The AI Power Boom Is Skipping the Utilities Sector
Dozens of AI power-infrastructure IPOs are filing, but none in Zorro's pipeline carry a Utilities sector tag. Here's the classification gap behind that.
The demand story behind AI power is, by now, familiar enough to recite from memory: data centers are a genuinely new and fast-growing source of electricity load, and the grid has to absorb it. JPMorgan strategists have projected that data centers could account for roughly 11% of US electricity consumption by 2030, up from about 5% in 2024, on overall demand growth of around 2.5% a year. The Energy Information Administration's September 2026 outlook puts total US electricity sales at roughly 4,135 billion kilowatt-hours for the year, nearly 2% above 2025, with commercial and industrial load — data centers chief among them — doing most of the pushing. Morningstar has estimated close to $1.4 trillion of utility capital investment between 2026 and 2030 tied to this buildout. None of that is in dispute.
What doesn't follow automatically from any of it is that the money shows up in the GICS Utilities sector. Checking the sector-weight trend on the rotation model for "AI power demand" is a reasonable instinct, but it's largely watching the wrong account — and the mismatch is visible directly in the kind of company actually going public to build this capacity.
What the pipeline says, company by company
Pull up the IPO pipeline and look specifically at the names tied to AI power and data-center buildout: SB Energy is tagged Energy / Data Center Infrastructure. Aggreko is Power Infrastructure. CoVolt Power is Solar / Battery Storage EPC. ProEnergy is Power Infrastructure / Gas Turbines. Firmus Technologies, pricing its ASX listing this month, is AI Data Centers. Switch and DayOne Data Centers are both tagged Data Centers. Nscale is AI Cloud / GPU Infra. That's eight separate filings or recent listings, each one a direct bet on the same electricity-demand story driving the headline numbers above — and not one of them carries a Utilities tag. Scan the rest of the pipeline and the pattern holds: across dozens of entries, zero are classified Utilities.
That's not a labeling oversight. It reflects something real about how this specific capacity is getting built and financed.
Why GICS doesn't catch the money
GICS Utilities is built around regulated and quasi-regulated electric, gas, and water holding companies — the kind of business where capital spending flows through a rate base that a public utility commission approves, and where returns are smoothed by regulation rather than set by a merchant market. That structure is exactly what most of the new AI-driven generation and data-center capacity is avoiding. SB Energy pairs power generation directly with the data-center campuses it serves, financed in part by a $3 billion Nvidia commitment rather than a utility's regulated balance sheet. Aggreko, majority owned by TDR Capital and I Squared Capital, rents modular generation and cooling equipment outside any rate-base framework entirely. CoVolt Power is an engineering and construction contractor, not a power owner. Firmus Technologies is backed directly by Nvidia and Blackstone equity, not ratepayer-funded capex.
Each of these is a merchant or private-capital vehicle sitting outside the regulated-utility structure GICS was built to classify — which is precisely why the sector shows up in Industrials, Energy, or a standalone "data center" bucket instead of Utilities, even though the underlying demand driver is identical to the one utility analysts cite. The capital is real and it is going into power infrastructure; it's just not booking into the sector an index-level weight chart would show you.
The sector didn't rally on the story either
The gap isn't only definitional — it shows up in performance too. Despite the demand backdrop, utility stocks have recently lagged the broader market, according to Morningstar's September 2026 update, which described the sector as slightly undervalued following a pullback earlier in the year even as near-term earnings growth estimates stayed firm. That's a strange outcome if the AI-power trade were simply flowing into GICS Utilities the way a sector-rotation narrative would suggest. It fits more easily once you separate the story into two different exposures: the handful of regulated utilities with direct, disclosed hyperscaler power-purchase agreements or transmission build-outs serving these loads, and the much larger set of merchant generation, EPC, and data-center companies capturing the same demand outside the regulated sector entirely. Headlines tend to average the two into one "AI power trade," which is also why a single week of utility-stock whipsaw doesn't settle into a clean trend even when the underlying demand mechanism is real.
Reading this correctly in practice
The practical upshot is that a sector-weight chart and an IPO pipeline are answering different questions, and neither substitutes for the other. GICS sector weight tells you how the existing, regulated incumbents — names like the large-cap utility holding companies most AI-power commentary defaults to — are being valued relative to the rest of the market, which is sensitive to regulatory lag, rate-base approval timelines, and interest rates, independent of how much new AI-driven generation capacity exists anywhere in the economy. The pipeline tells you where new, often unregulated capital is actually being deployed to build that capacity, long before — and sometimes instead of — any of it touches a regulated utility's books.
Reading backlog-heavy filings from this cohort carries its own caution, too: a large contracted-revenue figure in an S-1 is a claim about future capacity under contract, not cash the company has already earned, and that distinction matters as much for a power-infrastructure IPO as it does for the sector classification question above. Both are reminders that the label attached to a number — "Utilities," "backlog," "AI power demand" — can carry less information than it appears to, and that checking what sits underneath the label is usually the more useful exercise than reacting to the label itself.
If the AI-driven power buildout eventually gets absorbed into the regulated grid at scale — new transmission lines, utility-owned generation serving data-center load under long-term contracts — some of it may migrate into GICS Utilities over time. For now, the companies building the actual capacity are doing it from outside that sector almost entirely, which is worth knowing before treating a Utilities sector-weight chart as a scoreboard for the AI power story.