This group sells the machines that turn natural gas (and diesel) into electricity right where the power is needed. The "product" is generating capacity, measured in megawatts (MW — a unit of power; one MW runs roughly 750–1,000 average US homes est.). Three product families sit inside this group: heavy-duty industrial gas turbines (the giant jet-engine-like units, roughly 5–500 MW each est., made by GE Vernova / GEV); reciprocating engine gensets (genset = a piston engine bolted to an electrical generator, truck-engine-style, running on diesel or gas, made by Cummins / CMI, Caterpillar / CAT, Generac / GNRC); and the electrical gear around them — switchgear (the heavy-duty electrical switches and breakers that route power), UPS (uninterruptible power supply, battery-backed gear that keeps power flowing during an outage), and controls (made by Eaton / ETN, with BWXT / BWXT a nuclear-component outlier included here for breadth). The buyers used to be utilities and factories wanting backup. Now a large new buyer is the AI data center, which increasingly wants this equipment as primary, always-on power to bypass a grid that cannot connect it fast enough.
As of the live pull while building this, demand for large gas turbines is running ahead of the rate they can be supplied: roughly three firms make heavy-duty turbines at scale est., factory slots are reportedly sold out into the late 2020s est., and AI data centers are now competing with utilities for the same machines. Supply is limited mainly by manufacturing slots, multi-year lead times, and specialized know-how — not by cash or by a single scarce raw material. In money terms, the market is paying a wide range for exposure: roughly $2–3 of company value per $1 of yearly sales for the reciprocating-engine maker (Cummins) up to ~$6–7 per $1 of sales for the turbine maker with the largest installed base (GE Vernova). The reader can weigh what that spread implies; this sheet states the facts, not a verdict.
The product is on-site electricity generation hardware. It comes in two physical forms. A gas turbine is essentially a stationary jet engine: it burns natural gas to spin a turbine connected to a generator. Heavy-duty ones produce tens to hundreds of MW and are the backbone of large power plants and now of giant data-center campuses. A reciprocating engine genset is a large piston engine (like a scaled-up truck engine) coupled to a generator; these run on diesel or gas, come in smaller increments (hundreds of kW to a few MW), and are bolted together in arrays for backup or "peaking" (short bursts to cover demand spikes) power.
The companies make money three ways, and the mix matters for how durable the cash is:
Owner translation: you are buying (1) a near-term equipment order book, plus (2) a decades-long service stream per machine sold. The service stream is the part that compounds.
What is contracted/known today contracted: Per the 500-stocks energy scan, GE Vernova's heavy-duty gas turbine orders are described as at multi-decade highs, and hyperscalers (the big cloud/AI buyers such as the major cloud providers) are shifting from buying generators for backup to buying them as primary, behind-the-meter power ("behind-the-meter" = generated on the customer's own site, not bought from the public grid) — meaning the generator becomes the baseload plant (the always-running source), not the spare tire. Every MW of AI compute also still needs redundant backup (configurations described as N+1 or 2N — i.e. one or two full spare units per live unit), so each data center MW pulls multiple MW of generation equipment.
The forward picture (forecast) forecast: The driver is the AI/AGI build-out. Given that AGI is arriving, compute demand — and therefore electricity demand at data centers — rises faster than utilities can connect new load. Grid interconnection queues (the waiting line to be hooked up to the public grid) run multiple years; the data-center-power sector knowledge cites figures such as ~55-month queues and US data-center load growing from ~25 GW today toward ~50–70 GW by 2030 est.. When the grid cannot deliver power on the timeline a data center needs, the owner self-generates on-site with gas turbines or engine arrays. That converts a power-access problem into equipment demand for this exact group. The same physical-AI thesis (robotics, automated manufacturing) would raise industrial electricity demand on top.
Who the buyers are: hyperscalers and AI labs (directly or through their data-center developers), independent power producers building gas plants to sell power to those campuses, utilities adding gas peaking capacity, and traditional industrial/commercial backup buyers (the legacy Generac/Cummins base).
