Siphonophore

15/ Summary: Know which layer you're in. Know when to rotate. Software → zero (happening now) Hardware → scarce window, then commoditized (trade it while it's open) Fleet data → the consensus endgame that will be expensive by the time you can buy it cleanly The crowd will be right about the destination. They'll still lose money getting there.

14/ The honest framework: - Endgame: Fleet operators win. Data moat compounds. Consensus is right. - Problem: Untradeable today. Priced on arrival when it becomes tradeable. Therefore: The hardware transition window is the trade — not because hardware wins forever, but because it's the only layer with mispriced near-term catalysts and a clear rotation clock.

🧵 The AI robotics trade everyone is getting wrong. And I mean everyone — bulls and bears. 1/15 2/ The debate raging in AI right now: LeCun's JEPA world models vs. brute-force LLM scaling for robotics. LeCun says generative token prediction is fundamentally wrong for embodied AI. He's directionally right. But that's not where you make money.

Quite a run so far! Higher ->>

$AIXTRON ($AIXA.DE) – German MOCVD leader for GaN/SiC/GaAs compound semis. Thesis: Can power electronics tailwinds (EVs, AI data centers, renewables) + micro-LED/photonics upside drive growth through the cyclical trough and competition? Key checkpoints to watch: • Revenue back to double-digit YoY growth (SiC/GaN orders leading) • Gross margins holding >40% (pricing + tech moat) • Book-to-bill consistently >1.0 (demand visibility) • >50% MOCVD market share defended vs Veeco & Chinese players Bull drivers: • SiC/GaN power boom – EV/industrial/AI DC electrification (high conviction) • Micro-LED catalyst if Apple/Samsung scale displays • Asset-light model = big operating leverage on rev ramps • EU Chips Act subsidies + reshoring sentiment Quick business flow: End markets (EVs, 5G, data centers, LEDs, lasers) → customers (Wolfspeed, Infineon, STMicro, onsemi, Chinese fabs) → AIXTRON tools (R&D → design → build → install/service) → revenue (~85% systems, 15% service) Moat: Deep MOCVD/CVD process expertise in wide-bandgap materials. Switching costs sky-high after tool qualification. Risk/reward snapshot: • SiC growth: Bull 23–26%+ CAGR to 2030 • GaN adoption: Bull mainstream RF/power ICs • China (~30–40% rev): Bull big contributor • Cyclical trough: Bull 2025 priced in • Valuation: Bull below historical medians Bottom line: High-quality niche equipment play with genuine secular drivers, currently in a trough priced like growth is dead. I think the upcycle is coming. Patient conviction bet, not low-vol safe. Not advice – DYOR. #Semiconductors #SiC #GaN #AIXTRON #AIEquity

NFA DYOR

Foxconn's AI server segment is growing 30%+ YoY — and the margins are nearly 2x their legacy business. Key numbers on $2317.TW: 📊 AI/server revenue: ~25-30% of total (up from 15% two years ago) 📊 AI segment margins: 6-7% vs. 3-5% historical average 📊 Current valuation: ~10-12x forward earnings 📊 Target re-rating: 14-15x (30-40% upside) What makes this different from generic contract manufacturing: • Vertical integration into liquid cooling & power systems = higher BOM capture • Only company with balance sheet to finance 8,000-lb server racks at Nvidia-scale volume • Mexico/Vietnam expansion = built-in tariff hedge The May 2026 Q1 earnings print is the catalyst. Watch for: → AI server revenue growth confirmation (need ≥25% YoY) → Sequential margin improvement → Free cash flow stability (working capital strain is the hidden risk) Kill switches: Nvidia in-house assembly, AI server growth <25%, or inventory expansion for two consecutive quarters. This is a secular AI infrastructure play hiding in a "boring" hardware stock.

Violent divergence in robotics right now. #Palantir and #Nvidia: holding. Harmonic Drive (6324.T) and #Nabtesco (6268.T) — the Japanese companies actually manufacturing the joints and reduction gears inside humanoid robots — down 6% on the day and 8% over two weeks. The market is doing a BOM reset. Mass-market humanoids need MRL 7+ unit economics. Current valuations were priced for prototypes, not factory floors. The "brains" are still flying. The "muscles" are getting crushed. Watch which side is right.

Everyone is paying 40-60x earnings for AI chip designers. Foxconn builds the actual server racks those chips go into — Nvidia's largest manufacturing partner, 1,000 AI racks per week, OpenAI and SoftBank Stargate in the backlog — and it trades at ~15x P/E. FY2025: record revenue (NT$8.1T, +18% YoY), record EPS, record dividend. January 2026 revenue already up 35.5% YoY. Chairman guiding double-digit growth for the full year. ALL while the stock is down 23% from the highs. Cloud overtook consumer electronics as the #1 revenue segment. The business has fundamentally changed. The valuation hasn't caught up yet. $HNHPF / 2317.TW — the AI infrastructure play the market forgot to reprice. no investment advice