$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.
Saw something interesting this morning
$CRWD reports tonight with the stock +64% in 20 days and RSI ~86.
That is not a setup where I want to be a hero before the print. If recurring revenue, retention, free cash flow and guide all clear, great. If not, USD 681 matters.
The common thread: AI capex is moving from GPU headlines into boring power guts.
I like boring, but only when boring turns into orders, backlog, revenue, and margin. Otherwise it is just FinTwit cosplay with a better ticker list.
We'll see how it goes.
6 AI power and data-center bottleneck names I'm watching
Not the Mag 7. Weird suppliers around power, storage, RF, and chip test.
1. $AMPG — AmpliTech
Micro-cap RF/5G/quantum hardware. Q1 revenue USD 5.35M (+49%), GM 48%, backlog >USD 20M. Real orders; now prove conversion.
6. $BELFB — Bel Fuse
Power, magnetics, and connectivity parts for data centers, networking, and defense. Q1 revenue USD 178.5M (+17%), data-solutions apps grew 30%, and backlog hit USD 531M.
Needs a named AI/HPC design win before it becomes real conviction.
$MRVL +32.5%, LITE +13.7%, GFS +5.8%, Google -3.8%.
That is the AI capex tape in one line.
Market is paying the bottleneck suppliers and taxing the balance sheets funding the buildout. I want revenue proof, not vibes.
Saw something interesting this morning
$SIVE getting into GlobalFoundries silicon-photonics reference designs is real CPO validation.
But it is not a purchase order yet. This is tiny-scout territory, not victory-lap sizing.
Qualification first. Revenue/backlog next.
$XFAB deep dive:
X-FAB is not a zero-revenue science project. Q1 revenue was $195.6M (-4% YoY), or $205.8M excluding IFRS 15, above the $190-200M guide. EBITDA was $34.2M, a 17.5% margin. TTM revenue is about $747M, but net income was down hard and FCF was slightly negative.
The reason this is on my radar: it sits at the specialty foundry intersection of photonics, microsystems, SiC/GaN wide-bandgap power, and the European supply-chain reshoring trade. That is exactly the kind of boring bottleneck the market can suddenly decide is sexy.
The actual proof is mixed. Microsystems and Photonics revenue was $33.7M, +42% YoY and +35% QoQ. Wide-bandgap was $15.1M, +152% YoY and +49% QoQ. Those are real growth pockets.
But the backlog math is the wet blanket. Q1 order intake was $169.4M versus $205.8M ex-IFRS revenue, and backlog fell to $308.4M from $318.2M. If this is becoming an AI/photonics/power bottleneck, bookings need to lead revenue, not trail it.
Catalysts I care about:
1. Named photonics/datacenter/customer production order
2. SiC/GaN or 800VDC power customer ramp with real dollars
3. Q2/Q3 guide raise or estimate revisions tied to those high-growth segments
Risks are obvious: automotive visibility is still low, backlog is down, analyst targets are below spot, and this can easily become a European policy proxy trade with no revenue conversion. StockAnalysis had it around EUR10.19, EUR1.41B market cap, Hold-rated by 8 analysts, with EUR5.52 average and EUR7.98 high targets.
Entry/Verdict: HOLD / CATALYST WATCH / NO CHASE. I would not buy the social rerate here. Max size for me is 0% now; maybe a tiny scout only after a major pullback or hard order/backlog proof.
The company is real. The theme is real. The current price is asking me to pay for proof that has not arrived yet. We'll see how it goes.
Quantum is one of those themes where the TAM sounds enormous and the income statements look like a rounding error.
Policy money is nice. Orders, backlog, revenue, and dilution discipline decide whether this is investable or just an expensive science project.
4 quantum names I'm watching
Policy money finally hit the theme. Most of the bucket is still science-fair valuation until revenue shows up.
1. $GFS — GlobalFoundries
Foundry layer. Quantum Technology Solutions + cited USD 375M note. Need orders, backlog, or estimates.
4. $IONQ — IonQ
Best pure-play proof in the group: Q1 revenue was cited at USD 65M, FY26 guide USD 265M, and RPO USD 470M.
