; doesn't break the thesis
Auto cyclical drag — auto segment growth slowing; if EV demand cools in 2H 2026 the second-curve narrative weakens.
Thesis Killers
- AOSL or Infineon takes >50% Vera Rubin socket share (would invalidate the KeyBanc thesis)
- Feynman 2028 socket position goes to a competitor (long-cycle bear)
- Operating margin compression below 22% (would signal commoditization)
Downside Scenario
| If This Happens |
Stock Could |
| Q2/Q3 2026 disappoint; sentiment unwind |
$1,100-1,200 (-25%) |
| AOSL/Infineon socket-share gain on Rubin |
$900-1,000 (-40%) |
| Direct-conversion topology wins Feynman |
$700-800 long-term (-50%+) |
Market-Moving News
Historical (the lead-change arc)
| Date |
Event |
Impact |
Lesson |
| 2023-2024 |
Hopper PMIC thermal issues → Blackwell lost to Infineon VPD |
-40%+ drawdown |
The moat doesn't make you immune to architectural disruption |
| Sep 2025 |
KeyBanc $1,050 PT on B300 HGX recovery (~50% share) |
Reset higher |
Half a generation back is a re-rating event |
| Oct 2025 |
KeyCorp upgrade; new 52-week high |
+large daily move |
Sell-side amplification |
| Dec 2025 |
Added to NASDAQ-100 |
Index flow tailwind |
Passive demand kicks in |
| Mar 2026 |
KeyBanc $1,250; Vera Rubin ~70% socket / +60% ASP |
Re-rate higher |
The "scarred and back with pricing power" line |
| Apr 30, 2026 |
Q1 EPS $5.10 vs $4.90; revenue $804M vs $782M; raised enterprise data guide 50% → 85% YoY |
Beat-and-raise |
Guide raise > headline beat |
Recent News (cached)
| Date |
Headline |
Source |
Impact |
Relevance |
| 2026-04-10 |
"3 Companies Aggressively Raising Dividends" |
Investing.com |
🟢 |
28% div hike narrative |
| 2026-03-25 |
"A 28% Dividend Hike From the AI Stock You've Never Heard Of" |
Investing.com |
🟢 |
The "stealth AI" angle |
| 2025-12-13 |
Added to Nasdaq-100 Index |
GlobeNewswire |
🟢 |
Index inclusion = passive bid |
| 2025-12-12 |
Q4 2025 dividend declared |
Benzinga |
🟢 |
Cash-return story |
News Patterns
What moves MPWR: (1) NVIDIA architectural transition signals (any leak/note about socket share), (2) sell-side channel checks (KeyBanc/Barclays/Morgan Stanley), (3) NVDA results / GTC announcements. Sentiment overshoots in both directions on socket-share narrative changes — that's the whole 18-month story.
| Symbol |
Name |
Relationship |
Status |
| SOXX |
iShares Semiconductor ETF |
Holds MPWR |
🟢 Strong-up; semis parabolic |
| SMH |
VanEck Semi ETF |
Holds MPWR |
🟢 At all-time highs |
| FTEC |
Fidelity MSCI Tech |
Tech sector |
🟢 RSI 72 stretched |
| MGK |
Vanguard Mega Growth |
Mega-cap growth |
🟢 Stretched |
| PSI |
Invesco Dynamic Semi |
Semis |
🟢 |
Sector Context
Whole semis basket is parabolic (per today's full-scan brief: SOXL +50% in 30 days, INTC RSI 87, AMD RSI 79, MRVL RSI 76). MPWR is participating in a sector-wide blowoff, not a stand-alone breakout. Sector rolling over = MPWR rolls with it.
