Cal Conviction

$NBIS vs $IREN vs $CRWV: CRWV is cheapest on forward sales. NBIS has the best proof. IREN has the fattest upside tail but weakest evidence. The neocloud trade is not "which one has AI vibes?" It’s valuation vs financing risk vs proof of conversion.

@ren_stocks Warning sign when everyone and their moms are buying?

Seems like even my mom is following Leopold’s investing strategy. Except a simplified version. 50% $NBIS 50% $DRAM She called me today to help her understand CSP. Bruh… I wish I was making this up, good for her I just don’t do CSP.

$IQE deep dive: FY2025 revenue was about GBP 97M, down 17.6% YoY. Photonics was GBP 57.1M, up 15%, while wireless was GBP 40.1M, down 40%. The latest yfinance snapshot had GBP 97.3M TTM revenue, 1.39% gross margin, negative EBITDA around GBP 8.2M, GBP 15.7M cash and GBP 91.7M debt. The thesis is upstream photonics scarcity. IQE makes compound-semiconductor epiwafers/materials, including InP/GaN/GaAs, not the sexy final optical module. If AI data centers keep pushing 800G/1.6T optics and future co-packaged optics, the chokepoint can move upstream to hard-to-replicate wafer and epitaxy capacity. The new catalyst is Tower Semiconductor. IQE signed a multi-year InP epiwafer supply agreement for AI-driven optical connectivity, including Tower's 200Gb/s-per-lane pluggable platform, next-gen 400Gb/s-per-lane modulator prototyping, and optical-circuit-switch work. It includes a year-one minimum purchase commitment and later minimum volume commitments. That is way better than "trust me bro, AI optics." Second catalyst: MACOM's strategic investment and long-term supply agreements tied to the GBP 81M fundraising. That cleaned up some balance-sheet stress and validated the capacity. Third: the old consensus path was only about GBP 117M FY2026 revenue and GBP 140M FY2027 revenue, so any real AI/datacenter mix disclosure or guide raise would matter. Risks are not cute. The stock has already ripped from distressed levels to 57.71GBX. Fully diluted market cap is roughly GBP 750M on sub-GBP 100M revenue and barely-there margins. New shares/convertibles diluted holders. HSBC covenant waiver history says the old balance sheet was fragile. And strategic customers can secure supply without giving IQE huge pricing power. Entry/Verdict: CATALYST WATCH / optional tiny scout only, not a normal buy. I would not add size here after the vertical rerate. Clean entry is a pullback toward/under 30p with the thesis intact, or paying up only after Tower/MACOM revenue, backlog, guide, gross margin, or estimate revisions actually move. This is real proof, not just photonics basket noise. But the market is already asking IQE to become strategic scarcity, not a cyclical wafer supplier with a cool press release. We'll see how it goes.

The agent stack is not just models. It is data, monitoring, permissions and messy human context. That is where I want proof: NRR, RPO, usage revenue, margins and estimate revisions. If those do not move, the story is just AI wallpaper.

3. $RDDT — Reddit The weird AI-data angle is human context. Q1 rev was USD 663M, +69% YoY; ad rev +74%; adj EBITDA USD 266M and FCF USD 311M. Not cheap after the rebound. Proof gate is Reddit Max/Shopify/data deals moving estimates, not vibes.

3 agent-infrastructure names I'm watching Agents need observability, governed data and human context. Boring, useful stuff. 1. $DDOG — Datadog Monitoring gets harder when apps become model calls. TTM rev is USD 3.67B, +32% growth. Proof gate: AI usage turning into NRR/margin.

