$16.8 billion in 15-year contracted leases. Triple-net. Take-or-pay. Anthropic on the lease through Fluidstack. Google backstopping the payments. JPMorgan and Goldman financed the buildout at 6.192 percent non-recourse.

Market cap, $10.2 billion.

The entire enterprise trades at 0.64 times the contracted backlog the company has already signed. Add the option value extension and it’s 0.43 times.

Most growth stocks trade at five to ten times forward sales. This one trades at less than one times signed 15-year demand.

The discount is the trade.

The Power Problem

By 2030 the US needs roughly 40 new nuclear plants worth of new generation just to power the AI training that already has contracts attached to it. The Department of Energy projects data center electricity use to triple by 2028.

Greenfield data center buildouts take 4 to 7 years from a standing start. Power interconnect alone is a 2 to 3 year process.

So when a hyperscaler signs a contract for AI capacity online by 2027, they cannot build it themselves. They need a landlord. Someone with the land already energized. Sites already on the grid. Cash already in the construction phase. NVIDIA reference architecture already designed in.

The company I bought owns the actual megawatts.

597 megawatts contracted. Triple-net. Take-or-pay. To investment-grade tenants on 15-year terms. Sites in Texas and Louisiana already energized or with utility interconnect locked.

8,375 megawatts in pipeline. 5,315 under diligence. 1,680 under exclusivity. 550 under development. 830 under construction.

The Variant Perception

The market is still tagging this name as a Bitcoin miner. Beta 5.72. Fifteen percent short interest. Every quarterly print gets dragged down by unrealized digital asset losses on the balance sheet.

The mining is real. It’s also ring-fenced now. Sits inside a majority-owned subsidiary called American Bitcoin Corp (ABTC) with 28.1 EH/s of hash rate and a 13,696 BTC treasury worth roughly $1.1 billion at $80,000 BTC, per The Block’s treasury tracker.

The parent company is the data center landlord. The legacy hash rate is an asset-light cash cow funding the AI infrastructure buildout. The pivot is functionally complete on the contract side.

Wall Street has not re-rated the multiple yet.

The closest pure-play AI infrastructure comp in my book is NBIS, an anchor at 8.98 percent of book, +89.69 percent lifetime. NBIS trades at premium AI cloud multiples on signed Microsoft and Meta backlogs. The company I bought today is in the same Datacenters layer at a backlog discount because the market still cannot separate the BTC noise from the AI infrastructure substance.

This is a re-rate trade. Each new contract announcement drags the multiple toward where the pure-plays trade. Each pipeline conversion is a $1 to $5 billion catalyst. Each tranche expansion on the existing tenant relationships is another $5 to $10 billion of contracted backlog dropped into the disclosure.

The room I’m in keeps showing me things adults miss.

The next section names the company, the entry price, and the AVGO trim I made to fund it. Free subscribers see everything. Pro launches July 22. Subscribe to read it. Free. No upsell, no spam.

Subscribe now

The Reveal

The ticker is HUT. Hut 8 Corp. Dual-listed Nasdaq and TSX. Last $97.

Post-merger Hut 8 plus US Bitcoin Corp from late 2023. The “Hut 8” name is a relic of the BTC mining roots. The actual business today is gigawatt-scale AI data center development.

The Counterparty Stack

This is the part the market is mispricing.

River Bend campus, Louisiana. Tenant is Fluidstack. End user is Anthropic. Payment is backstopped by Google. Tranche 1 signed with up to 1,000 megawatts of right-of-first-refusal capacity available for Tranches 2 and 3. Energization target Q2 2027. Source, Hut 8’s own partnership press release.

Beacon Point campus, Texas. Tenant is an undisclosed “high investment grade” customer. Base term contract value $9.8 billion on the first 352 megawatts. With options the deal stack goes to $25.1 billion. 1 gigawatt utility interconnect locked. Energization target Q1 2027. Designed to NVIDIA’s DSX reference architecture for gigawatt-scale training clusters. Data Center Dynamics covered the lease structure in depth.

The financing. $3.25 billion in senior secured notes at 6.192 percent. 16.5 year tenor. Non-recourse to the parent. That is investment-grade project finance from JPMorgan and Goldman at 85 percent loan-to-cost.

You do not get that paper on a Bitcoin mining business. You get it on contracted AI infrastructure with investment-grade tenants.

Total liquidity at the parent is $1.3 billion. $795 million parent cash plus $489 million at ABTC.

The Q1 Print

Reported May 6, 2026 per Hut 8’s earnings release:

• Revenue $71 million, up from $21.8 million a year ago. Plus 226 percent.

• Compute revenue $66 million. Power $3.7 million. Digital Infrastructure $1.3 million.

• Gross margin 64 percent, up from 47 percent.

• Net loss $253 million, mostly $295 million in unrealized digital asset losses on the BTC mark.

• Adjusted EBITDA negative $250 million.

The income statement looks ugly because the BTC mark drags it every quarter. The cash flow picture and the contract backlog tell a completely different story. This is the first quarter where the AI pivot is the primary revenue line.

The Valuation Math

EV divided by contracted backlog equals 0.64 times. Add the option value extension and you are at 0.43 times.

