Most coverage of data centers and electricity asks whether the buildings will raise your bill. It is the right question and I have written about it before. But it skips something that has been happening quietly in utility commission dockets for the last two years: in a growing number of states, regulators have approved arrangements where a large data center load pays enough into the system that everyone else's bill goes down.

Not down in theory. Down in a signed contract, approved by a state commission, with a number attached.

That does not happen automatically and it is not a property of data centers. It is a property of how the deal is written. The same load that lowers bills under one structure raises them under another, and the difference is a handful of unglamorous contract terms that almost never make the news.

Indiana Is the Cleanest Example

Start with the one where a regulator has already signed off.

NiSource, which operates NIPSCO in northern Indiana, reached agreements with Amazon and Alphabet and took them to the Indiana Utility Regulatory Commission. On June 17, 2026, the IURC approved the settlement, the Amazon special contract, and the associated power purchase agreement. The Amazon contract runs 15 years, service begins January 1, 2027, and the load ramps to 2,400 megawatts by the end of 2032.

NiSource's stated expectation is approximately $1.4 billion in customer savings over 15 years, up to $124 a year for a residential customer, delivered as bill credits as the load ramps. Amazon alone accounts for roughly $1 billion of that.

Two things about that number are worth saying plainly. It is the company's own projection, not a regulator's audit. And the credits arrive as the load arrives, which means most of the money is in the future, not on a bill anyone has opened yet.

Georgia: Real Numbers, Honest About the Calendar

Georgia is the second cleanest case, and it is a good test of whether you are being told the truth about timing.

On July 1, 2025, the Georgia Public Service Commission unanimously approved a base rate freeze for Georgia Power through at least 2028. That freeze covers base rates only. Fuel costs and storm recovery can still move a bill, so a customer can see a freeze announced and a bill go up, and both facts can be true.

The larger figure being quoted, $1.7 billion of customer benefit, is a projection covering 2029 through 2031 from serving new large loads. Georgia Power projects about $102 a year in additional residential savings beginning in 2029, on top of an approved reduction of roughly $50 a year, funded by at least $556 million a year of incremental large-load revenue.

That is a real mechanism with real approval behind it. It is also four years out, and anyone telling you Georgians are saving money on data centers today is describing 2029.

Wisconsin: The Part That Matters Is the Ordering

Wisconsin is where I would point a city council, because the commission did something more useful than approve one contract.

On May 7, 2026, the Public Service Commission of Wisconsin approved Alliant Energy's special contract with Meta for the Beaver Dam data center, 220 megawatts over ten years, with customer-protection modifications attached. Meta's investment is roughly $1 billion, it underwrote about $200 million in network improvements, and it put $15 million into the Hometown Care Energy Fund, which helps customers who cannot pay their bills.

The more consequential part of that order gets less attention: the commission also directed Alliant to create a standard rate for future data center loads above 100 megawatts. That moves the state from negotiating each deal privately to publishing terms in advance, which is the difference between a favor and a rule.

PSC Chair Summer Strand criticized the black-box, nondisclosure-driven approach these contracts usually take. That criticism is worth repeating even by people who support the projects, because secrecy is what makes a good deal look like a bad one.

One correction, since it circulates constantly: the widely repeated five-year rate freeze tied to these projects is in Iowa, at Alliant's Iowa utility, not Wisconsin. If you see it attributed to Wisconsin, whoever wrote it did not check.

Where It Is Contested

Two more states get cited in these lists and both deserve qualifiers rather than applause.

In Michigan, the Public Service Commission conditionally approved DTE's contract with an Oracle-linked data center in December 2025, 1,383 megawatts in Saline Township, with DTE calculating a net benefit to other customers of approximately $300 million. The contract terms are the interesting part: a 19-year minimum, 80 percent minimum billing demand, and a termination payment of up to ten years of minimum billing. A separate Google contract was still contested as of mid-2026, and Attorney General Dana Nessel filed testimony against it in June 2026. Anyone presenting Michigan as settled is presenting an unapproved contract as done.

In Mississippi, Entergy projects more than $2 billion in savings over twenty years from its Amazon agreements. There is no public docket to check it against, because a 2024 state law made the agreement confidential. Meanwhile a Synapse Energy Economics analysis prepared for Earthjustice estimates the projects have already raised residential rates by about $10.60 a month. Entergy's counter is that bills will be lower by 2030 than they would have been otherwise, which is a different claim than bills going down.

That is the honest state of the evidence. Some of these deals are approved and audited. Some are projections from the company that benefits from them. Some are actively being fought by state attorneys general and consumer advocates. Treating all of it as one triumphant pile is exactly the credibility problem this industry keeps creating for itself.

The Four Terms That Decide It

Strip away the state names and the same handful of contract terms show up everywhere the deal protects residential customers.

Minimum billing. The customer pays for a floor of demand whether they use it or not. In Michigan the Oracle contract sets 80 percent minimum billing demand. It stops a utility from building generation for a load that shrinks or never shows up, which is the single fastest way for costs to land on everyone else.

Upfront infrastructure payment. The substation and the lines get paid for by the customer that needs them, before service starts, rather than rolled into general rates and recovered from everyone.

Long commitments with exit costs. Nineteen years with a termination payment behind it is a different risk profile than a five-year deal on a thirty-year asset. Stranded cost is the risk here, and the contract length is how it gets managed.

A published tariff rather than a private negotiation. This is the one Wisconsin just moved on. A standard rate for loads above a threshold means the next project does not get a bespoke deal nobody can inspect.

Arizona Is Building the Same Machinery

Every one of those four terms already has an Arizona counterpart, which is why this is not a story about somewhere else.

SRP's E-67 price plan requires accounts of 20 megawatts or more to pay the greater of actual demand or 80 percent of forecasted load. That is the same mechanism as the Michigan minimum billing term, arrived at independently. SRP's Large Customer Integration Process, introduced in 2025, requires new large-load customers to pay upfront for the infrastructure needed to serve them so those costs are not shifted to residential customers. As of 2025, data centers accounted for about 5.1 percent of SRP's system peak.

APS has had a high load factor rate class since 2017. In its pending rate case, APS proposes a rate increase of more than 45 percent for extra high load factor customers, along with formula rates with annual true-ups, minimum bills, collateral requirements, and long-term contracts with upfront infrastructure contributions. Worth stating alongside it: the same case requests roughly a 14 percent net system increase and drew hours of public protest when the hearing opened in May 2026, and Arizona's residential utility consumer advocate has argued data centers belong in a separate customer class entirely.

In southern Arizona, TEP currently serves no data centers at all, with one under contract and a speculative queue far larger. That is the useful case: a framework being built before the load arrives rather than after.

The piece that ties it together is Arizona Corporation Commission docket E-00000A-25-0069. At the Commission's large-load workshop in April 2026, the consensus across utilities, consumer advocates, and industry was that the cost causer should pay, and the Commission is actively studying a universal large-load tariff. Chairman Nick Myers put the thesis of this entire article in one sentence: "With properly structured tariffs and ESAs that prevent cost-shifts, data centers and other large load customers can help drive down rates and serve as powerful engines of local economic growth."

Note the first three words. Not data centers lower rates. Properly structured tariffs.

What to Actually Watch

If a project is proposed near you and someone tells you it will lower your bill, the claim is testable. Ask whether there is a minimum billing provision and what the percentage is. Ask who pays for the substation and when. Ask how long the commitment runs against the life of whatever is being built to serve it. Ask whether the terms are published or sealed.

If the answers are good, the claim is probably fair.

The docket is public. Anyone can read it, and anyone can file a comment.

Sources