Still not cheap. SkyWater helps the manufacturing story, but I want revenue conversion, not quantum vibes.
Weekly conviction list update:
Added BE, PENG and VLN to the watch/pullback board since last week. BWA stays removed because it is a fine company but the asymmetry math is not the game. Main change: the tape is still rewarding AI infrastructure, but most of the list is now proof-gated after vertical moves.
$VST @ 154.62 — 🔴 ENTRY ZONE. Cleanest liquid AI power/nuclear core. Entry: hold existing 2028 calls; no fresh call add. Common add only around 146-151 or after fresh PPA/load-growth/Cogentrix/guide proof.
MRAM @ 28.45 — MRAM/defense/edge-memory option with Q1 product growth, Microchip capacity path and defense agreement. Entry: no add on pullback alone; add only after Q2 revenue above 16.5M, GM above 50%, and customer/defense milestone proof.
CEG @ 271.68 — Nuclear plus Calpine AI-power scarcity platform. Entry: first watch 255-265, preferred 240-255, or add above only after named premium PPA / PJM-FERC unlock / Crane de-risking.
PDFS @ 49.01 — Advanced-packaging yield software with Q1 proof, 246.4M RPO and offering overhang absorbed. Entry: starter only; add at 35-38 if intact or after Q2/Q3 revenue, RPO, eProbe, Exensio and FCF proof.
CAMT @ 164.97 — Real AI advanced-packaging inspection winner, but not cheap. Entry: add zone 135-140 intact reset or after Q2/Q3 order and estimate conversion.
ONTO @ 257.48 — Dragonfly/3Di/Atlas metrology proof is real; convert/Rigaku complexity keeps adds disciplined. Entry: 220-239 if thesis intact, or after order/margin/estimate proof.
POET @ 13.38 — Lumilens 50M PO / possible 500M framework is real, but revenue is still qualification-gated. Entry: no add-size before late-2026 samples, qualification and payment-backed PO conversion.
OSS @ 18.64 — Edge AI/defense compute with Q1 validation: revenue +55%, 51.6% GM, strong bookings and positive adj. EBITDA. Entry: no chase; add after Q2/Q3 GM, book-to-bill and EBITDA proof or intact reset to 13.70-15.40.
OUST @ 45.04 — Clean physical-AI LiDAR/perception play, but still negative EBITDA with ATM and concentration risk. Entry: no fresh add; preferred pullback around 35.95-32.44, then 28.93, or add after Q2 proof.
RDW @ 21.15 — Defense/space order proof is real, but spot is vertical and dilution/EBITDA risk still matters. Entry: backlog-to-revenue, margin and cash-burn proof; no chase above target math.
VPG @ 121.77 — Robotics sensor proof is real, including measurable humanoid orders, but the stock already rerated. Entry: wait for Q2 proof or 91-80 / better 75 intact reset.
KEEL @ 5.74 — AI/HPC powered-site optionality with 2.2GW pipeline and 533M liquidity. Entry: tiny only around 4.30-4.90 if intentionally buying lease-event risk; better reset 3.35-3.85.
AEHR @ 88.71 — AI/SiPh test order signal is real, but valuation already wants FY2027 to work. Entry: 80-83 first watch, 70-75 better, or add after FY2027 guide/consensus resets above 100M.
FLNC @ 24.63 — Hyperscaler storage MSAs plus 5.6B backlog. Entry: hold existing; fresh add waits for first hyperscaler order economics or intact reset to 17.45-18.00 / better 14.75-16.85.
RDDT @ 179.13 — Q1 proof remains excellent: revenue +69%, ad +74%, big EBITDA/FCF. Entry: no add after bounce; add only after Q2 beat/raise or controlled pullback into 148-155 / 154-163.
NVTS @ 23.99 — 800V GaN/SiC optionality is real, but the stock ran ahead of named order proof. Entry: trim-risk / hold token; proof-gated add only around 20.50-23.80, better 16.00-17.65.