Changed this week: CEG moved to buy-small common after the FERC Crane waiver, VRT briefly hit a cautious add zone but is back above it, and ONDS/IQE were added to the tiny catalyst-watch research pile. BWA stays removed. $VST @ $150.57 — 🔴 ENTRY ZONE. Cleanest liquid AI-power/nuclear scarcity play. Entry: common add only around $146-$151 with thesis intact; no fresh calls. MRAM @ $27.49 — Hold existing sleeve only. Product growth, defense agreement, and Microchip capacity path are real, but Kerrisdale/insider-sale/GM risk block add-size. Entry: do not add before Q2 proof. CEG @ $260.00 — 🟡 APPROACHING / BUY SMALL common. Nuclear plus Calpine AI-power platform; FERC Crane waiver de-risks the setup. Entry: common preferred $240-$255, 0.5%-1.0% starter if underweight; no new calls. MNDY @ $80.67 — Software reset with real Q1 proof, but AI monetization/NDR still need receipts. Entry: $86-$90 / $79.50-$83 only if Q2 ARR, NDR, and AI proof stay intact. PDFS @ $65.46 — Semiconductor analytics proof is real, but valuation/FCF leave thin urgency. Entry: $35-$38 if thesis intact, or after Q2 revenue/RPO/eProbe proof. CAMT @ $198.10 — Real HBM/advanced-packaging inspection winner. Entry: no auto-add; wait for 2H revenue conversion, order/backlog strength, and margin recovery. ONTO @ $337.95 — Quality metrology name, but price is through the old proof zone. Entry: $220-$239 if thesis intact or after FY26/Dragonfly shipment proof. POET @ $13.24 — Photonics option, not earnings yet. Entry: tiny tracker only; add-size waits for module qualification, payment-backed PO conversion, and 2027 deployment proof. OSS @ $17.53 — Defense/edge-compute proof is improving. Entry: no chase; Q2 must show GM above 40%, book-to-bill above 1.2x, and positive adjusted EBITDA. OUST @ $46.47 — Physical AI perception play with real revenue growth. Entry: keep starter; add only after Q2 guide beat, margin stability, and narrower EBITDA losses. RDW @ $15.77 — Space-defense order proof is real, but dilution/EBITDA risk remains. Entry: no chase; wait for backlog conversion, GM above 25%, and cash burn narrowing. VPG @ $141.40 — Robotics sensor proof, but the stock already did the fun part. Entry: no chase; needs Q2 beat/raise and sensors conversion. KEEL @ $5.87 — AI/HPC powered-site optionality. Entry: no chase; needs named lease with counterparty, MW, term, economics, and financing clarity. AEHR @ $122.70 — AI/SiPh burn-in signal is real, but valuation is still science-project-ish. Entry: tiny tracker only; add after FY2027 guide/order conversion resets estimates. FLNC @ $24.23 — AI data-center design validation is useful, not a purchase order. Entry: no add; wait for hyperscaler order/backlog/economics proof. RDDT @ $173.53 — Q1 proof is excellent, but current sleeve is enough. Entry: add only on $148-$155 / $154-$163 support or after Q2 beat/raise. NVTS @ $24.89 — 800V GaN/SiC optionality, but narrative is ahead of backlog. Entry: token only; proof-gated add $20.50-$23.80, better $16-$17.65. BE @ $271.36 — Real AI-power leader, still no chase. Entry: first look $274-$285, preferred $220-$244 reset, or new order/guide proof. VRT @ $311.50 — Quality 800VDC compounder; the $287-ish tactical zone came and went fast. Entry: cautious only on $275-$307 retest, better $242-$275; no calls. AEIS @ $367.84 — Real AI PSU beneficiary, but no named hyperscaler backlog yet. Entry: optional tiny tracker $288-$304, better reset $253-$260. NBIS @ $250.24 — Event validated, but neocloud concentration risk is not cute. Entry: do not add here; $221-$208 intact reset or new financing/customer proof. PENG @ $64.03 — Five AI/HPC customers and MemoryAI proof are real, but spot is above the average target. Entry: token only; 0.5%-1% on $44-$52 reset or Q3 order/backlog/estimate proof. VLN @ $2.42 — Connectivity semi with automotive/robotics optionality. Entry: tiny scout only after Q2/Q3 revenue guide, GM, and OEM production proof. AMPG @ $7.94 — Micro-cap RF/5G O-RAN/quantum-LNA watch. Entry: optional tiny scout only; add after Q2/Q3 revenue, backlog, MNO conversion, and GM proof. LPTH @ $14.81 — Backlog proof is real, but valuation and customer opacity leave thin asymmetry. Entry: first look $15, better $13.90-$14 / $13.10. Highest conviction today: VST common in the red entry zone, with CEG the cleaner new buy-small change if it gives a better $240-$255 entry. We'll see how it goes.

My core holdings right now: AI infra / neocloud — $NBIS, IREN, NVDA Fintech / software — HOOD, NOW, CRWD, RDDT Memory — MU, DRAM AI power / grid — VST, CEG, VRT, FLNC Photonics / optical — MRVL, LITE, GLW, COHR Metals hedge — GLD, SLV calls Changes since last portfolio post: Added GLD/SLV call basket. Added/pressed some NVDA, NBIS, META, NOW, AMZN/VRT call exposure. Trimmed IREN. SG is out. SOXL/SOXS are not core — just tactical trading junk, not a thesis. Most of the real research here is still around AI infra bottlenecks: neocloud power, memory, photonics, and grid capacity. But the actual book is broader than this account makes it look: software/fintech and a metals hedge matter too. The honest state: winners are working, but options/cash leverage is high enough that I'm not trying to broad-add into green tape. New money has to earn it with proof or a pullback. We'll see which ones earn their way in

This week I learned the AI tape is not rewarding stories anymore. $ORCL had USD 638B of RPO and still got hit because the funding/FCF math was ugly. BKSY had real defense wins but a USD 250M ATM. Proof matters. Financing matters. Vibes are getting taxed.

Indexes looked fine today, but the tape was picky underneath. $NBIS +4.6% while RKLB -10.8%, ADBE -6.8%, and GDX +3.0%. I don't read that as an all-clear chase signal. It is proof getting paid, crowded stories getting slapped, and hedges still working.

Saw something interesting this morning $GLD had the cleaner read than equities: gold +3%, silver +5%, while QQQ opened red and the gold model still stayed mildly bearish. That says hedge demand is real, but this is not the candle I want to chase.

This is the part of AI infra where I want receipts, not vibes: customer orders, backlog conversion, guide raises and margins. The theme is real. The trade is not “buy every optical thing with a chart.” Which is annoying, because that would be way easier. We'll see how it goes.

6. $IQE — IQE Upstream epitaxy and compound-semiconductor wafers for InP, GaN and GaAs. FY2025 revenue was about GBP 97M; MACOM's strategic investment and long-term supply agreements de-risked the balance sheet. Still needs AI datacenter revenue and margin proof.

6 photonics chokepoints I'm watching AI clusters need lasers, glass, InP wafers, burn-in, packaging and epitaxy. 1. $SIVE — Sivers DFB lasers for Jabil 1.6T LRO. Q1 sales -22%; pipeline USD 799M. ⚠️ science project — USD 36M rev, USD 2.9B mcap. 10x or zero.

Today's tape was not generic risk-on. SMH ripped +6.8% and MU +11.7%, while $ADBE beat, raised, said AI-first ARR tripled to >USD 500M... and still sold off about 6% after hours. My read: the market still pays AI infrastructure proof before software AI stories.

$ORCL is the cleanest AI-capex tell today. RPO jumped to USD 638B and OCI IaaS grew 93%, but the stock still got hit because FY26 FCF was -USD 23.7B and FY27 needs about USD 40B financing. AI demand is real. Funding it is the trade.