✓ VERIFIED — the following figures were confirmed from primary sources after initial publication:
Heavy-duty gas turbines are the genuine bottleneck. Per the scan and general knowledge, roughly three firms in the world make them at scale est.: GE Vernova, Siemens Energy, and Mitsubishi Power. Lead times are ~2–3 years est., and the binding constraint is manufacturing slots — the number of large turbines a factory can build per year — not a single scarce raw material. A 300 MW turbine cannot be produced quickly; it needs specialized casting, precision machining, skilled labor, and a long test cycle. Adding a new turbine line takes years and large capex (capex = capital expenditure, the upfront cash to build plant and tooling). That is why slots are reportedly sold out well into the late 2020s est. and why, per the scan, pricing power has shifted toward the makers.
Reciprocating engines are tight but not as scarce. Cummins, Caterpillar, and Generac can scale piston-engine genset output faster than turbine lines. The scan describes this supply as "less constrained but still tight," because the same engines are demanded simultaneously by data centers, grid peaking, and traditional backup. So lead times have stretched, but there is no three-supplier chokepoint.
Market-share structure (who controls supply) est.: In heavy-duty turbines, the roughly three makers split the global market, with GE Vernova holding the largest installed base globally per the scan — meaning the biggest existing fleet to service, which itself raises switching costs for customers. In reciprocating gensets the field is broader (Caterpillar and Cummins large in big engines; Generac historically residential, now pushing into commercial/industrial and grid-scale storage). BWXT sits outside both — it makes naval nuclear reactors and SMR (small modular reactor — a compact factory-built nuclear unit) components, included here only as an adjacent on-site/long-duration generation option, not as a gas-turbine maker.
As of this build, the product is short — demand exceeds the rate at which it can be built, most acutely in heavy-duty turbines. The evidence is physical: a handful of makers, ~2–3 year lead times est., factory slots reported sold out est., and an order pace described as at multi-decade highs (all per the scan or general knowledge). When a product is sold out years forward, the maker tends to set price and the buyer waits — the cash signature of a shortage.
| Sub-product | Demand trend | Supply state | Short or long? | When could it flip to oversupply? |
| Heavy-duty gas turbines (GEV) | Rising (AI primary power + utility peaking) | ~3 makers, slots sold out, ~2–3 yr lead est. | Short | Plausibly only after new turbine lines finish (multi-year builds) AND AI load growth slows forecast |
| Reciprocating gas/diesel gensets (CMI, CAT, GNRC) | Strong (backup + behind-the-meter) | Tight, but scalable faster | Short (mild) | Sooner than turbines — engine output can ramp; a demand pause could close the gap quicker forecast |
| Electrical gear / switchgear (ETN) | Rising (every MW needs it) | Moderate lead times | Short | Tied to overall data-center build cadence forecast |
The oversupply risk to keep in view: this is a cyclical capital-goods group. If the AI build-out pace slows, or if grid interconnection catches up, today's sold-out backlogs could normalize and the makers could lose pricing power. On lead-time logic, turbines (longest lead time, fewest suppliers) would stay short longest; reciprocating engines and gear would loosen first. Nothing in the provided files dates a flip — treat the timing as an open forecast.
Figures below: live price/market-cap pull and trailing-twelve-month (TTM) revenue done while building this; revenue-mix percentages for this specific product are approximate est. because segment detail was not in the provided files. Market caps and revenues are stated as rounded ranges to avoid false precision.
