This group makes the electrical "last mile" inside a data center: the boxes that take incoming utility power, condition and protect it, and hand it off cleanly to the racks of computers. The two main products are UPS systems (uninterruptible power supplies — battery-backed units that keep servers running through the few seconds-to-minutes gap between a grid outage and the backup generator firing up, while also smoothing out voltage dips and surges) and PDUs (power distribution units — the strips and busways that split the conditioned power out to each individual server rack). The unit of demand is electrical capacity, measured in megawatts (MW): every MW of computing capacity a data center installs needs roughly a matching MW of UPS protection and a set of PDUs to feed the racks. The names that make these products are Vertiv (VRT), Eaton (ETN), Hubbell (HUBB), and, on the backup-generator side, Generac (GNRC), with Schneider Electric and Legrand as large non-US-listed competitors.
On the forecasts below, demand for UPS and PDUs is set to run ahead of easy supply for the next several years, because each new megawatt of AI compute mechanically pulls roughly a megawatt of UPS and a rack's worth of PDUs behind it. Supply is limited less by factory floors and more by shared inputs — lithium-ion battery cells (competing with electric vehicles) and power semiconductors — plus the engineering and service know-how to deploy at data-center scale. PDUs are the more commoditized, more competitive end; UPS is the higher-priced, more concentrated end. In money terms: the market currently prices the purest names (those whose revenue is almost all data-center power) at a high multiple of revenue and earnings est. — that is, several dollars of market value per $1 of current annual sales — so today's buyer pays for a stream of future sales, not for today's cash flows alone. The diversified names trade at lower revenue multiples but carry more diluted exposure to this specific product. The reader judges whether that math is worth it.
Picture the power path inside a data center as a relay race. Utility power comes in at medium voltage; switchgear and transformers (a separate group) step it down; then it hits the UPS, which holds a battery charge so that if the grid blinks, the servers never notice — the UPS carries the load for the seconds it takes a diesel or gas generator to start. From the UPS, power flows through busway (rigid copper bars overhead) or cabling to PDUs, which are the metered strips that plug into each rack. Modern data-center UPS units are modular blocks of roughly 1-3 MW each, increasingly built around lithium-ion batteries rather than the older lead-acid (VRLA) type.
How the cash actually comes in: these companies sell the hardware as engineered equipment, typically as part of a larger power-and-cooling order tied to a specific data-center build. A single 100 MW campus can carry tens of millions of dollars of UPS and distribution gear. The sale is mostly one-time equipment revenue, but it drags a long tail of higher-margin services behind it — battery replacement, monitoring, maintenance contracts, and spares — because a UPS that fails defeats its entire purpose, so operators pay to keep it serviced for its 10-15 year life. The "intelligent" PDU adds software-flavored value (per-outlet metering, remote power cycling), which lifts the average selling price (ASP — the price per unit) versus a dumb power strip. So the money-in shape is: large lumpy equipment orders up front, plus a recurring services annuity that compounds as the installed base grows.
Source: 500-stocks scan, /Users/ravf/projects/work/.claude/worktrees/sector-hub/research/investments/500-stocks/04-data-centers-infrastructure.html (UPS and PDU sub-sections).
The mechanical link. Unlike many AI-adjacent products, demand here is almost arithmetic: UPS demand scales linearly with data-center power capacity. Every new MW of compute requires a corresponding MW of UPS, and a matching set of PDUs to feed its racks. So if you believe data-center capacity is growing, you can largely read UPS/PDU volume straight off it. The scan pegs hyperscaler data-center construction at $200B+ annually and calls it "just the opening salvo."
The AGI lens. Reasoning from the premise that AGI is arriving: compute demand does not plateau, it compounds, because each smarter model generation trains on larger clusters and then gets deployed (inference) at population scale, and physical-AI/robotics adds a second wave of compute on top. The scan's own framing — "recursive self-improvement means each generation of smarter models demands larger training clusters with denser racks" — points the same way. Two demand multipliers stack on top of raw MW growth: (1) power density is rising from legacy 10-15 kW racks toward 50-100+ kW, which concentrates and increases UPS and high-amperage PDU content per rack; and (2) the shift to lithium-ion batteries raises the ASP per UPS unit even when MW are flat. Net effect: dollar demand grows faster than MW demand.