Cross-References (Where This Appears in Our System)
| Location |
File |
Context |
| Perspectives |
AI Power Delivery |
Primary catalyst — listed as the "intersection" winner; ~70% Rubin / +60% ASP |
| Perspectives |
AI Power Bottleneck |
Parent perspective (delivery is child of generation) |
| Inputs |
inputs/articles/2026-04-19-nuttycld-ai-power-crisis-part-3-last-few-centimeters.md |
The lead-change arc (Hopper → Blackwell → GB300 → Rubin) |
| Inputs |
inputs/articles/2026-04-28-nuttycld-economics-of-the-last-1cm.md |
The five-company intersection metaphor |
| Journal |
the May 1 full-scan market brief |
AI Power Delivery section — "coverage gap" call-out |
| Watchlists |
none yet |
Coverage gap — needs to be added to ai-power.json or new vrm-pmic.json |
| Holdings |
not held |
Not in any paper or real account as of 2026-05-01 |
| Ticker |
Why Related |
Suggested Placement |
Priority |
| AOSL |
OpenVReg co-leader on GB300; $900M cap = highest asymmetry |
new vrm-pmic watchlist |
🟢 High |
| VICR |
Factorized architecture originator; $176.9M record backlog |
new vrm-pmic watchlist |
🟢 High |
| IFNNY |
Infineon — VPD competitor; current Blackwell winner |
new vrm-pmic watchlist |
🟢 High |
| NVTS |
Navitas — 800V→6V GaNFast; intermediate-bus disruption |
new vrm-pmic watchlist |
🟡 Med |
| POWI |
Power Integrations — 1,250V/1,700V PowiGaN direct conv |
new vrm-pmic watchlist |
🟡 Med |
| BESI |
Hybrid bonding for HBM + BSPDN — same physics frame |
already in supply-chain-traces |
🟢 confirmed |
| MRAAY |
Murata — MLCC bottleneck (GB300 uses ~30K MLCCs) |
new mlcc watchlist |
🟡 Med |
| ON |
onsemi — broader power portfolio |
semis |
already tracked |
| STM |
STMicro — analog power, 800VDC partner |
semis |
already tracked |
| NVDA |
The customer — every architectural shift is downstream of NVIDIA |
already focus |
🟢 |
| AVGO |
Custom silicon adjacency |
already tracked |
— |
| INTC |
18A PowerVia BSPDN — die-level power |
already tracked |
— |
| RNECY |
Renesas — filled the Blackwell gap |
new vrm-pmic watchlist |
🟡 Med |
Catalysts & Timing
Upcoming
| Date |
Event |
Impact |
Watch For |
| Jul/Aug 2026 |
Q2 2026 earnings |
🟢/🔴 |
$900M revenue guide; enterprise data growth pace; Vera Rubin commentary |
| Oct/Nov 2026 |
Q3 2026 earnings |
🟢/🔴 |
First quarter with material Rubin shipment commentary |
| Late 2026 / Early 2027 |
NVIDIA Vera Rubin volume ramp |
🟢🟢 |
The thesis-confirming print |
| 2027-2028 |
NVIDIA Feynman socket-share signals |
🟢/🔴 |
Multi-year thesis renewal or break |
Key Dates
| Event |
Frequency |
Notes |
| Earnings |
Quarterly |
Beat-rate 100% trailing |
| Dividend |
Quarterly ($2.00) |
Q4 2025 hiked 28% |
Entry Strategy
Position Sizing
| Conviction |
Allocation |
| Speculative |
— |
| Low |
1-2% |
| Medium (current) |
3-5% |
| High |
5-10% |
My conviction: Medium — thesis is real, comeback is real, valuation is full at the entry tape.
Target allocation: 2-3% starter on pullback, scale up if/when Rubin ramp confirms.
Entry Approach
| Strategy |
Details |
| Entry zone |
$1,505 - $1,580 (current -5% / RSI mean-reversion) |
| Better entry |
<$1,300 on broader semis pullback (full reset) |
| Starter position |
1/3 of target on RSI <55 |
| Add on |
Rubin ramp confirmation (Q3 2026 earnings) |
| Full position at |
Confirmed Rubin socket share + holding margin |
Technical Levels
| Level |
Price |
Notes |
| Resistance / 52wk high |
$1,661.79 |
-4.7% from current |
| Current |
$1,583.48 |
RSI 75 |
| 20-day SMA |
$1,153 |
-27% — would be deep pullback zone |
| 50-day SMA |
$961 |
The bear-case reset level |
| Stop-loss |
$1,267 |
-20% from current |
Sources
Research Log
| Date |
Update |
| 2026-05-01 |
Created deep dive — triggered by NuttyCLD AI Power Crisis series; new ai-power-delivery perspective |
The Gold
Key Discoveries
| Discovery |
Implication |
| MPS sells the entire intersection — controller + DrMOS + module + qualification — that's why ASP +60% on Rubin |
Income statement asymmetry vs single-component peers (AOSL has position; MPS has integration) |
| Architecture-cyclical moat — every NVIDIA generation is a new audition |
The "socket war" repeats every 18-24 months; we should re-evaluate the thesis at each transition, not assume incumbency |
| KeyBanc $1,250 PT is already in the price |
The asymmetry now lives in the peers (AOSL, VICR, NVTS) and adjacencies (MRAAY MLCC), not in MPWR itself |
| Founder-CEO 28 years + zero LTD + 17-19% R&D |
Compounder DNA in a cyclical industry — that's rare |
Open Questions
- How much of the Q1 +85% enterprise-data guide is Rubin pre-build vs Blackwell sustain?
- What's the Feynman 2028 socket-share read? (TSMC A16 Super Power Rail interaction with VRM topology — does BSPDN make MPS more or less relevant?)
- AOSL co-leader trajectory on Rubin — does the $900M-cap small-co retain reference position, or does NVIDIA consolidate to MPS as Rubin matures?
- Should we open a
vrm-pmic watchlist (MPWR, AOSL, VICR, IFNNY, NVTS, POWI, RNECY, ON, ADI, TXN) — Nutty's open question, still open
- The -$751.9M insider net — does a parsed view (10b5-1 vs discretionary) change the signal?