BE @ 276.42 — AI-power leader with Nebius/AEP/Brookfield proof stack, but valuation is spicy and Form 144 supply says be patient. Entry: first look 274-285, preferred 220-244, breakout only above 323 with new proof.
VRT @ 318.24 — Quality 800VDC/data-center power compounder. Entry: first tactical 275-307 if thesis intact, preferred 242-275; add above spot only with named 800VDC/customer backlog or guide raise.
AEIS @ 292.84 — Real AI PSU/DCC proof, no named 800VDC production order yet. Entry: optional tiny tracker in 288-304 support; normal starter waits for Q2/Q3 DCC, margin and order proof.
NBIS @ 263.04 — Neocloud ramp validated, but valuation/capex/crowding say hold, not double. Entry: 199-210 intact or after Q2/Q3 ARR, power, financing and margin proof.
PENG @ 56.00 — Baby-DELL / AI-factory integrator candidate with real Q2 memory proof. Entry: event-risk zone 47.80-50.30, better 38.50-43.80; add-size only after named order/backlog/guide proof.
VLN @ 3.28 — Edge-vision connectivity optionality, not a data-center CPO trade. Entry: scout only around 2.35-2.65 or after Q2/Q3 revenue, GM and named production proof.
LPTH @ 16.11 — Defense/thermal imaging proof with 110.6M backlog and G5 program optionality. Entry: first look 15.00, better 13.90-14.00 / 13.10, or after named order/backlog conversion proof.
SIVE @ price unavailable in Yahoo check — laser-diode optionality, but Q1 failed add-size proof: revenue down, Photonics down, cash burn high, no named AI datacenter backlog. Entry: tiny tracker only; add-size waits for named orders/backlog/product revenue/burn proof.
LPKF @ €22.10 — LIDE glass-substrate chokepoint option with Q1 book-to-bill 1.4 and first AP capacity-expansion order. Entry: optional tiny tracker only; preferred scout zones €20.60 / €18.00 / €15.20.
AIXA @ €58.24 — Equipment-layer CPO bottleneck candidate with Lumentum G10-AsP orders and Q1 order/backlog proof. Entry: no chase; tiny only 48-52, preferred 40-45 if thesis intact.
Highest-conviction pick is still VST, but only because the price is finally back in the common-stock zone. The theme is obvious. The discipline is not paying vertical-chart prices for unconverted proof.
We'll see how it goes.
Portfolio update, because this thing has mutated a bit since the last post.
Biggest change: I promoted software/fintech beta to core instead of treating it like a side quest. $NOW is now the main expression, with CRWD, PLTR, RDDT, CRM, META, HOOD, and LMND around it.
AI infra and bottlenecks are still the research spine:
Neocloud and compute — NBIS, IREN, HIVE, NVDA, AMZN
Memory and semis — MU, DRAM, MRAM, TSM, CAMT, ONTO, PDFS
Photonics and optical — LITE, AAOI, COHR, GLW, VIAV, HLIT
Power and grid — VST, CEG, FLNC, NVTS, ENPH
Physical AI and edge — OUST, OSS, AOSL, SANM
Crypto wrappers and hedges — IBIT, ETHA, XRP, SOXS
What changed: added CRM, NVDA, SANM, SG, HIVE, IBIT/ETHA wrappers, and a tiny SOXS hedge. Removed GGLL, SOXL, MSFT, VEEV, and the old direct BTC/ETH sleeve.
Most of my research here is still around photonics, memory, power, and AI infra bottlenecks. But the actual portfolio is broader because the market keeps paying software and fintech too. Annoying, but profitable is allowed.
We'll see which ones earn their way in
This week I learned AI rack demand is real, but the money does not automatically flow to every supplier nearby.
Dell printed $24.4B of AI orders. $PENG has the rack exposure, but still needs named order/backlog proof before I care.
Receipts first. Sympathy trades second.
Today's tape was the rotation test.
$TEAM +15% on real numbers: revenue +32%, Cloud +29%, RPO +37%, 31% FCF margin. HOOD was +9% too.
AI hardware is still the obvious trade. The harder trade is who turns all that capex into actual software revenue.