
82 TWh of Data-Centre Demand Is Not a Scarcity Price
The electricity is real enough to matter. The premium investors want to attach to access still lacks a price tag.
Satellite images identified 403 hyperscale data centres across the United States. A satellite can find a roof; it cannot read an electricity contract. For May 2024 through April 2025, a Harvard-led study in Nature Sustainability estimates their consumption at 82 TWh in its central scenario. We would pass on paying a scarcity premium for data-centre power access on that evidence.
There is a sharp distinction between identifying where electricity goes and identifying who has the leverage to charge more for it. According to Harvard's account of the research, Virginia, Oregon, Ohio and Iowa account for more than half of the estimated load. The Four-State Lens is simple: a national electricity total hides regional concentration, but scarcity becomes investable only with evidence of available capacity and its price at individual sites. The researchers found the concentration; the contract economics remain to be established.
The quantifiable exposure belongs to the sector rather than a public-company income statement. Across those 403 centres, the study's lower-load scenario is 68 TWh and its upper-load scenario is 99 TWh, against 82 TWh centrally, or about 1.8% of US electricity consumption over the same period. In a scenario in which the higher load assumption replaces the lower one, 99 TWh minus 68 TWh puts 31 TWh of annual electricity demand at issue. There is no verified operator-specific bill or upgrade charge to multiply by that volume. For an investor, this is an enormous physical input with an unpriced financial consequence.
The best case for paying more for early grid access is a business argument, not an environmental one. A facility with a usable connection gets to put its computing equipment to work while a competing facility still waits for service. With contracted customer demand, that head start brings revenue forward while an idle site ties up capital. Its value depends on the duration of the delay and the customer purchases secured during it. Concentration magnifies the mechanism if a particular network reaches capacity, because an early connection then provides an advantage that later entrants must recreate. You would pay something for that timing option even before a utility starts turning applicants away.
The Nature researchers give this argument its heft: an 82 TWh central scenario, bounded by 68 and 99 TWh, for the sampled facilities in the study year. Harvard supplies its geographical punch, with more than half assigned to four states. On the study's own period, however, 82 TWh divided by 8,760 hours is 9.36 GW of average equivalent load. Peak requirements, contracted capacity and network headroom are different quantities. The modeled 9.36 GW average establishes energy use, while a premium for early access requires a documented constraint or connection price.
Ireland put conditions on new data-centre connections while keeping a route open. Its Commission for Regulation of Utilities published a connection policy on 12 December 2025 after data centres accounted for 22% of national electricity demand in 2024, with a projected 31% by 2034. The regulator wrote, "The objective of the new policy is to provide a pathway for data centres to connect to the electricity system." That pathway requires new projects to provide generation or storage matching requested maximum import demand, plus additional renewables covering at least 80% of annual use. A connection remains available in principle, but the operator must bring costly supporting assets to obtain it.
That mechanism deserves a price in any serious investment model. Ireland's 2034 projected demand share of 31%, minus its 22% share in 2024, is a nine-percentage-point scenario increase, with connection conditions already specified. The 80% annual renewable requirement puts an identifiable obligation ahead of any revenue from a new connection. Yet the regulator offered conditional connections rather than an outright ban on new load. Ireland's network, demand profile and regulatory terms differ from America's, so its prescription cannot be copied across the Atlantic. Its lesson is economic: the developer can bear an obligation to supply capacity before anyone denies a connection.
That is precisely why we distrust the easy valuation story. A power connection commands a premium when you know which costs it avoids, which competitors it excludes and for how long. No named operator has a verified price for those advantages. The researchers could not separate AI-specialized electricity use from general-purpose activity at each facility, and their 403-site inventory omits some US hyperscale centres. Guidi, the study lead, told Harvard that current consumption could already be substantially larger because of expansion. That is a view about 2026, rather than a new reading of meters. It changes the argument about future supply, without changing a known operator's earnings.
The emissions accounting in the study reinforces the importance of place without resolving the investment question. The central scenario attributes 25.7 million metric tons of CO2 to the sampled centres, within a scenario range of 21 million to 31 million. Harvard says roughly 54% of their attributed electricity generation was fossil-fuelled, including about 38% from natural gas and 15% from coal, with rounded shares. That helps explain why siting, electricity mix and grid conditions attract scrutiny. A carbon accounting figure and an electricity connection price occupy different lines of an investment model. You can map the power mix without turning it into an operator's electricity bill.
Imagine two owners underwriting identical electricity volumes. One has a firm agreement spelling out service, timing and costs; the other has only a place on a map of regional demand. Their theoretical need for power is the same, but their assets should command different prices. The Four-State Lens keeps that distinction honest. A state with many data centres is a sensible place to investigate scarcity; it is a poor substitute for a public connection agreement, a utility filing with an upgrade obligation or a grid operator's capacity study. You are buying the commercial terms of access, not a colourful map of demand.
We would pass on paying extra for a presumed grid bottleneck while leaving room to buy an asset whose actual power terms justify its price. A public facility-level reconstruction released by the end of November 2027 that puts the four states at half or less of the same 403 sites' electricity use during the study year would falsify our concentration claim. Until an investable shortage is documented in site-level costs and capacity, 82 TWh is a reason to investigate those contracts, not to pay for scarcity in advance.