For the comp set:

NBIS — 2.10x EV/backlog — Anchor: Microsoft + Meta — In book: 8.98%

CIFR — 1.05x — Anchor: AWS pivot — In book: 1.80%

APLD — 0.92x — Anchor: CoreWeave — Not held

IREN — 0.85x — Anchor: Microsoft — Was held, rotated (Held by dad)

HUT — 0.64x — Anchor: Anthropic via Fluidstack — New today, 1.78%

Analyst consensus is Strong Buy per Public.com’s forecast aggregation. Average target $115.75. Jefferies at $156. Citizens at $140. Rosenblatt at $124. 26 percent upside to consensus. 70 percent upside to the high target.

The Street sees the re-rate coming. They just have not finished pricing it.

The Risk

Three kill vectors. Written before the position filled.

One. BTC crash. The treasury and ABTC subsidiary mark down together. 15 percent short interest is positioned for exactly this scenario, per StockAnalysis.com.

Two. Anthropic or Fluidstack contract delay or modification. River Bend is 41 percent of contracted backlog by value. Any restructure resets the multiple. The Google payment backstop mitigates credit risk, not optical damage.

Three. Energization delay. River Bend is Q2 2027. Beacon Point is Q1 2027. Any slip on power interconnect, NVIDIA supply, or construction takes the multiple down hard.

Those are real risks. They are also why the stock trades at 0.64 times backlog instead of NBIS class multiples. The discount IS the risk premium. The thesis is the discount is too wide.

If any of these fire cleanly, the position trims or exits on the four-quarter rule. Not on vibes. Not on a red day. On a written kill that tripped a written line.

The Position

This morning I sold 3 shares of AVGO at the open and bought 10 shares of HUT at $97.48.

• 10 shares at $97.48 = $974.80 cost basis

• 1.78 percent of book (NAV $54,442)

• Target weight 3 to 4 percent ($1,633 to $2,178)

• Remaining capex about $660 to $1,200

Cash is tight at $539 (one percent of book, no dry powder). Tranche 2 and Tranche 3 require another rotation, not new capital.

Add plan:

• Tranche 2 at $85 to $88 on a 50-day pullback

• Tranche 3 at $80 on a BTC drawdown (not a thesis break)

• Stop chasing here. I already paid up at $97 because the May 6 Beacon Point disclosure is the structural re-rate that justifies a starter

Trim trigger pre-set at +100 percent from cost, around $195. Kill at break of $70 for two consecutive days or any Anthropic / Fluidstack contract modification.

Horizon 12 to 18 months. The re-rate happens between now and River Bend energization in Q2 2027.

Why I Trimmed AVGO For This

AVGO is still an anchor. One of the five. The trim was 3 shares of 8. The thesis on Google TPU custom silicon and roughly 90 percent Ethernet switch share is intact.

I trimmed because the position was up 24 percent (well below the +100 percent trim trigger) and because HUT at 0.64 times backlog is a better risk-reward setup than adding more AVGO at eight times sales.

This is what rotation looks like inside my book.

Anchors stay anchors. When a Pivot opportunity trades at a re-rate discount with the right counterparty stack, capital rotates. The book is alive. Positions you see today are not forever holdings. I sell to redeploy into the next conviction.

The framework was on the page before this trade fired. Kill vectors written. Trim triggers set. Entry zone defined. Capital constraint acknowledged.

Discipline isn’t what you do when things go right. It’s what’s already in place when they go wrong.

What I’m Watching Next

Early August 2026, Q2 earnings. First print with Beacon Point and the senior notes in disclosure. Watch backlog reaffirmation and pipeline conversion language.

Q2 to Q3 2026, the Coatue convert. $150 million mandatory redemption begins. Clears an overhang.

Q3 to Q4 2026, River Bend construction milestones. Any data hall topping-out or NVIDIA shipment announcement is a tape-mover.

Tranche 2 and 3 Anthropic announcements. Up to 1,000 megawatts of ROFO capacity at River Bend plus 1,050 megawatts jointly diligenced beyond. The market treats expansion as a $5 to $10 billion catalyst per tranche.

Beacon Point Phase 2. Remaining 648 megawatts of the 1 gigawatt campus.

Pipeline conversion. 5,315 megawatts under diligence is the optionality.

The Receipt

The market sees a Bitcoin miner. The deal stack sees Anthropic, Google, JPMorgan, Goldman, NVIDIA. The contract value sees $16.8 billion. The price sees 0.64 times that number.

This is what reads as a Bitcoin miner from the outside and a hyperscaler landlord from the inside. The framework was on the page before the contracts dropped. The framework is what called the rotation.

The full HUT live thesis, the daily watch, the kill-vector status, and the rotation receipt timeline live at youngbullinvests.com/research/HUT. The full book lives at youngbullinvests.com/book. Where HUT sits in the eight-layer Physical Layer stack is at youngbullinvests.com/map.

Real money. Real positions. Real receipts.

You’re in the room or you’re not.

Young Bull on Substack · The Book · The Map · HUT research card · Get Pro (July 22)

Disclosure

I am long HUT. 10 shares as of 2026-05-20, day one of the position.