| Company | What it makes | Exposure to THIS product | Size (live pull) | Position |
| GE Vernova (GEV) | Heavy-duty gas turbines, plus grid gear and wind | High; gas power + electrification are the core, but the company also has grid and (lower-margin) wind est. | ~$260B cap · ~$39–40B rev | Largest installed turbine base globally (per scan); one of ~three heavy-duty makers; large service stream. Most concentrated large-turbine exposure in the group. |
| Cummins (CMI) | Diesel/gas engines, gensets, power systems; also truck powertrains | Medium; power-generation gensets are one segment of a broader engine company est. | ~$90–95B cap · ~$34B rev | Major reciprocating-engine supplier to data centers; diversified across on-highway engines, so AI power is a slice, not the whole. |
| Generac (GNRC) | Backup generators (residential → commercial/industrial), storage | Medium and shifting; historically residential backup, moving into commercial/industrial and grid-scale est. | ~$16–17B cap · ~$4B rev | Smallest of the group; most tied to the backup/distributed niche; the commercial/industrial and storage shift is the swing factor. |
| Eaton (ETN) | Electrical gear: switchgear, UPS, power distribution, controls | Adjacent; sells the gear around generation, not the generator. Large data-center exposure across the power chain est. | ~$160B cap · ~$29B rev | Broadest electrical exposure; sells into every MW regardless of who makes the generator. |
| BWXT (BWXT) | Naval nuclear reactors, SMR components, nuclear fuel | None to gas turbines; an adjacent long-duration on-site generation option (nuclear) est. | ~$16–17B cap · ~$3–3.5B rev | Outlier in this group; relevant only as a different answer to the same "firm on-site power" question. |
Source: 500-stocks energy scan (sectors 08 "Diesel/Gas Generators & Backup Power," 07 "Grid Equipment," 03 nuclear/SMR) at /Users/ravf/projects/work/.claude/worktrees/sector-hub/research/investments/500-stocks/01-energy-power.html; live price/market-cap/revenue pull (yfinance) done while building this.
Plain-money framing: "price-to-sales" (P/S) means how many dollars of company value (market cap) you pay per $1 of this year's revenue. It is a rough rent-multiple on the top line, not a measure of profit. Using the live pull while building this (rounded):
On profit: the trailing price-to-earnings (P/E — dollars of market value per $1 of last year's net profit) from the same pull runs roughly GEV ~28, CMI ~35, ETN ~41, BWXT ~50, GNRC ~89 (rounded, point-in-time). In plain terms, the market is currently paying the most per dollar of profit for the smaller/niche names (GNRC, BWXT) and less for the largest-installed-base turbine maker (GEV) and the diversified engine maker (CMI). These are point-in-time ratios reported as-is; the reader draws the conclusion.
Money-in / money-out shape: this is a capital-goods group, not a capital-light one. The makers spend real cash on factories, tooling, and working capital (parts and half-built machines sitting on the floor) to fill backlog — money-out comes first. The owner cash shows up later and more durably through the higher-margin service stream on the installed fleet. GE Vernova's owner-cash therefore depends heavily on converting its backlog and on its large service base; the reciprocating-engine makers (CMI, GNRC) generate cash on a shorter, more product-sale cadence; Eaton sits closest to a steady, broad-based electrical-products cash engine. None of these are utilities — there is no guaranteed regulated return; cash depends on the order cycle holding.
Source: live yfinance pull (price, market cap, TTM revenue, trailing P/E and P/S) for GEV, CMI, GNRC, ETN, BWXT done while building this report.
Where a company-level deep-dive would be most informative, factually:
Hard / grounded facts (used directly): the 500-stocks energy scan section — sector 08 "Diesel/Gas Generators & Backup Power," with supporting context from sector 07 "Grid Equipment" and sector 03 (nuclear/SMR) — at /Users/ravf/projects/work/.claude/worktrees/sector-hub/research/investments/500-stocks/01-energy-power.html. From it: the multi-maker turbine structure, GEV's largest-installed-base and multi-decade-high order claim, the ~2–3 year turbine lead time, the backup-to-primary demand shift, and the company roster. Live price, market cap, TTM revenue, P/S and P/E for GEV/CMI/GNRC/ETN/BWXT/CAT came from a yfinance pull done while building this report (point-in-time, rounded here).
Approximate / not-live-verified (general knowledge, cutoff ~early 2026, flagged with the est. tag and the unverified note above): the precise number of global turbine makers being exactly three, "factory slots sold out into the late 2020s," the GW-scale data-center demand trajectory, homes-per-MW, per-company revenue-mix percentages for this specific product, and any market-share split. No prior deep-dive report exists for this group; the data-center-power sector knowledge file informed the demand framing but its grid-queue figures are themselves dated. Treat every figure carrying the est. tag as directional, and confirm backlog/segment specifics against the latest 10-K/10-Q (annual/quarterly SEC filings) before relying on them.
Source: /Users/ravf/projects/work/.claude/worktrees/sector-hub/research/investments/500-stocks/01-energy-power.html (sectors 03, 07, 08); live yfinance quote pull; general knowledge (cutoff ~early 2026, labelled approximate where used).