EdgeCore says it will bear the full power infrastructure cost of its data centre campuses
Who pays for grid upgrades is becoming a design constraint for hyperscale campuses, and defining 'full cost' is harder than it sounds.
EdgeCore Digital Infrastructure says it will fund the generation, transmission and delivery infrastructure required by its data centre campuses, Data Center Knowledge reported on 20 August. Julie Brewer, the developer-operator's executive vice president of finance, told the publication that the company expects to bear 100 percent of those costs. She described signing the White House's Ratepayer Protection Pledge, introduced in March 2026 and expanded in July, as a continuation of how the company has historically operated.
The scale involved has changed quickly. According to the report, power infrastructure typically represents 5 to 10 percent of a project's upfront cost, which at campus scale can run to tens or even hundreds of millions of dollars. Five years ago a large data centre needed around 25 MW; EdgeCore now concentrates on campuses above 300 MW. Its pipeline includes a campus of more than 1.1 GW in Louisa County, Virginia, with planned investment above $17 billion, a 496 MW campus in Mesa, Arizona, and a 216 MW campus in Reno, Nevada.
The article also explains why "full cost" is difficult to define. Neil Osnato of Persistence Analytics Group noted that infrastructure serving a single customer can be assigned to it, but upstream investments that change power flows, congestion and reserve requirements are much harder to attribute prospectively.
Arizona utility Salt River Project offers one model. Its Large Customer Integration Process, introduced in 2025, requires large customers to pay upfront for identified upgrades, and its updated E-67 rate plan requires customers forecasting 20 MW or more who take service after November 2025 to meet minimum billing based on actual use or 80 percent of forecast demand, plus pay for dedicated transmission and substation service. At Mesa, EdgeCore paid for a dedicated Hartman Substation and a five-pole 69 kV line extension for its first 26 MW building. Brewer said EdgeCore would walk away from a project if the power economics became unworkable.
Why it matters
Electricity has become the gating resource for data centre development, and the argument over who pays for grid expansion now shapes where and whether campuses get built. Utilities are responding with tariffs that shift risk onto large customers, while developers are volunteering to absorb costs to keep approvals moving. The unresolved question of how to attribute system-wide costs will define the next round of regulatory fights. Understanding these mechanics helps anyone working near infrastructure read the true economics of an AI build-out.
Electricity has become the gating resource for data centre development, and the argument over who pays for grid expansion now shapes where and whether campuses get built.
- Louisa County, Virginia1.1+ GW
- Mesa, Arizona496 MW
- Reno, Nevada216 MW
Figures: Figures reported by Data Center Knowledge, citing EdgeCore
What you can learn from this
Cost allocation separates customer-specific from system costs. A substation and a dedicated line that serve one campus are easy to attribute, so tariffs commonly make the customer pay for them directly. Upstream assets such as new generation or a reinforced transmission corridor serve many customers at once and their costs are usually spread across all ratepayers. The debate in this story exists because a very large load changes system conditions in ways that are real but hard to prove in advance.
Minimum-bill clauses exist to manage stranded-asset risk. When a utility builds capacity for a customer that never arrives or arrives smaller than forecast, the investment becomes a stranded asset that other ratepayers must fund. A minimum bill tied to a percentage of forecast demand, like the 80 percent figure in SRP's tariff, guarantees a revenue stream regardless of actual consumption. It behaves like a take-or-pay contract and forces developers to forecast honestly, because an inflated forecast now carries a direct financial penalty rather than being a free option on future capacity.
The grid is layered, and each layer has an owner. Generation produces power, high-voltage transmission moves it over long distances, substations step voltage down, and distribution lines deliver it to a site. A 69 kV line, as in the Mesa example, sits at the sub-transmission level, above local distribution but below the bulk system. Knowing which layer an upgrade touches tells you which regulator, which utility and which tariff govern it. Transmission is typically overseen at the federal level in the United States while retail rates are set by states, which is one reason cost-allocation disputes rarely have a single decision-maker.
Campus size follows accelerator density. A modern AI rack can draw tens of kilowatts, many times more than a traditional enterprise rack, so the same floor area needs far more power. That is why a 25 MW facility was large five years ago and 300 MW campuses are now the target. Cooling scales with power as well, since almost every watt delivered to a server eventually becomes heat that must be removed.
Forecast risk cuts both ways. Developers announce gigawatt-scale campuses years before servers arrive, and utilities must decide how much of that demand is real before committing capital. If loads fail to materialise, ratepayers may carry the cost; if utilities under-build, projects stall. Tariffs with upfront deposits and minimum bills are the instrument both sides use to share that uncertainty.
We teach this
How to use this in practice
Read one large-load tariff end to end. Download a published rate schedule for large customers from any utility website, such as the SRP E-67 plan mentioned in the story, and summarise in one page the minimum-bill terms, deposit requirements and who owns the substation. Note any threshold, such as a megawatt figure, that triggers the large-customer terms. Done looks like a summary you could hand to a colleague that answers the question of who pays if the load never shows up, and at what size a customer starts to be treated differently.
Draw the path from generator to rack. Sketch generation, transmission, substation, distribution and the on-site electrical room, then mark each segment with whether the customer or the utility typically pays for it under the tariff you read. Done looks like a labelled diagram with at least one segment marked as contested, matching the attribution problem described in the article.
Build a two-tab energy cost model. In a spreadsheet, enter a campus capacity in MW, an assumed utilisation such as 70 percent, and a price per megawatt-hour, then calculate annual energy cost. Add a second tab that applies a minimum bill at 80 percent of forecast demand and shows the cost when actual load comes in at 50 percent. Done looks like a chart showing the penalty for over-forecasting, which explains why EdgeCore says it would abandon an unworkable project.
Estimate the heat load for a single AI rack. Take a rack power figure of 40 kW, convert it to BTU per hour by multiplying by 3,412, and compare it with the output of a typical residential air conditioner. Then multiply by the number of racks in a 26 MW building to see the total heat that must leave the facility every hour. Done looks like a short note showing why liquid cooling appears in every large campus plan and why power and cooling are budgeted together rather than as separate line items.
Sources
- EdgeCore Says Data Centers Should Pay Their Own Power Costs — Data Center Knowledge
Our reporting is an original summary; full coverage is at the links above.
Don't just read about it — build it.
Square 1 teaches the skills behind the headlines, with every line of your work graded by AI. Find your starting point in 3 minutes.
Get your free skill report