I am long AVGO. 5 shares (trimmed from 8 today to fund HUT). AVGO is one of the five anchors.

I am long NBIS (Datacenters anchor, 8.98 percent of book), and a number of other Physical Layer of AI positions on /book.

I do not own CIFR at anchor weight (1.80 percent Pivot, day-of disclosure). I do not own APLD or CORZ. I do not own SanDisk or any of the names in the broader bitcoin miner sleeve. I used to own IREN and rotated out of my personal position earlier this year.

I rotate. Positions shown today are not forever holdings. The book is alive. Lifetime gain compounds, individual positions do not. Anyone tracking should understand the rotation is the strategy, not the side effect. The book at /book is what I actually hold today.

Young Bull covers the Physical Layer of AI broadly. I do not own every name I mention. Research and watch list names are exploratory, not positions. The /book page is authoritative.

The Substack stays free. Pro launches July 22, 2026 at $199 per year. Founding 50 seats first.

I’m 17. The runway is 50 years. The Physical Layer of AI is the build. The framework was on the page before this trade fired.

Sources

• Hut 8 Q1 2026 earnings press release. Primary source on Q1 revenue, gross margin, liquidity, $3.25B senior notes close.

• Beacon Point $9.8B / 352 MW lease announcement (Morningstar / PRNewswire). Beacon Point deal mechanics, IG tenant, NVIDIA DSX reference, 1 GW utility interconnect.

• Hut 8 / Fluidstack / Anthropic partnership press release. River Bend tenant identity, 245-2,295 MW tranche structure, Google payment backstop, JPMorgan / Goldman 85 percent loan-to-cost financing.

• Data Center Dynamics on the 352 MW Texas lease. Independent confirmation of Beacon Point deal structure and triple-net, 3 percent annual escalator.

• StockAnalysis.com HUT statistics. Market cap, EV, beta 5.72, short interest 15.26 percent, Altman Z 1.71, TTM net loss.

• The Block, Hut 8 BTC treasury 13,696 BTC. Treasury holdings and NAV reference.

• Public.com HUT analyst forecast. Consensus Strong Buy, $115.75 average target, Jefferies $156 initiation.

• DOE 2028 data center electricity projection. External AI power demand reference.

Internal: [[Research/Tickers/HUT]], [[Research/Tickers/NBIS]], [[Research/Tickers/IREN]], [[Research/Tickers/AVGO]], [[MASTER_THESIS_v2_2026-05-05]], [[System/Discipline-v2]], [[Reports/PORTFOLIO_SNAPSHOT_2026-05-20]].

People keep asking what I actually own.

Here it is. The whole book. 31 positions. Net account value $56,092.27 as of the May 16 close. Up $12,209.56 lifetime, +27.84%. Real money. Real receipts.

I am starting deep dives on these names. Anchors first, then new adds and sells as they happen. The heavy per-position work goes behind the Pro paywall when Pro launches July 22. Before any of that lands, you should see the whole board.

One thing upfront. I am 17. I run real money against my own framework in public and I learn out loud. Some of these positions will work. Some will not. The whole book is the receipt of what I think. It is not a guarantee of what is right. Read it that way or get burned.

The promise has always been receipts before takes. This is the receipts page.

The numbers up front

• Account value: $56,092.27

• Cost basis: $43,858.63

• Lifetime gain: +$12,209.56 (+27.84%)

• Realized profit YTD: ~$9,000 (trim triggers and trades closed this year)

• Cash: $24.08. Fully invested.

• Position count: 31 (5 anchors, 4 high-conviction sleeves, 20 working, 2 ETF parking)

The book has been live the better part of this year. No kill vectors firing. No trim triggers firing. I am not adding here. I am writing this post.

Layer by layer

Compute & Memory     ████████████ ~30%   MU AVGO ARM ON AEHR AXTI CRDO EUV
Power & Grid         ████ ~10%           VST BE SMR CIFR
Datacenter / Cloud   █████ ~12%          NBIS BRUN WYFI
Networking           ███ ~7%             ANET NOK LUMN
Photonics & Quantum  ████████ ~20%       ALMU XNDU LITE SIVEF LPKFF
Space & Defense      █████ ~12%          RKLB ASTS
Sensing / Autonomy   ██ ~5%              OUST PENG TEM
Working / Honesty    ██ ~2%              HIMS
ETF parking          ██ ~2%              MDY SMH

Exact weights on the live book at youngbullinvests.com/book. Compute and memory is the heaviest by design. Every other layer feeds compute or routes its output. Photonics and quantum is second-largest because that is where I think the next mispricing lives.

The 5 Anchors

The anchors are the spine. Five names. Five thesis floors I do not blink on. The trim trigger on every anchor is +200% lifetime (ALMU set lower at +100% because of smaller cap). I am not adding to anchors at these prices. I am not selling. I am letting the receipts compound.

