A Dallas Fed working paper put a number on the thing your neighbors are arguing about at the county building. Data centers that are already built have raised wholesale electricity prices by 3 to 5 percent on average across the continental United States.
Not the ones on next month's agenda. The ones already humming.
That figure is a modeled result, not a meter reading, and the gap between those two things is most of this post. Four people usually end up in the same room over this: a household holding a statement, a commissioner who has to vote, a utility planner holding a 500-megawatt interconnection request, and a developer who needs speed to power. You are arguing about the same megawatts in four different dialects.
So here is the honest answer — and then the part almost nobody covers: which half of the fight a funded living system can actually touch, and which half it cannot.
the short answer, in three findings
Wholesale prices: yes, measurably. The Dallas Fed paper (Kay, Reaser and Taylor, March 20, 2026) runs an hourly, unit-level least-cost dispatch model across the continental U.S. Existing data centers account for a 3 to 5 percent average increase in wholesale prices, with substantially larger effects in the corridors that host them. Extended through 2028, a moderate build-out yields about 20 percent; a high-utilization build-out yields roughly 50 percent. Those last two are scenarios, not forecasts — the spread between them is an argument about how hard the chips actually run.
Capacity prices: the market monitor is blunt. In PJM — the wholesale market covering all or part of 13 states and D.C. — the Independent Market Monitor writes that "data center load growth is the primary reason for recent and expected capacity market conditions, including total forecast load growth, the tight supply and demand balance, and high prices." That is the Monitor's own language, in its 2026 State of the Market reporting. In a separate counterfactual analysis of the 2026/2027 Base Residual Auction, the Monitor puts the effect of including existing and forecast data center load at $7.27 billion, or an 82.1 percent increase in that auction's revenues, and $23.1 billion across the 2025/2026 through 2027/2028 auctions combined, per the same IMM report. Capacity payments are what generators are paid to promise availability. They land in bills.
Your residential statement: less clean than either side wants. A New Jersey State Policy Lab study (Kiran Garimella, June 2026) built a ZIP-level panel of 22,834 units across 24 states from 2014 to 2024, linked to 2,277 operating data centers. First pass: bills about $2.09 a month higher near a data center, against an average bill of $164 — borderline. Matched against similar places, it fell to $0.90 and lost significance. At the utility service-territory level, which is where residential rates are actually set, the estimate was $0.35 a month unweighted, negative when household-weighted, and statistically indistinguishable from zero.
why the third finding is not a defense
If you are a developer, do not walk into a hearing waving that last paragraph. Read the author's own caveat: the data ends in 2024, before the hyperscale wave, so the sample is mostly traditional facilities. The campuses now in queue run to hundreds of megawatts, sometimes near a gigawatt, and they require new substations, transmission, and generation contracts that take years to build and years more to enter rates.
The three findings do not contradict each other. They describe a cost that is real in the wholesale and capacity markets today and has not yet finished traveling to the household. Rate cases are slow. That is the whole finding.
Data centers raise wholesale power prices measurably and residential bills unevenly. The difference between those two sentences is rate design — and rate design is the fight.
who pays for the wires
The ratemaking principle here is old and has a name: cost causation. Whoever causes a cost should carry it. Everything useful on the bill question is a version of that sentence, written into a tariff.
The most visible instrument is the Ratepayer Protection Pledge, announced March 4, 2026, and signed initially by seven hyperscalers — Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI. Signatories commit to build, bring, or buy the new generation their load requires and pay its full cost; to pay for all new power delivery upgrades their facilities require, including network upgrade costs; and to negotiate separate rate structures with their utility and state, paying for that power and infrastructure whether or not they use it. A July 23, 2026 expansion added 187 utilities, cooperatives, and developers plus 23 governors.
It is also voluntary and non-regulatory. Brookings, broadly supportive of the framework, says exactly this: the pledge needs enforcement, and the starting point is utility commissions, state legislatures, and governors. As of February 20, 2026, more than 300 data center bills had been filed in 30-plus states; lawmakers in at least 18 states introduced 30-plus bills creating a separate rate class for large energy users. In Congress, the House Energy and Commerce Committee advanced the Ratepayer Protection Act (H.R. 9340) 52-0 on July 21, 2026, directing state commissions to consider requiring loads of 100 MW and larger to cover the full cost of the grid upgrades that serve them.
Ohio shows what the enforceable version looks like. AEP Ohio's data center tariff, approved in July 2025, sets a 25 MW threshold for a separate rate class and requires those customers to pay for 85 percent of contracted capacity regardless of actual consumption — a take-or-pay floor. In August 2026 the Public Utilities Commission of Ohio added a 180-day notice requirement before grid connection, with the data center customer expected to cover the costs of the generation procured to serve it. Data Center Dynamics reports the utility's data center pipeline fell by more than half after the tariff took effect.
That is the honest cost of a good tariff, and developers should say it out loud rather than let someone else say it for them. Structure filters speculative load, and filtering speculative load is most of the point.
None of this is a knock on the utility. A utility is legally obliged to serve, holds a queue of requests it did not create, and can only change rates through a formal case. It has the least discretion of anyone in the room.
