The Other Bottleneck: Who's Actually Selling The Power? A Fundamentals-Only Map Of The AI Power Chain

The "picks and shovels" framing for AI infrastructure has a blind spot: it mostly stops at the rack. Chips, networking, and rack-level power/cooling get the headlines, but every one of those racks still needs to be plugged into something — generation capacity, transmission and distribution, and the electrical contractors who physically build the interconnection. That's a different chain, with a different capital structure, and — as this piece will show — a materially different resilience profile than the one we mapped in [Who's Actually Selling the Shovels?].

This is a direct sequel, not a rehash. Two names carry over from that piece — ETN and VRT — because they sit at the boundary between "rack-level power" and "grid-level power," and dropping them here would leave a gap in the chain. Everything else is new. The question is the same one we've been asking all series: not where the cycle is headed, but what the numbers already on the books say about who is capturing the buildout right now, and how much of that capture is already priced into DUEL's own valuation math.

What do already-filed 10-K/10-Qs say about who, in the US-listed "power for AI" chain, is currently capturing growth, margin, and balance-sheet strength — and how much of that is already reflected in DUEL's DCF math, with no load-growth forecast involved?

Part 1 — Building the Basket

Powering an AI data center is not a single business. It's at least three: the company that generates or procures the electricity, the company that makes the switchgear, transformers, and thermal management equipment sitting between the grid and the servers, and the company that physically builds and wires the substations, transmission lines, and electrical rooms connecting the two. We built a ten-company basket across those three layers, plus the same four hyperscale buyers used as a control group in the prior piece:

• Generation / Utilities: VST, CEG, NEE, SO

• Power / Industrial (grid-to-building electrical equipment): ETN, VRT, EMR, ROK

• Construction / Electrical & T&D Contracting: PWR, EME

• Buyers (control group, not scored as "beneficiaries"): MSFT, GOOGL, AMZN, META

That's ten unique supply-side names and four buyers — 14 companies in total, each run through DUEL's Battle Report, DCF Report, and Resilience Report exactly like any other duel on the site. The ten new duels behind this piece were generated on August 17, 2026. The four buyer profiles are reused from the July 26–August 13, 2026 window used in the prior piece — DUEL's per-company DCF and Resilience outputs are opponent-independent, so this does not affect comparability, but the underlying filing snapshots are a few weeks apart, which is worth naming plainly.

Part 2 — Growth and Margin: Where the Capital Is Actually Landing

The first thing this table shows is that "power" is not one growth story — it's at least three uncorrelated ones. The generation/utility layer is barely growing on a trailing basis: VST (4.31%), CEG (1.47%), SO (0.31%), and NEE (-3.05%) — a regulated, heavily contracted, capital-cycle business where revenue simply does not move the way chip or networking revenue does, regardless of how much incremental demand is showing up in interconnection queues. The power/industrial equipment layer is the opposite: VRT is growing revenue at 21.59%, closer to the AI-infrastructure names in the prior piece than to its own sector peers, while ETN (9.77%) and ROK (2.44%) grow far more slowly, and EMR is outright shrinking (-2.82%) on a trailing basis. The construction/T&D layer sits in between but skews high — PWR (18.60%) and EME (15.32%) — consistent with a physical buildout boom in the segment of the chain that actually pours concrete and pulls cable.

Margin tells a second, separate story. VRT's 18.51% FCF margin and 27.99% ROIC look like an infrastructure-supplier margin profile, not a utility one — it is, structurally, closer to the compute/networking layer from the prior piece than to its own "power" label. SO's FCF margin is negative (-9.93%): CAPEX is currently running ahead of operating cash flow, a pattern that recurs across the generation layer and that Part 3 below shows has real consequences for DCF modeling. NEE's operating margin of 58.08% looks like an outlier worth flagging rather than trusting at face value — regulated-utility accounting (allowed-return recovery, deferred regulatory assets) can produce operating-margin figures that are not comparable, factor-for-factor, to an industrial or software company's operating margin, even though the underlying formula (Operating Income / Revenue) is applied identically. We surface it rather than smooth it over.

Part 3 — How Much Growth Is Already in the Price

This is where the two halves of this basket diverge sharply — and where the model's own assumptions become part of the finding. DUEL's DCF engine builds fair value and a 3-year price projection from each company's own filings alone: no market price as an input, no analyst growth estimates, just revenue CAGR, FCF, ROIC, and Sloan Ratio run through a fixed formula (full methodology in our earlier [DCF confidence band] piece).

Seven of ten names produce an ordinary, interpretable upside figure, ranging from VRT's 11.5% (the same figure as in the prior piece — confirming the model's per-company outputs really are opponent-independent) up to VST's 41.7%. But three names — CEG, NEE, and SO, all in the generation/utility layer — don't. This is the headline finding of this piece, and it deserves to be stated precisely rather than smoothed into a "low upside" bucket.