MU. Micron Technology

Layer: Compute & Memory · Position: 12.93% (largest single) · Entry: 03/27/26 · Today: +105.08%

Why it is here. MU is the HBM trade. High Bandwidth Memory has been sold out before chips ship. Samsung and SK Hynix have signaled oversupply concern since February. Their February guide said HBM shortages last into 2027. SK Hynix already allocated its entire 2026 supply to NVIDIA. Customers are paying prepayments to lock 2027 allocation. Micron is the third leg of that stool, and the stock has done what a sold-out book does. The +105% lifetime is not a peak. It is what the cycle is paying right now.

Kill vector. Micron, Samsung, and SK Hynix would all have to guide oversupplied in the same quarter. Today all three guide undersupplied through 2027.

Trim trigger. +200% lifetime. 95 points away. Holding through any soft tape that does not break the supply-demand math.

Variant perception. Bears price MU off the historical DRAM cycle. Bulls price it off the AI-specific HBM cycle, where customers pre-pay for capacity that does not exist. Different cycle. Different math.

NBIS. Nebius Group

Layer: Datacenter / neocloud · Position: 9.80% · Entry: 02/09/26 · Today: +113.43%

Why it is here. NBIS is the cleanest pure-play on neocloud. AI compute infrastructure-as-a-service for hyperscalers and frontier labs. The Q1 2026 print (May 13): $399M revenue, +684% YoY, with the Nebius AI segment at $390M at 45% adjusted EBITDA margin. 2026 capex guide raised to $20-25B. A 1.2 GW Pennsylvania site is now on the page. The backlog is anchored by Meta ($27B, five-year) and Microsoft ($19.4B). Roughly $50B locked.

Kill vector. Hyperscaler concentration. Two customers carry the backlog. Either Meta or Microsoft pulling capacity in-house is the structural risk the Q1 print did not retire. The Eigen + Tavily + Clarifai acquisitions tell me Volozh is building the inference layer as back-door diversification, but customer concentration is not yet solved.

Trim trigger. +200% lifetime. 87 points away. Not adding into the post-print pop. Not chasing the gap.

Variant perception. Bears price NBIS as a capex burn. Bulls price it as a customer-prepayment-funded buildout with $9.3B liquidity and an asset-backed financing structure. Two P&L pictures, same balance sheet.

AVGO. Broadcom

Layer: Compute & Memory + Networking · Position: 6.07% · Entry: 03/31/26 · Today: +22.75%

Why it is here. AVGO is the only public name with a contracted custom-AI-silicon backlog at hyperscale. $73 billion of custom-silicon backlog with Google TPU, Meta MTIA, and OpenAI confirmed as customers. The networking side (Tomahawk and Jericho switching silicon) is the other half. Every hyperscale AI fabric in production runs on Broadcom switching silicon today.

Kill vector. Hyperscalers internalize the custom-silicon design. Google has Pixel Tensor and TPU. Meta has MTIA. The kill case is the next generation of custom AI chips moves off Broadcom's design service onto in-house silicon. Today the backlog says that is not happening on the timeline the bears price.

Trim trigger. +200% lifetime. 177 points away. Sized smaller than MU because the entry was late.

Variant perception. AVGO prints on both sides of the AI capex curve. Custom silicon goes up with training. Switching silicon goes up with inference. Most analysts pick one of those two lines. I own the company that owns both.

ALMU. Aeluma

Layer: Photonics & Quantum · Position: 5.10% · Entry: 04/22/26 · Today: +25.56%

Why it is here. ALMU is the smallest of the five anchors and the most asymmetric. Indium phosphide quantum-dot photonic ICs. Commercial laser shipping FY27. The thesis floor is the NASA award on April 21 for integrated quantum dot lasers, plus six government R&D contracts. The same photonic stack that builds AI datacenter optical transceivers builds the inter-satellite-link hardware in the orbital layer. One supply chain, two altitudes.

Kill vector. Execution. ALMU is pre-commercial-revenue at scale. The kill case is FY27 product slip past 2027 or a critical NASA program loss.

Trim trigger. +100% lifetime (lower than other anchors because of smaller cap). 74 points away. Adding more is off the table until the FY27 commercial revenue line.

Variant perception. Most photonics analysts cover ALMU as a venture-stage materials story. The bull case is to read the customer list. NASA. DoD. AI hyperscaler R&D groups. That is what rerates this name when the FY27 laser ships.

VST. Vistra Energy

Layer: Power & Grid · Position: 2.99% · Entry: 01/19/26 · Today: −12.42%. The worst anchor on the page.

Why it is here. VST runs the largest US nuclear fleet outside Constellation. Microsoft signed a 20-year PPA with VST for nuclear capacity. Meta did the same. Behind-the-meter nuclear is the only way certain hyperscaler workloads come online before 2030. The grid cannot deliver the megawatts hyperscalers have already contracted for (ERCOT has 226 GW of large-load requests in queue, three-quarters data-center demand). Nuclear capacity already built and already on the grid is irreplaceable. VST owns it.

Why it is red. Regulatory pendulum. The market punished the entire power-and-grid AI trade through April on natural-gas weakness, hyperscaler capex concern, and headline risk around behind-the-meter PPAs (the FERC review of Talen / Amazon spooked the cohort). VST got dragged even though the long-dated Microsoft and Meta PPAs are intact.