For contrast: five data center firms put $18 million into AEP Ohio's Neighbor to Neighbor bill-assistance program. Real money reaching real households — and not a tariff. It does not decide who pays for the next substation.
the table nobody puts in the slide deck
Two different toolkits get pitched at the same hearing. They do not do the same work, and pretending otherwise is how a project loses trust it will need later.
| question at the hearing | tariff / cost-causation tool | living-system tool | what neither one does |
|---|---|---|---|
| who pays for the substation and the line? | yes — developer-funded network upgrades, interconnection deposits | no | — |
| will my rate class subsidize a 500 MW customer? | yes — separate large-load rate class, take-or-pay floor | no | — |
| what if the project is cancelled mid-build? | yes — collateral, exit provisions, cost-allocation rules | no | — |
| does the wholesale price come down? | slowly, and only if new supply actually gets built | no | clear a capacity auction, or un-tighten regional supply and demand in one permit cycle |
| who replaces the water the cooling towers evaporate? | no | yes — same-basin recharge, wetland and riparian funding | make a rate case move faster |
| who answers the heat the campus throws at the next street? | no | yes — living cover, canopy, permeable ground | replace a chiller |
| who keeps the acres under the new generation alive? | no | yes — pollinator ground, grazing, protected working land | stop the land-price bid next door |
| does the 3 a.m. hum stop? | no | partly — a living buffer takes decibels off | remove the need for setbacks and generator rules |
| does the county trust you? | no | not by itself | substitute for terms that are recorded and enforceable |
Read the columns as different venues, not competing vendors. The left one is decided by a public utility commission; the middle one is decided on the ground, by whoever holds the land and the money.
the hard no
Nothing in the ensurance stack lowers a wholesale power price. Not a coin, not a certificate, not a funded watershed. Wholesale and capacity prices are set by dispatch, supply, and auction rules. Say that plainly before you say anything else at a rate hearing, because a commissioner who catches you overselling on the bill question will discount everything you say on the water question.
Ensurance is not a substitute for the Ratepayer Protection Pledge, a large-load tariff, or a cost-allocation rule. It answers a different objection, in a different venue, on a different timeline.
the half a funded place does answer
Here is the connection most bill coverage misses: the water answer and the energy answer trade directly against each other.
Evaporative cooling is energy-efficient and drinks water. Closed-loop and air-cooled systems save the water and burn more power. A campus that resolves the water objection by going air-cooled has just made the bill objection worse — and it now draws hardest on the hottest afternoons, which are the same hours that stress the grid and set peak prices. The water hearing and the bill hearing were always the same trade-off.
Same-basin water. If the campus draws from a stressed basin, the durable answer is funding the systems that produce water there — recharge, wetlands, riparian ground — rather than replenishment credits bought somewhere else. That argument has its own post: data centers drink water — ensurance refills the glass.
Living cover at the site. Ambient air temperature at the intake is not a rounding error for an air-cooled plant. Shade and evapotranspiration measurably lower local temperature, which shaves cooling load at the margin on exactly the days that matter. Treat it as site engineering with an ecological dividend, not as a rate strategy — living cover is not a chiller. The mechanism is covered in the infrastructure that cools itself.
The ground under the new generation. "Build, bring, or buy" means acres. Solar arrays, gas plants, substations, battery yards — all of it lands on ground that someone farms, hunts, or looks at. What lives under a solar array is a choice: scraped gravel, or deep-rooted native cover that holds soil, holds water, and feeds the pollinators the surrounding farms depend on. Neither version changes the clearing price. One of them gives a county something to point at that is not a tax figure. See what is pollination worth.
what ensurance actually is, plainly
The watershed, the working ground, and the living cover exist whether or not anyone buys anything. ensurance is how a campus funds them in the same basin it draws from. It is not what they are.
Ordinary insurance pays after the loss. Ensurance funds the living system before it. The instrument is a certificate — a holding tied to a specific natural asset, a named watershed or wetland or stretch of working ground, whose proceeds route to the people stewarding that asset, with the flows visible to anyone who wants to check them. You can see the live instruments at specific ensurance.
Our stage, honestly: live solutions page, live certificates, small volumes. This is not a syndicate that will refinance a gigawatt campus. It is a way to put a named, funded, checkable local commitment on the table before the vote instead of after the denial.
what to do with this
If you are a household or a neighbor: the strongest question at a rate case is not "are data centers raising my bill." It is "what rate class is this customer in, and who pays for the interconnection." Ask for the tariff, not the press release.
If you are a commissioner or econ-dev staff: separate the two questions on the agenda. The bill question belongs to your public utility commission and your legislature. The water, heat, and land questions belong to your conditional use permit — and they are the ones you can actually condition.
If you are a developer or a tenant: arrive with both. A large-load tariff and a funded local system are not competing line items; they answer different objections, and a hearing where only one is present will find the missing one.
frequently asked questions
do data centers raise electricity prices?
Yes at the wholesale level, and the magnitude depends on how much gets built. The March 2026 Dallas Fed dispatch model finds existing data centers have already raised wholesale prices 3 to 5 percent nationwide, with roughly 20 percent under moderate build-out and roughly 50 percent under high-utilization build-out through 2028. Whether that reaches a residential statement depends on rate design in your state, not on physics.
who should pay for data center grid upgrades?
The load that caused them. That is the ratemaking principle of cost causation, implemented through separate large-load rate classes, take-or-pay minimums, interconnection deposits, and developer-funded network upgrades. The Ratepayer Protection Pledge commits signatories to this voluntarily; state commissions and legislatures are what make it binding.
can nature-based projects lower data center power bills?
Not the wholesale price and not the capacity auction — those are set by dispatch and auction rules. Funded living systems answer the other half of the hearing: the water the cooling towers evaporate, the heat the campus throws, and the ground under the new generation. Site-level living cover can shave cooling load at the margin, which is engineering, not a rate strategy.