The bottleneck is real.
The trade is not ready.
AI can finance chips faster than the physical power system can add generation, transmission, electrical equipment, cooling, and interconnection. That is a strong observation. It is not yet a reason to buy these stocks at any price.
Do not force a basket because the theme is persuasive. Capital moves only after the security-level work shows that expected return compensates us for valuation, cycle, and execution risk. Until then, QQQ remains the unchanged opportunity-cost comparison.
No basis, weights, or result are sealed. The page is an internal underwriting record: what we believe, what we do not know, and the exact work required before a Call can begin. HOUSE is a contestant, never the referee.
INVESTMENT CONCLUSION FIRST
Right observation. Unproven edge.
Power availability is becoming a binding constraint on new compute capacity.
The companies are exposed to the bottleneck, but they do not carry the same economics or risk.
The current page does not yet prove that expected returns exceed what prices already imply.
No HOUSE Power basket until at least two solver names clear. One passing name belongs in its own Call.
From a future seal date, an equal-weight basket of the eligible power-solvers should outperform QQQ held unchanged over 36 months because time-to-power remains scarce and those companies convert the scarcity into durable cash flow faster than the market expects.
ATTRIBUTION RAIL. Ribbit Capital supplied the Make / Move / Store / Sell framework and the bottleneck idea. Ribbit did not recommend these securities. The security map, underwriting hurdles, comparison, horizon, and challenges are HOUSE judgments.
HOW THE THESIS MAKES MONEY
Follow the cash, not the megawatts.
Compute demand grows faster than efficiency reduces electricity per unit of useful work.
Generation, grid connection, switchgear, transformers, cooling, labor, and permits remain constrained.
Customers pay for certainty, speed, capacity, and execution—not merely for more equipment.
Orders become revenue, margins, free cash flow, and returns on incremental capital.
A break anywhere can leave the macro story intact and destroy the investment return.
- Demand must be real. Announced campuses, queue requests, and capex plans cannot be counted as identical evidence.
- Scarcity must last. A multi-year thesis needs more than one strong order cycle.
- Economics must accrue to suppliers. Revenue growth without margin, cash conversion, or return on capital is not the thesis.
- Price must leave room. A great business bought above a plausible value range can still be a bad trade.
MAP THE SECURITIES
Four solvers. One adjacent scarcity trade.
These are not five independent bets. PWR, ETN, GEV, and VRT share a data-center and grid-capex factor. CEG is different: it benefits when scarcity persists rather than primarily getting paid to remove it.
Self-perform labor and infrastructure execution across transmission, substations, interconnection, and mission-critical facilities.
WHAT MUST PROVE · BACKLOG CONVERTS WITHOUT MARGIN OR WORKING-CAPITAL DAMAGEElectrical distribution and power-management equipment between the grid, building, cooling system, and rack.
WHAT MUST PROVE · ORDER GROWTH OUTLASTS CAPACITY ADDITIONS AND MIX STAYS PROFITABLEGas turbines, grid equipment, and services. The broadest direct exposure to new firm generation and electrification.
WHAT MUST PROVE · LONG-CYCLE ORDERS BECOME NORMALIZED CASH, NOT CUSTOMER-FUNDED WORKING CAPITALPower and thermal infrastructure that determines how much useful compute a delivered watt can support.
WHAT MUST PROVE · GROWTH SURVIVES CUSTOMER CONCENTRATION, NEW CAPACITY, AND MULTIPLE COMPRESSIONPRICE IS PART OF THE THESIS
Strong fundamentals. Demanding starting prices.
This is a first screen, not a valuation conclusion. It shows why the observation cannot be promoted directly into a trade. Adjusted metrics differ by company and are not comparable without reconciliation.
| NAME | AUG 13 CLOSE | CURRENT COMPANY GUIDE | ROUGH SCREEN | WHAT THE PRICE DEMANDS |
|---|---|---|---|---|
| PWR | $672.78 | 2026 adjusted EPS $16.45-$16.95 | 40.3x midpoint | Sustained compounding plus clean conversion of a $53.4B backlog. GAAP EPS guidance is materially lower. |
| ETN | $453.33 | 2026 adjusted EPS $13.40-$13.60 | 33.6x midpoint | Electrical growth must stay well above normal industrial growth while margins remain near peak levels. |
| GEV | $1,049.42 | 2026 revenue $45.5-$46.5B; FCF $11.5-$12.5B | ~23.3x guided FCF | The FCF guide includes major working-capital benefit. We need normalized cash and segment value, not a headline multiple. |
| VRT | $287.07 | 2026 adjusted EPS $6.65-$6.75 | 42.8x midpoint | Thirty-percent organic growth must persist long enough to offset normalization in growth and valuation. |
| CEG | $278.64 | 2026 adjusted operating EPS $11.50-$12.50 | 23.2x midpoint | Scarcity rents must survive regulation and integration risk. This is a separate generation Call. |
| QQQ | $732.07 | Doing Nothing | Benchmark | The candidates must beat this after taking concentration, cycle, and execution risk. |
PRICE BASIS. Settled closes for August 13, 2026, retrieved August 14. These are research observations, not sealed Call bases.
GUIDANCE SOURCES. PWR Q2 · ETN Q2 · GEV Q2 · VRT Q2 · CEG Q2.