Kill vector. A regulator unwinds an existing behind-the-meter nuclear PPA on retroactive grounds. The headlines so far have been about new deals, not existing ones. Different bar.

Trim trigger. +200% lifetime. The trim rule does not apply when a position is red. The kill trigger has not fired. I am holding through this drawdown. This is the position that tests whether I run my own discipline rules in real time. I do.

Variant perception. Bears say nuclear is a regulated utility wrapped in an AI multiple. Bulls say nuclear is the only economically interconnectable base-load left, and the existing fleet is worth more than the comp set prices it. Twenty-year PPAs do not get unwound on a single regulator's tweet.

The 4 high-conviction sleeves

Done the work on, not promoted to anchor status. Sizing smaller. Conviction real.

ASTS. AST SpaceMobile

Layer: Space & Defense / Direct-to-cell · Position: 7.46% (second-largest after MU) · Today: +13.01%

This is my current vibe pick. The name I am leaning hardest into right now.

ASTS is the satellite direct-to-cell trade. Six BlueBirds in orbit (BlueBird 6 launched April 19 on Blue Origin's New Glenn). Target 45 by end of 2026. $3.5B cash. 60 MNO partners covering 3 billion subscribers globally. FirstNet authorization April 2025. T-Mobile direct-to-cell live across 500,000+ square miles at $10/user/month.

Q1 missed on revenue (60% miss, $14.7M vs $36.6M expected). FY 2026 guide reaffirmed at $150-200M. Back-half loaded by design. Calendarization story. Not a thesis break.

I added 10 shares mid-April on the BlueBird 7 failed-insertion drawdown. Position is back in the green. The lean: operational ramp visible, cash runway real, carrier ecosystem the moat. Next four quarters are about converting 60 MNO partners into revenue.

Kill vector. BlueBird operational failure or MNO revenue not appearing by Q4 2026.

XNDU. Xanadu Quantum Technologies

Layer: Photonics & Quantum · Position: 6.06% · Today: +64.04%

This is the name everyone in my circle already knows.

XNDU is the only public photonic-quantum pure-play. Room-temperature qubits on fiber. DARPA, Lockheed, AMD on the customer list. Five-year thesis.

The honest receipt: the +64% lifetime is the post-dilution number. The S-1 resale that hit May 5 crashed the stock −63% in 48 hours and reset the lifetime gain from a peak of +335% down to roughly +62%. The Xanadu + EVG manufacturing partnership announced the same week is what is keeping the thesis floor intact.

I trimmed XNDU once already (83 shares on April 15 at $25.22, $2,094 realized). Position now at the sleeve-rules-compliant size. Never full exit on XNDU. The discipline rule is in the playbook. This is the conviction moonshot.

Kill vector. A second dilution event before the manufacturing partnership produces commercial revenue.

BE. Bloom Energy

Layer: Power & Grid · Position: 4.92% · Today: −4.29%

This is the safe play.

BE makes solid-oxide fuel cells. Behind-the-meter power generation for data centers that cannot wait three years for substation interconnection. Same thesis as VST. The grid is the bottleneck. BE attacks it from the on-site generation angle instead of existing base-load.

I doubled the position on May 5 on a drawdown. Equinix already runs Bloom at scale. The receipts I am watching: hyperscaler announcements using Bloom for primary or backup AI datacenter power. The question is whether the AI-specific cycle accelerates the order book.

Kill vector. Hydrogen-economy unit economics break. Bloom's pivot to hydrogen is the long-dated lever. If green hydrogen costs do not fall on schedule, the multiple compresses.

LITE. Lumentum Holdings

Layer: Photonics (terrestrial + orbital crossover) · Position: 3.46% · Today: +1.23%

This one is backed by the deepest NotebookLM-sourced research in the book.

Lumentum builds the optical components for both AI datacenter transceivers and (per the Orbit thesis) the inter-satellite-link photonic stack going to orbit. Same indium-phosphide supply chain. Same coherent modulation. Different altitude.

Near-flat lifetime because the entry was recent and the optical-component cycle has been choppy on China-channel concerns. Structural thesis intact. Every hyperscaler optical transceiver upgrade goes through Lumentum or a Lumentum-adjacent vendor.

Kill vector. A coherent-modulation packaging shift that bypasses Lumentum's IP. Today there is no credible alternative.

The rest, by layer

Working positions. Smaller sizing, variable conviction, tighter kill vectors. The book is alive. The bench is not a museum.

Chips bench

• ARM. IP royalty on every chip. Q4 FY26 record $1.49B. v9 royalty moving 25% → 60-70%. Deep dive May 12.

• ON. SiC power for datacenters and EVs. Green.

• AEHR. SiC burn-in test. Bull case: SiC capacity coming online 2026-2027 needs Aehr-class volumes.

• AXTI. Compound semi substrates (InP, GaAs). Cross-pollinates ALMU.

• CRDO. High-speed AI datacenter connectivity. AEC line is SerDes-adjacent.

• EUV. Corgi's new ETF (May 6). EUV-litho / semicap-photonics supply chain. 41 holdings. Tiny starter.