Sizing it (forecast, approximate). The scan states the data-center UPS market is growing 15-20% annually with AI as the primary driver est.. PDU/busway volume "tracks data-center construction linearly" with rising ASP per rack as density climbs. These are growth-rate statements, not contracted orders — treat them as forecasts. Hard, company-grounded evidence of live demand: the scan notes UPS lead times remain elevated (20-30 weeks in 2025), which only happens when orders exceed easy supply.
Who the buyers are. Hyperscalers (the largest cloud and AI operators) are the dominant buyers, followed by colocation providers (companies that build data centers and rent space) and large enterprises. The buyer base is concentrated, financially strong, and currently spending aggressively — which supports volume but also gives big buyers negotiating leverage on price.
✓ VERIFIED — the following figures were confirmed from primary sources after initial publication:
Current capacity and expansion. The makers are established industrial firms with global factories, and they have been adding capacity through the AI build-out. The clearest live evidence that supply is tight but improving: data-center UPS lead times fell from 40+ weeks in 2023 to 20-30 weeks in 2025 — still above the historical norm, meaning the order book is fuller than the factories can immediately clear.
The real bottleneck. For UPS, the constraint is not the metal box — it is two inputs the makers do not fully control: lithium-ion battery cells (the same cells the electric-vehicle industry buys, so the two compete for supply) and power semiconductors (IGBTs and silicon-carbide MOSFETs — the chips that switch high power). Add a third, softer constraint: the engineering and field-service talent to design, integrate, and commission systems at data-center scale. PDUs are a different story — the scan calls them "relatively straightforward to manufacture," with multiple competing vendors and manageable 8-16 week lead times; the only tight spot is newer high-density PDUs (60A+ per circuit), which have fewer qualified suppliers.
Who controls supply. UPS is concentrated: Vertiv and Eaton together hold roughly 60% of the global market est., with Schneider Electric the other major force. That concentration acts as a barrier to entry (an obstacle that makes it hard for a new competitor to take share) — capacity, customer qualifications, and field-service networks take years to replicate. PDUs are more fragmented (Vertiv's Geist, Eaton's ePDUs, Legrand, Schneider, and others), so the scan describes this end as more competitive, where more sellers compete on price.
Source: 500-stocks scan UPS and PDU sub-sections (lead times, ~60% combined share, bottleneck list). Share figure not live-verified.
Putting the two sides together: the product is currently short — demand is running ahead of frictionless supply. The evidence is the price-and-time signals rather than a published shortage number: lead times are still elevated (20-30 weeks for UPS), demand is mechanically tied to a data-center build-out the scan calls the "opening salvo," and the bottleneck inputs (Li-ion cells, power chips) are contested by other industries. None of this is a contracted, take-or-pay guarantee (take-or-pay = a contract where the buyer must pay whether or not it takes delivery) — it is a tightness inferred from lead times and stated growth, so treat the "short" conclusion as well-supported but not contractually locked.
| Factor | UPS systems | PDUs & busway |
|---|---|---|
| Demand growth | ~15-20%/yr est. | Tracks DC build, + rising ASP |
| Supply tightness | Moderate (Li-ion + power chips) | Low (except high-density) |
| Lead time (2025) | 20-30 wks | 8-16 wks |
| Lead-time direction | Improving (was 40+ wks) | Stable / manageable |
| Short or long today? | Short | Roughly balanced |
| What could flip to oversupply | Li-ion/chip supply catches up + makers add lines | Already competitive; commoditization |
When could it flip? The honest answer is that the falling UPS lead times (40+ down to 20-30 weeks) show supply is already catching up. If battery-cell and power-semiconductor capacity continues to expand and the makers keep adding lines, the UPS shortage could ease within a couple of years even as volumes grow — the tightness is a transient of a fast ramp, not a permanent geological scarcity like, say, large power transformers (which the same scan flags at 100-150 week lead times). PDUs are the more competitive end and are closest to balanced already. So the gap is real now but is structurally easier to close than the most acute parts of the power stack.
Source: 500-stocks scan (lead times, growth statements, bottleneck inputs). Forecasts labelled; not live-verified.
