On January 13, 2026, Microsoft’s president, Brad Smith, offered a phrase that is starting to sound like the new social contract for the AI era: the tech industry must “pay its way.”
It is a neat line — and, politically, a useful one. Electricity demand in the US is rising again after years of slow growth, and the Energy Information Administration expects consumption to set new records in 2026 and 2027. The biggest, fastest-moving load is not a steel mill or a chemical plant. It is the data center — the physical infrastructure behind cloud computing and, increasingly, generative AI.
For households, this boom can feel like a bill they did not vote for. The physical footprint is local: transmission lines, substations, diesel backup generators, water pipelines, and giant buildings that employ fewer people than their size suggests. The cost footprint is often regional: once the grid expands for a handful of very large customers, the question becomes whether those costs are fully pinned to the customers that caused them — or smoothed across everyone else.
Microsoft says it wants to end the argument before it hardens into a wider revolt. The company pledged to pay utility rates “high enough” to cover its power costs, to work with utilities to expand supply, and to publish water-use information by data-center region while replenishing more water than it consumes.
The problem is that “paying its way” is not one thing. It is a spreadsheet of contested definitions — and, increasingly, a fight over who gets to see the spreadsheet.
The demand shock is real. The accounting is the battleground.
Start with the scale. In a recent forecast from 451 Research (S&P Global), utility-supplied power to hyperscale, leased and crypto-mining data centers rises to 61.8 gigawatts in 2025 (up about 11.3 GW year-on-year), then 75.8 GW in 2026, and reaches 134.4 GW by 2030.
Grid operators are building their own versions of this story. PJM, the largest US grid operator, has said its 2025 long-term load forecast implies 32 GW of peak-load growth from 2024 to 2030, with around 30 GW attributed to data centers.
This surge is showing up in markets. PJM’s capacity auction prices for 2026–2027 hit the regulator-approved cap, a sharp jump from just two years earlier — a signal that the system is paying more to ensure enough power is available during peaks.
Now comes the awkward part: how those costs reach your bill.
In much of the US, electricity prices are set through public “rate cases,” but large customers often negotiate special deals. Ari Peskoe, who studies electricity law and regulation, notes that many data centers do not pay standard tariff rates; they sign special contracts that are often confidential.
Confidentiality matters because it obscures a simple question: are data centers paying the full incremental costs they trigger — new substations, upgraded feeders, transmission reinforcements, system planning, reliability services — plus a fair share of the sunk costs of the grid they benefit from?
Where the costs are obvious, the politics becomes combustible. Bloomberg has documented places where wholesale power prices near significant data-center activity were dramatically higher than five years earlier, and argued that these pressures can ripple across wider regional grids.
That is the background to Microsoft’s attempt to get ahead of the backlash. It is also why regulators, state lawmakers, and grid operators are racing to write a new set of rules for an old industry.
“Pay your way” can mean four different things
When a tech executive says a data center will pay its way, they usually mean some combination of:
1) Paying for the wires to the front gate
This is the easiest piece to explain and, often, the easiest to enforce: interconnection upgrades and local distribution equipment needed to physically serve the load.
2) Paying for the system upgrades the load forces onto everyone else
This is where arguments begin. Utilities may need upstream transmission reinforcements and additional generation to serve large, fast-growing load. If those upgrades are treated as “system benefits,” some costs can be socialised — even if the immediate driver is a cluster of new data centers.
3) Paying even if the demand turns out to be a mirage
Forecasts for data centers are notoriously uncertain. Some projects never materialise; some get scaled back; some move jurisdictions. If utilities build for a load that evaporates, someone is left holding the bill. This is why regulators are increasingly forcing “take-or-pay” style commitments — minimum payments tied to contracted demand, regardless of actual usage.
4) Paying for the externalities communities notice
Water, land use, noise, and local tax arrangements are not line items on a utility tariff, but they are often the triggers for public opposition. A typical data center can use around 300,000 gallons of water a day, and very large sites can reach millions of gallons per day, depending on cooling design and climate.
Microsoft’s new initiative touches (2), (3), and (4). But whether it changes bills depends on how regulators define the “way” in “pay your way” — and whether they can verify compliance.
States are starting to write the fine print
The most interesting regulatory action is happening at the state level, because that is where utility rate design lives.
Virginia’s “minimum bill” approach
Virginia, home to the world’s densest cluster of data centers, has moved toward a clearer rulebook. In November 2025, the Virginia State Corporation Commission approved a new large-user rate class (GS-5) for customers demanding 25 MW or more, effective January 1, 2027.
Crucially, the SCC requires certain large customers to pay minimum shares of contracted demand — 85% of contracted distribution and transmission demand, and 60% of generation demand — as a way to insulate other customers from the cost of rapid infrastructure build-out.
This is regulation by blunt instrument — and it is deliberately blunt. It treats the right to reserve large quantities of power like a product you must pay for, even if you do not fully consume it.
Georgia’s contract-and-commitment model
Georgia is taking a more contract-led route. Georgia Power has highlighted nearly 2 GW of new customer contracts filed under rules designed to serve large-load customers while “safeguarding” residential and small business customers. The new framework emphasises upfront infrastructure payments, long-term commitments, and financial guarantees — a way to reduce the risk that other customers subsidise speculative projects.