Power & Grid bench

• SMR. NuScale. First NRC-approved SMR. DOE backstop on Idaho. Pre-revenue, regulatory moat real.

• CIFR. Cipher Mining. BTC miner pivoting grid interconnect to AI datacenter buildout.

Networking bench

• ANET. Datacenter switching. Q1 +35.1% YoY, AI guide $3.5B doubling. CEO Ullal on the call: "our demand is actually the best I have ever seen in my Arista tenure." Bought the margin-flush dip.

• NOK. AI optical pivot finally printing. Q1 AI/cloud revenue +49% YoY. Optical Networks +20%. Network Infra guide raised 12-14%. Infinera integrating. Argus upgraded to Buy. Holding through the re-rate.

• LUMN. Fiber infrastructure. Position underwater. Dark-fiber asset thesis hasn't played out. Small, holding, would not add today.

Photonics & Quantum bench

• SIVEF. Sivers Photonics. Indium phosphide adjacency. Small base, lifetime positive.

• LPKFF. LPKF Laser & Electronics. German photonics laser equipment. Recent buy near breakeven.

Space & Defense bench

• RKLB. Rocket Lab. Q1 $200M revenue, +63.5% YoY. 31 launch contracts signed in Q1 alone. Backlog $2.2B. Neutron debut Q4 2026. Mynaric acquisition puts RKLB in the photonic-orbital stack. Already trimmed at +100% per Adjacent Tech rules.

Sensing & Autonomy bench

• OUST. Ouster. Lidar and perception. Position held.

• PENG. Penguin Solutions. AI infra for tier-2 enterprises. The "every AI data center that is not a hyperscaler" trade.

• TEM. Tempus AI. Q1 $348.1M, +36.1% YoY. Insights +44.1% on pharma. Six $100M+ partners on a 500-petabyte database. FY26 guide $1.59-1.60B. Variant: market reads Tempus as precision-oncology lab. I read it as the AI training-data layer for healthcare.

Datacenter / Neocloud small-cap bench

• BRUN. Boost Run. Just-IPO'd neocloud (Nasdaq May 11, via Willow Lane SPAC). $940M contracted revenue at IPO. Targeting $375M+ ARR by FY2026 end. Pairs with NBIS at different scale.

• WYFI. WhiteFiber. AI/HPC colocation + cloud. Recent IPO. −5.16%. Same thesis as BRUN, different angle. Kill vector: first full-quarter print missing on utilization or contract velocity.

Working positions

• HIMS. Telehealth + GLP-1 distribution. Position underwater. Channel is real. Multiple compressed. Holding small, not adding.

ETF parking

• MDY / SMH. Index parking. Small. Long-term auto-reinvest base and broad-AI-cycle exposure.

On the watchlist

Same rigor as held positions, not bought yet.

• HLIT. Pure-play broadband-edge networking. Q1 +43% YoY, backlog +87%, FY26 raised.

• COHR. Photonics + laser systems. Deeper portfolio overlap with ALMU and LITE.

• ONTO. Chip inspection and metrology. Semicap layer beneath ASML.

• MTSI. High-perf analog/RF for AI datacenter, aerospace, defense. Overlaps CRDO.

• NVT. Busbars, enclosures, thermal. Unsexy but load-bearing. Pairs with VST and BE.

• VRT. Datacenter cooling and power. Classic AI-tailwind. 2026 prints tell whether entry waited too long.

None of these are buys today.

The questionable ones

Five positions are underwater: VST, BE, HIMS, LUMN, WYFI. Four covered above. The fifth (HIMS) is on the page because I want this list to include positions outside the Physical Layer thesis. GLP-1 channel intact. Multiple is not.

LUMN I would not enter today. Holding existing share count because the fiber-asset thesis is still on the page, but sizing is small enough it does not move the book.

Every newsletter that shows you only winners is doing a different job than the one I signed up for. You should see the red names. That is the receipt that matters.

How to read the book

Every ticker is clickable. Lands on the per-ticker research page (notes, kill vectors, trim triggers, latest agent runs).

Open the live Book

See the 8-layer Map

Visit the Room

The vote

Anchors first on the deep-dive queue. Tell me which one to write first:

• MU. The HBM trade at +105% lifetime.

• ALMU. The photonics + quantum laser anchor.

• AVGO. Custom AI silicon + networking, both at once.

• ASTS. The high-conviction sleeve I am leaning hardest on.

Drop the answer in the Substack poll above (or in the comments. Write-ins land in my notebook). Most-voted name gets written next.

The closer

Free: my book. Paid: the notebook.

The Substack stays free. The book page stays free. This post stays free.

Pro launches July 22 at $199/year. Founding 50 seats first. Pro is the per-position kill vector, trim trigger math, variant perception, NotebookLM audio briefings, and the Vestor agent watching your portfolio against mine. That is the differentiator. "Quinn sees your book." Not the other way around.

Join Pro Waitlist (July 22)

I am 17. The runway is 50 years. The Physical Layer of AI is the build. The whole book is the receipts.

When I am wrong, the postmortem lands before the next deep dive. That is the deal.

You have the whole book now.