What's actually happening: DUEL's DCF defines FCF as CFO − CAPEX. For all three of these companies, current CAPEX is large enough relative to operating cash flow (and, in SO's case, actually exceeds it) that projecting this FCF base forward under the model's fixed formula pushes modeled enterprise value below total debt. The model does not report a negative fair value; it floors displayed equity value at $0.00 and flags the case explicitly ("total debt exceeds modeled enterprise value... signals significant balance-sheet risk, not a precise valuation"). SO's case is the most extreme: its modeled 3-year price trajectory is negative ($0.00 → -$2.67 → -$5.34 → -$8.03), because its FCF base itself is negative and the model has no mechanism to distinguish "temporarily negative FCF during a regulator-approved buildout cycle" from "structurally cash-burning business." Both would produce the same output.

This is not a claim that Vistra, Constellation, NextEra, or Southern are overleveraged or poorly run — regulated and quasi-regulated generation utilities routinely finance multi-year capital programs with debt against a rate base or long-term contracted cash flows that a pure trailing-FCF DCF was never built to see. It is a finding about the model, applied honestly: a discounted-cash-flow engine calibrated on CFO − CAPEX, with no rate-base, regulatory-asset, or dividend-discount treatment, structurally cannot produce a usable fair value for this kind of balance sheet. We'd rather report that limitation clearly than force a number that implies more precision than the model can support.

Part 4 — Who Actually Has the Balance-Sheet Armor

This table is the sharpest contrast with the prior piece. In the chips/networking/cooling basket, 55% of companies cleared all three internal consistency checks with a clean OK/OK/OK. Here, only two of ten do — VRT, and EMR with a caveat below. Every one of the four generation/utility names carries a Debt-vs-Cash flag, and three of the four carry the more severe RISK level rather than CAUTION. This lines up directly with Part 3: the same debt-financed capital intensity that floors CEG, NEE, and SO's DCF fair values at $0.00 is also what's driving their Debt-vs-Cash flags here. The two findings are not independent discoveries — they are the same underlying balance-sheet structure showing up in two different parts of DUEL's model.

EMR's headline FRI of 100 needs its own caveat, in the same spirit as GEV's in the prior piece. Its Resilience Report shows several raw balance-sheet components — Cash, Total Debt, Operating Income, Equity — as unavailable ("—") in the filing period captured, leaving only the Investment Self-Sufficiency (FCF/CAPEX) sub-score computed; the other three sub-metrics (Cash-to-Debt, Earnings Quality, Liquidity Runway) could not be scored at all. A 100 built on one populated sub-metric out of four is a different kind of 100 than VRT's, which is built on four fully populated sub-metrics that all scored cleanly. The most likely explanation is a fiscal-calendar mismatch — Emerson's fiscal year ends in September rather than December, which can leave certain XBRL-tagged fields sparsely populated relative to calendar-year peers at any given snapshot — but until a filing period with complete data is captured, EMR's FRI should be read as provisional rather than definitive.

PWR and EME both carry a single CAUTION on Growth vs. FCF: both are growing revenue in the mid-to-high teens (18.60% and 15.32%) while FCF margin sits in the mid-single digits (5.69% and 6.77%) — a straightforward cash-conversion-lag flag typical of contractors scaling headcount and working capital ahead of billings, not an accounting-quality concern.

Part 5 — The Scorecard: Upside vs. Resilience

Putting Parts 3 and 4 on one page sorts the seven names with an interpretable DCF output into a 2×2 grid. The split point on each axis is the median across those seven names only (upside median 17.0%, FRI median 68) — CEG, NEE, and SO are excluded from this grid entirely rather than forced into a quadrant their modeled fair value doesn't actually support; they get their own category below.

Attractively priced and structurally sound: EMR, PWR (both above-median FRI; PWR's upside sits almost exactly at the 17.0% median line)

Modeled upside, thinner buffer: VST, ROK (above-median upside, below-median resilience)

Already priced in, but rock-solid: VRT (below-median upside, but the highest fully-populated FRI in the basket)

Already priced in, comparatively thinner buffer: EME

Sitting exactly on the crossover: ETN — its 17.0% upside and 68 FRI land almost precisely on both median lines, making it this basket's pivot point rather than a clean member of any quadrant.

Model-floor cases (excluded from the grid): CEG, NEE, SO — see Part 3. All three carry a Debt-vs-Cash RISK or CAUTION flag in Part 4 and a $0.00 modeled fair value in Part 3; read together, this is the single clearest signal in this basket, and it belongs to the generation/utility layer specifically, not to power/industrial or construction/T&D.

The contrast with the prior piece's scorecard is itself a finding. In Shovels, only two of eleven names landed in the least-favorable quadrant, and none of the eleven broke the DCF model outright. Here, three of ten names break the model outright before a quadrant conversation is even possible, and they're concentrated in exactly one layer of the chain — generation. The power-for-AI story, at the fundamentals level, is not one story; it's a resilient equipment/contracting layer sitting on top of a highly levered generation layer.

Part 6 — The Control Group: What the Buyers Look Like

All four hyperscale buyers land in DUEL's HIGH resilience tier — the same result as in the prior piece, and the same set of numbers, reused rather than re-run (see Part 1's note on the reused snapshot window). What's different this time is the contrast on the other side of the transaction. In Shovels, the eleven chip/networking/cooling suppliers ranged from FRI 11 to 100 with a median of 75 — close to the buyers' own 81–100 range. Here, the ten power-chain suppliers range from FRI 42 to 100 with a median of 68, and three of them don't clear DUEL's DCF model at all. The buyers financing this build sit on dramatically stronger balance sheets, by this measure, than the generation-side companies actually building the power for it.