Sizes below are approximate, general-knowledge figures (cutoff ~early 2026), not live quotes. They are meant to convey order of magnitude — pure-play vs. conglomerate — not precise current market value.
| Company | What it makes here | Exposure to this product | Rough size est. | Position |
|---|---|---|---|---|
| Vertiv (VRT) | UPS, PDUs (Geist), busway, plus cooling, racks, modular, monitoring | Purest large play; data-center power & cooling is essentially the whole company | ~$50-60B mkt cap | Co-leader in UPS (~60% combined with Eaton); full-stack DC vendor |
| Eaton (ETN) | UPS, ePDUs, busway, plus switchgear and broad electrical | Diversified industrial; DC power is a large, fast-growing slice but not the majority of revenue | ~$130-150B mkt cap | Co-leader in UPS; deep electrical breadth across the power path |
| Hubbell (HUBB) | Switchgear, cable management, electrical connectors (adjacent to, not core, UPS/PDU) | Diversified electrical; DC is a growth slice, lighter on UPS/PDU specifically | ~$20-25B mkt cap | Electrical infrastructure breadth; less of a pure UPS/PDU play |
| Generac (GNRC) | Backup generators (the engine that the UPS bridges to), plus liquid cooling via acquisitions | Backup power, not UPS/PDU; adjacent in the same power chain | ~$8-12B mkt cap | Strong in generators (residential/commercial heritage); newer to DC scale |
| Schneider Electric (SBGSF, non-US-listed) | UPS, PDUs, switchgear, DCIM software | Diversified; major global DC power competitor | Large-cap | The third major UPS force globally |
| Legrand (LGRDY, ADR) | PDUs, busway, cabling | Diversified electrical; PDU-weighted | Large-cap | Significant PDU/busway competitor |
Source: company products from the 500-stocks scan ticker grids; market-cap ranges from general knowledge, approximate and not live-verified.
The plain-money question for an owner is: how many dollars of market value am I paying for each dollar of this year's sales (and ultimately of owner cash)? The two ends of this group are priced very differently.
The pure play (Vertiv) trades at a high multiple. As a company whose revenue is almost entirely data-center power and cooling, and growing fast, VRT trades at a high multiple of revenue and a high multiple of earnings est. — a buyer pays several dollars of market value per $1 of current annual revenue. Mechanically, that means today's price embeds a stream of future sales, not just this year's cash flows. One structural fact about the money-in/money-out shape: making UPS and PDUs is more capital-light (it ties up less money in factories and equipment) than building chip fabs or transformer plants — it is assembly and engineering rather than heavy smelting — so a growing order book converts to free cash flow (cash left after running the business and reinvesting in it) at a higher rate than a capital-heavy manufacturer would. The arithmetic an owner is weighing: at a high multiple, the embedded future growth has to materialize for today's price to be earned back.
The conglomerates (Eaton, Hubbell) are diversified industrials. Because data-center UPS/PDU is one slice of a much larger business, their overall valuation reflects the blend — a lower revenue multiple than VRT, broader and more diversified cash flows, and more diluted exposure to this specific product: a dollar invested buys less of the UPS/PDU gap and more of general electrical equipment. Eaton's EBITDA (earnings before interest, taxes, depreciation and amortization — a rough proxy for operating cash generation) and free cash flow are large and broad-based; the AI-power demand is one input among many to the total business.
Generac is the odd one out. It is a backup-power (generator) name with a residential/commercial heritage now reaching for data-center exposure; its multiple has historically swung with the home-generator cycle, so paying for it as an "AI power" play means accepting that the data-center piece is still a developing minority of the business est..
Net arithmetic, no verdict: the more direct the exposure to the UPS/PDU gap, the higher the multiple paid (VRT); the more diversified the company, the lower the multiple but the more diluted the exposure to this specific product (ETN, HUBB). All of these are established, cash-generating businesses with current revenue, not pre-revenue companies. What the reader is left to judge is whether the demand-vs-supply gap is wide enough and lasts long enough to earn back the higher multiple on the pure plays.
Source: valuation shape from general knowledge, approximate and not live-verified; product mix and competitive position from the 500-stocks scan.
For a company-level deep-dive, here is what each name lets an owner observe, stated factually. This is a map of where to look, not a ranking.
In short: VRT and ETN carry the most direct exposure to this product; HUBB and GNRC are useful mainly to gauge how diluted or adjacent their exposure is. The reader decides which is worth pursuing.
What was used:
Hard vs. approximate: Company product mixes, competitive positions, and the directional facts (UPS demand scales with MW; UPS is more concentrated than PDUs; lead times improved 2023→2025; Li-ion and power chips are the bottleneck) are grounded in the provided scan and are reasonably hard. Every quantitative figure — the 15-20% growth rate, the ~60% combined share, the specific lead-time weeks, all market-cap ranges, and all valuation multiples — is approximate and not live-verified, because live web/market data was unavailable when this was built. Treat numbers as order-of-magnitude. Forecasts (future demand, when the gap could flip to oversupply) are labelled as forecasts and are not contracted facts.