Young Bull on Substack

His first reaction was the same as everyone’s:

“Nokia? They don’t even make phones anymore.”

That’s exactly the point.

The bet on NOK

Nokia isn’t a phone brand. Hasn’t been for years.

What they own is some of the most important networking and optical IP on the planet — Bell Labs heritage, Alcatel-Lucent acquisition, real silicon, real fiber, real hyperscaler relationships. AI data centers don’t run on vibes. They run on networking fabric and optical interconnects between GPUs. That’s exactly what Nokia builds.

The market still prices it like a fading consumer brand. The asset is already there. The repricing hasn’t happened yet.

NOK — five years of going nowhere. The setup.

That’s the gap I’m watching.

Why I think I can spot these — and the trade that proved it

I started investing at 13, putting a fixed monthly amount into the market. My dad’s an ER doctor and a long-time investor. I’d sit at the kitchen table and pitch him positions until he’d push back hard enough that I had to come back sharper the next week.

The first time I really pitched him — not asked, pitched — was during the DeepSeek selloff in January 2025. The market was panicking that AI training costs would collapse and U.S. semis were dead. Tariff noise was layered on top.

I told my dad the panic was wrong. The thesis hadn’t broken — the price had. AI compute demand wasn’t shrinking, it was being mispriced for a week.

I made the case for buying the dip on TSM, NVDA, and AMD.

He listened. We bought.

I sold all three earlier this year for huge profits and rolled the proceeds into my next plays. That trade was the first time my dad treated me like a peer instead of a kid asking questions. It’s also where Young Bull really started.

In March he handed me $20K to deploy on tickers I was finding. Right now that book is up around $8K.

He’s not just my dad anymore. He’s a client of the strategy.

The market my dad invested in is gone

This is something we go back and forth on constantly. He’ll tell me the market is at highs and a crash is coming. I’ll tell him the rules he learned to invest under don’t apply the same way anymore.

A generation ago, you bought stocks the way you picked a career or a town to settle in — based on stability, brand, and certainty. Coca-Cola was Coca-Cola. IBM was IBM. The safe path stayed safe for forty years.

That world is over. Stable jobs aren’t stable. Safe industries aren’t safe. The biggest companies on earth a decade ago aren’t even in the top ten anymore. The new edge isn’t picking the most certain bet — it’s recognizing patterns inside the uncertainty before the rest of the market does.

Old game was rewarded for conviction in stability. New game rewards conviction in change.

How I find these — and the edge

Most investors form opinions inside the same five outlets. Same analysts, same consensus, same crowded trades. Their edge is depth in a narrow lane.

I’m 17. The room I’m in looks completely different.

I’m in conversations adults aren’t in. Group chats, Discords, what my friends are building, what nobody on CNBC is talking about yet. Most of it is junk. But every once in a while a real signal shows up before Wall Street has a name for it.

Think about the crowd you had to be in to know Bitcoin in 2013. Not finance Twitter. Niche internet corners. The edge wasn’t intelligence — it was exposure to a different signal layer.

That’s the bet I’m making with Young Bull. Broad input beats narrow consensus when the world is changing fast.

Introducing: The Pivot Sleeve

This is the system, formalized.

The thesis: most of the best returns in the next five years won’t come from new AI startups. They’ll come from companies the market still prices as their old business while their assets get repositioned for the new one.

Nokia fits the pattern exactly. Old label. New asset. Gap between the two.

The discipline is what separates a strategy from a hot take:

• Position size stays small — this is asymmetric, not core

• Trim 25% at +100%, 25% more at +200%, let the rest run

• Cut at –40% if the pivot thesis hasn’t shown up in filings or earnings within four quarters

• No averaging down on losers

NOK is the first active candidate.

The bet

The book runs live at young-bull-site.vercel.app. Every position, every pivot, time-stamped before the order goes in. If I’m wrong, you see it the same day I do.

I’m 17. I’ve been wrong before and I’ll be wrong again.

But the receipts stay public, the room I’m in keeps showing me things adults miss, and the strategy now has a name.

— Quinn

If this lands, restack it.

The question that comes back every week: if you had to pick five names — five — and not touch them for two years, what are they?

This is the list. Not the highest-flyers. The ones that hold the thesis up. Everything else is satellite.

The frame

Every dollar that pours into AI training has to land somewhere physical. It needs a chip to run on, a wire to travel through, a building to live in, and a power plant to keep it alive. That’s the Physical Layer. It’s not glamorous. It’s mechanical, capex-heavy, and slow-moving — until it isn’t.

The software guys get the magazine covers. The infra guys get the cash flows. Be early on the cash flows.

Here’s the spine.

1. AVGO — the boring one that’s actually lethal

AVGO — Compute Conviction Anchor.

AVGO is what NVDA wants to be when it grows up. NVDA is one product cycle from a hyperscaler ASIC eating its lunch. AVGO is the hyperscaler ASIC — Google’s TPU silicon, Meta’s MTIA, custom AI silicon for whoever shows up with a checkbook. Plus the networking stack that ties it all together. Plus VMware. Plus a dividend.

This is the picks-and-shovels trade dressed up in a $1.3T market cap.