Part 7 — What This Actually Answers, and What It Doesn't

To be precise about the limits of this exercise, since overclaiming here would undercut the point of doing it at all: this piece does not say, and cannot say, whether US grid capacity will keep pace with AI-driven load growth, whether specific generation or transmission projects will get built on schedule, or whether any of these companies' capital plans will succeed. Those are forward-looking questions, and every figure above describes only the past and present.

What the data does show, on its own terms: growth and margin in this chain are genuinely three separate stories layered on top of each other — a slow-growing, low-margin-variance generation layer; a fast-growing, high-margin equipment layer; and a fast-growing, thin-margin construction layer. It also shows that resilience, measured the same way across all fourteen companies in this and the prior piece, is not evenly distributed across this chain — it's concentrated in the equipment layer and largely absent from the generation layer, where DUEL's DCF model breaks down entirely for three of four names. That is a statement about the balance-sheet structure of these specific companies today, processed through one specific model, not a prediction about tomorrow's power market.

Limitations

Three of ten names produced no usable DCF fair value. CEG, NEE, and SO all hit DUEL's model floor (displayed equity value of $0.00, full detail in Part 3). This reflects a mismatch between the model's fixed CFO-minus-CAPEX FCF definition and the debt-financed, rate-base-driven capital structure typical of regulated and quasi-regulated generation utilities — not a claim about those companies' actual financial health or investment merit.

EMR's Resilience score is built on incomplete raw data, flagged directly in Part 4 — three of four sub-metrics could not be computed in the captured filing period, most likely due to a fiscal-year-end mismatch (September for Emerson vs. December for most peers). Its FRI of 100 should be treated as provisional until a filing period with complete balance-sheet data is captured.

Snapshot, not a live ranking, and not a single snapshot window. The ten supply-side duels were generated on August 17, 2026. The four buyer profiles are reused from the July 26–August 13, 2026 window used in the prior piece in this series; DUEL's per-company outputs don't depend on the opponent, but the underlying filing snapshots are several weeks apart.

No market price, deliberately — and this cuts both ways. DUEL's DCF engine deliberately ignores current share price and analyst estimates, which is exactly what makes "how much growth does a company's own trajectory already assume" a clean, checkable question. It also means the "modeled upside" here is not a market-implied expectation and should not be read as one.

Basket construction is a judgment call, not a census. Ten supply-side names and four buyers are a reasonable cross-section of the US-listed AI power chain, not an exhaustive one; a different, equally reasonable basket could shift which names land in which part of this analysis.

Fourteen companies is a diagnostic snapshot, not a statistical sample — read the patterns above as descriptive, not as evidence generalizable beyond this specific basket.

Bottom Line

Across ten companies supplying power for AI infrastructure and four of the hyperscale buyers financing it, growth and margin split cleanly into three separate stories by layer — slow, capital-heavy generation; fast, high-margin equipment; fast, thin-margin construction. Resilience splits even more sharply: the equipment layer looks like the strongest part of the entire AI-infrastructure chain examined across both pieces in this series so far, while three of four generation/utility names carry a genuine leverage flag and none of the three produce a usable DCF fair value under DUEL's model. None of this is a forecast about grid buildout, load growth, or which projects get financed. It's what the filings already say, run through the same formula for every company in the room — including an honest account of where that formula stops working.

Every figure in this piece traces back to a duel you can pull and re-verify yourself on duelstocks.com.

Further Reading — Same Series, Same Method

This piece is a direct sequel to, and reuses methodology and buyer-side data established in, the following pieces in this series:

• Who's Actually Selling the Shovels? A Fundamentals-Only Map of the AI Infrastructure

• Absolute vs. Relative Valuation: A Tale of Two Questions

• Beyond the Pair: Ranking a Full Sector with Pairwise Comparisons

• A Point Estimate Isn't a Forecast: Building a Confidence Band for DCF

• How Fragile Is a Relative Stock Score?

• How Many Independent Signals? A Principal Component Analysis

• The Flag Nobody Talks About: A Base-Rate Analysis of Internal Contradictions

References

Damodaran, A. (2012). Investment Valuation: Tools and Techniques for Determining the Value of Any Asset (3rd ed.). Wiley.

Damodaran, A. (2018). Valuing regulated and capital-intensive businesses: why standard DCF assumptions break down for rate-based utilities. NYU Stern School of Business working note.

U.S. Securities and Exchange Commission — EDGAR full-text search and structured XBRL filing data, sec.gov.

A DuelStocks methodology deep dive. Not investment advice. All data sourced from public SEC EDGAR filings.

A DuelStocks fundamentals deep dive. Not investment advice — and, like the rest of this series, not a forecast either. All figures below come from public SEC EDGAR filings (10-K/10-Q), processed by DUEL's Battle, DCF, and Resilience (STR) reports.