ROUGH-SCREEN MATH. Price divided by the midpoint of company-defined adjusted 2026 EPS. GEV uses market value from 266.3M June 30 shares divided by guided FCF midpoint; this is especially poor as a normalized value measure.
THE CAPITAL RULE
We buy securities, not themes.
For each solver, estimate 2028-2029 revenue, margin, cash conversion, capital needs, dilution, and a defensible exit value.
Calculate what growth and terminal value today's price already requires. The question is not “is 40x high?” but “what must be true for 40x to work?”
Base-case annualized return must exceed the QQQ base case by at least 3 percentage points, with bear-case total downside no worse than 25%.
If two or more solver names pass, equal-weight only the eligible names. One passing name becomes a separate single-name Call; it cannot validate a basket.
Broad, diversified ways to monetize physical execution and electrical control. Cleaner expressions do not mean attractive prices.
More direct exposure to the capacity build, with greater normalization, concentration, and valuation sensitivity.
A scarcity beneficiary. Never use it to make the solver basket look diversified.
No HOUSE Power allocation. This is the research benchmark, not a statement of actual holdings.
The 3-point / 25% rails are proposed underwriting rules, not sealed methodology. They make the current decision auditable. Howard's methodology must approve or replace them before a Call can lock.
IS THE CALL REALLY BETTER THAN DOING NOTHING?
Grade the money and the thesis separately.
The mechanism is physical and multi-year. Six-, 12-, 18-, 24-, and 30-month checks observe evidence; they do not decide the Call.
It captures the opportunity cost of owning the obvious liquid AI exposure without selecting power bottleneck winners. XLI or XLU may add context; neither replaces the sealed grade.
Return of one starting dollar in the fixed basket versus one starting dollar in QQQ, using the same close and corporate-action treatment.
Show every constituent's contribution, the median excess return, and how many names beat QQQ. One winner can carry the money grade while breadth fails.
Hold the starting lots. Rebalancing would introduce a second, unregistered timing strategy. Drift remains visible rather than quietly corrected.
Demand, duration, supplier economics, and conversion are graded independently. A correct mechanism never rescues a losing money verdict.
CHALLENGE THE THESIS
The strongest attack is not demand. It is expectations.
The models cannot vote on a moving thesis. Once methodology, valuation cases, and the evidence packet lock, each model receives the same material blind and before the outcome. HOUSE remains outside their tally.
What if bookings are not load?
Forecast campuses may be duplicated, speculative, delayed, or never energized. Chip and model efficiency may outrun useful-demand growth.
What if scarcity rotates?
Generation, grid reform, behind-the-meter supply, and equipment capacity can ease one bottleneck while creating another. Yesterday's constraint may not own tomorrow's margin.
What if backlog is not cash?
Cancellation rights, fixed-price exposure, labor, project complexity, customer advances, and working capital can separate orders from economic profit.
What if all of this is priced?
At roughly 34-43x adjusted current-year EPS for three solver names, good execution may only defend the current price. Outperformance requires results above expectations, not merely above last year.
REALITY GRADES
Evidence must travel from demand to cash.
Central estimate for data centers' share of U.S. electricity use in 2030; the official range is 9.5%-15.3%.
DOE / LBNL -> DELIVERY BASELINE2,060 GWGeneration and storage actively seeking U.S. interconnection at the end of 2025.
LBNL QUEUED UP -> PWR · Q2 2026$53.4BTotal backlog. Quanta separately reports $33.6B of remaining performance obligations.
QUANTA Q2 RESULTS -> ETN · Q2 2026+41%Electrical Americas 12-month rolling average orders; total backlog was up 33% year over year.
EATON Q2 RESULTS -> GEV · Q2 2026116 GWGas Power equipment backlog and slot reservation agreements, up from 100 GW.
GE VERNOVA Q2 -> VRT · FY2026 GUIDE+31%Organic sales growth expected at the midpoint of raised full-year guidance.
VERTIV Q2 RESULTS ->These metrics are evidence of activity, not interchangeable proof. Backlog, orders, guidance, queue volume, and load forecasts have different cancellation, timing, accounting, and conversion risks.
The hard questions stay visible before the basis exists. Final definitions and sources must freeze with the methodology packet.
- DEMAND BREAKS. The official 2030 central data-center electricity-share estimate falls at least 25% from the 11.8% baseline.
- CAPEX BREAKS. At least three of Microsoft, Alphabet, Amazon, and Meta lower forward AI or data-center capital-spending guidance in two consecutive reporting seasons.
- CONVERSION BREAKS. At least three solver names show two consecutive quarters of relevant order or backlog contraction together with margin or cash-conversion deterioration.
- SCARCITY BREAKS. Lead times, interconnection delays, or pricing weaken enough that supplier growth normalizes before the underwriting case expects.
- THE TRADE LOSES. At the sealed 36-month horizon, the fixed eligible basket fails to outperform QQQ. The money verdict stands even if the physical thesis survives.
WHAT REMAINS BEFORE CAPITAL
The work is specific. None of it is ceremonial.
Revenue, margin, free cash flow, capital intensity, dilution, exit value, and reverse-implied assumptions.
No five-name minimum. No prestige allocation. No position because the theme needs a representative.
QQQ, same close, 36 months, no rebalance, money plus breadth, and explicit corporate-action rules.
Every model sees the same facts, valuation cases, bear case, and unanswered questions before any outcome.
No backfilled basis. If prices move before the work finishes, rerun the underwriting.