The reason AVGO sits at the top of this list isn’t the upside — there are higher-beta names on this list. It’s that AVGO compounds whether the AI cycle keeps ripping or stalls. Custom silicon revenue is sticky, contracted, multi-year. Networking is monopolistic. The dividend is a cushion. If AI capex slows in 2027, AVGO trades flat for a year and you collect the yield. If AI capex re-accelerates, AVGO is the silent 3x.

The five-year read: $421 → $700–900.

2. VST — the most underowned power name in the market

Why it matters: 41 GW of generation. 6.4 GW of nuclear. Texas IPP exposure. Capacity prices repricing higher.

You can’t run a data center without power. We are heading into a window — call it 2026–2030 — where power becomes the binding constraint on AI deployment. The hyperscalers are signing 20-year PPAs at $80–110/MWh. The Texas grid is repricing capacity. Nuclear is being restarted at sites everyone wrote off in 2015.

VST is one of the two or three companies in America that owns a fleet big enough to matter. They have a $5B+ buyback running. They have the largest unregulated nuclear position outside of CEG. And they’re sitting in the state where the data center buildout is most aggressive.

This is a utility that trades like a tech stock and prints cash like a utility. The market is still pricing it as if power is a commodity. It’s not. Not anymore.

The five-year read: $155 → $220–280.

3. ANET — every AI ethernet build runs through them

ANET — Networking Conviction. The 800G AI ethernet socket.

Here’s the math. The 800G ethernet upgrade cycle is happening now, every hyperscaler AI cluster is being wired with it, and there are basically two companies that build the switches that matter: Arista and Cisco. Arista is winning the AI sockets. Their hyperscaler revenue is growing 25%+ and their gross margins are 60%+ on a hardware business.

The dirty secret of AI infrastructure: networking is becoming a bigger line item than compute in some clusters. When you’re tying 100,000 GPUs together, the bottleneck isn’t the GPU — it’s the fabric between them. Every dollar of AI capex that goes to a cluster has a fixed percentage that has to come out as switches, optics, and cables.

Arista is the compounder pure-play on that.

The five-year read: $172 → $260–340.

4. NBIS — the AI cloud nobody talks about until it’s $200

NBIS — Data Center Infrastructure Conviction. Cost basis $94.14, last close $141.19. +50% from cost.

NBIS is the spinoff from Yandex’s data center business. Netherlands-domiciled. Already signed a $19B Microsoft deal. Already signed a $3B Meta deal. Revenue grew 503% year-over-year last quarter. They are building out 1+ GW of AI training capacity across Finland, Israel, Kansas City.

It is the closest thing to a pure-play AI cloud in the public market that isn’t already $4T. And it is still mispriced — most of Wall Street files it as “Russia-adjacent” or “Yandex” and clicks past.

It is not Russia-adjacent. It’s a Netherlands company with European compliance and US hyperscaler customers. The discount to peers will close. It already started.

The five-year read: $158 → $300–400 if execution holds.

5. XNDU — the only quantum name on this list, and the only one I’d defend

XNDU — Frontier. Cost basis $8.29 → +292% as of the last receipt.

Most quantum stocks are lottery tickets dressed up as companies. XNDU is different. Photonic quantum hardware — room-temperature, fiber-compatible, integrable with classical AI compute in a way that ion-trap and superconducting platforms aren’t.

When Nvidia opened up CUDA-Q to support quantum-classical hybrid models, XNDU was one of three companies on the integration list. That isn’t a press release — that’s the kind of structural relationship that determines who’s still standing in 2030.

The position is up 292% from cost. Trimmed once. Not selling the rest. The Physical Layer of AI thesis ends at the physics frontier, and XNDU is how you own the physics frontier without buying ten lottery tickets that don’t ship product.

The five-year read: optionality. Could be $0. Could be $80. Sized accordingly.

The Risk

The whole thesis is one Fed pivot away from a multiple compression. AI capex doesn’t slow on a flip — it slows on a credit shock or a capacity-glut quarter from a hyperscaler. If MSFT or META guides down on capex, every name on this list trades down 15–25% before the bid comes back. That’s not the bear case. That’s the entry.

The real risk is the one nobody writes about: execution risk on power. If interconnection queues stretch from 3 years to 6, the entire data center buildout slows, and VST + ANET both feel it before the rest of the book. Watch FERC capacity auctions and the Texas ERCOT reports. That’s where the early signal comes from.

The Play

You don’t need a 21-position book. You don’t need a 50-name watchlist. You don’t need to predict the next Nvidia earnings call.

You need to own the spine.

AVGO for the cash flow. VST for the power. ANET for the wires. NBIS for the cloud. XNDU for the frontier.

That’s the Physical Layer of AI distilled into five tickers. Every agent signal, every weekly post — they orbit those five. The other 16 names are leverage on the thesis. These five are the thesis.

Young Bull Pro is live. Every trade on these five — and the 16 satellites — documented in real time. The thesis before the price.

Subscribe to get the receipts before the chart catches up.

— Young Bull

Not financial advice. Not personalized. Public companies, public filings, my own money on the line. Make your own call.