New Jersey’s tariff push
New Jersey lawmakers have advanced efforts to create special tariff treatment for “large load” data centers, explicitly framed as consumer protection. One legislative summary described tariffs meant to compensate for the costs these customers impose on other utility users, applying to large-load users above a stated threshold.
The political message is consistent across states: if you are arriving with a power demand comparable to a small city, you are going to be treated less like a normal customer and more like an infrastructure project.
Grid operators, not just communities, are pushing back
Communities have been the face of data-center opposition — moratorium proposals, contentious zoning hearings, “not here” campaigns. More than 230 environmental groups have even called for a national pause on new data centers until stronger rules are in place.
But the harder pushback is coming from grid operators whose mandate is reliability.
PJM’s own materials have framed large-load additions as a critical issue, because the system is tightening: in one filing, PJM noted that it committed nearly all offered supply in a recent capacity auction and flagged that its updated load forecast implies even tighter conditions, driven mainly by large loads.
At the federal level, the Federal Energy Regulatory Commission is trying to keep up with a new phenomenon: data centers seeking to co-locate with generation and take service in ways that blur the boundary between “behind-the-meter” and grid-connected load. In December 2025, FERC directed PJM to create clearer rules for co-location arrangements, arguing PJM’s existing tariff lacked clarity and consistency and needed transmission service options that reflect loads able to limit withdrawals from the grid.
This matters because it changes the bargaining power. If a data center can bring its own power (or be physically adjacent to it), it can credibly threaten to bypass the grid — which weakens the utility’s ability to spread costs and potentially shifts risk back to the project developer. It also raises new questions: who pays for reliability, planning, and backup if a large load is “mostly” islanded but still wants the grid as an insurance policy?
Microsoft is responding to a legitimacy crisis, not just a rate case
It is tempting to read Microsoft’s pledge as corporate virtue — a large company doing the responsible thing. The more useful interpretation is strategic: the industry has discovered that social licence is now a constraint as real as power supply.
In Wisconsin, Microsoft pulled plans for a new data center after local opposition, and the company has said it supports a rate structure meant to prevent data center power costs from being passed on to consumers. That is not charity. It is a recognition that the “cheap power + tax incentives” playbook is becoming politically toxic.
Microsoft’s policy package also reads like a pre-emptive concession to lawmakers. President Donald Trump, for his part, has publicly echoed the “pay your own way” framing — signalling that the politics of AI infrastructure are now national, not just local.
Still, the pledge leaves open the key question: pay for what, exactly?
If paying “rates high enough” means simply paying a higher per-kilowatt-hour price while grid expansion is still booked into the broader rate base, households may see little relief. If it means covering the full incremental cost of transmission and distribution upgrades, plus capacity and reliability costs that follow from rapid load growth, then it could become a genuine turning point.
The difference is not rhetorical. It is billions of dollars, allocated one docket at a time.
What a credible “pay its way” framework would require
If states want the AI economy without an electricity revolt, they will need something more durable than corporate promises. Four changes stand out.
Make large-load contracts visible by default
Confidentiality is the quiet enabler of cost shifting. If the public cannot see the deal, they cannot know who is paying. Researchers have warned that opaque rate structures can allow cost shifts, particularly through demand charges and special contracts.
Require long-term commitments tied to build timelines
Virginia’s minimum-demand rule is a form of “reservation fee.” It is not elegant, but it is understandable. Similar “minimum bill” approaches have spread because they reduce the risk that utilities build for load that never arrives.
Treat flexibility as an asset, not a loophole
Data centers can sometimes curtail load or shift compute. The grid should reward that — but only if it is contractually enforceable and visible to planners, not used as a backdoor to avoid paying for system peaks. Academic work has noted that cutting peak demand can reduce costs for everyone if utilities plan for a lower peak and can rely on the customer’s ability to turn off or self-power.
Publish water use and cooling choices like emissions data
Microsoft’s commitment to publish regional water-use data moves in this direction. Given the scale — hundreds of thousands of gallons a day for a typical site, and far more for the largest — water is becoming as politically salient as electricity.
The deeper issue: who benefits from the AI build-out?
Underneath tariff design is a harder moral question: should households subsidise the infrastructure of a digital economy that they also consume?
The truthful answer is messy. Consumers do benefit from cloud services, better digital tools, and new products built on AI. They also bear the risk when demand forecasts run hot, when grid upgrades are misallocated, or when local water systems strain.
That is why “paying its way” is likely to become less a voluntary promise and more a regulated requirement — a new class of electricity customer with its own rules, minimum commitments, and public scrutiny.
The US is heading into an era where electricity demand is rising again and the grid must expand. The EIA expects consumption to keep climbing, even as the generation mix shifts toward more renewables over time. The question is not whether the build-out happens. It is whether the public believes it is fair.
Microsoft’s pledge is best read as an early sign that Big Tech sees what utilities and regulators already know: in the AI era, the limiting factor is not just chips and talent. It is legitimacy — measured, in the end, on the monthly electricity bill.
