Ohio’s utility regulator just threw down a gauntlet that could reshape the economics of AI cloud infrastructure across the country. On July 9, 2025, the Public Utilities Commission of Ohio (PUCO) approved a first-of-its-kind tariff forcing large new data-center customers to pay for at least 85% of their contracted energy costs for up to 12 years. The ruling directly targets the hyperscale data centers that power generative AI workloads, and it signals a seismic shift in who pays for the grid upgrades needed to sustain the AI boom.

The decision came after months of debate involving AEP Ohio, tech giants like Google and Microsoft, and consumer advocates. It’s a clear answer to a question that’s been simmering from Virginia’s “Data Center Alley” to the PJM regional grid: when a handful of corporate behemoths cause a surge in electricity demand that requires billions in transmission and generation investments, should those costs be socialized across all ratepayers, or should the companies themselves shoulder the burden?

The New Math of AI and Electricity

Data centers have always been power-hungry, but the advent of large-scale AI training and inference workloads has changed the equation entirely. The U.S. Department of Energy and Lawrence Berkeley National Laboratory estimate that data centers consumed about 4.4% of U.S. electricity in 2023, with projections showing that figure could jump to between 6.7% and 12% within a few years under current growth trajectories. That’s not just a blip—it’s a structural shift driven by clusters of facilities that can each demand hundreds of megawatts, concentrated in regions like Northern Virginia, central Ohio, and parts of Texas.

These instantaneous loads stress local transmission systems in ways that legacy utility planning never anticipated. Utilities typically map out demand growth in modest increments over years; now they face developers requesting gigawatt-scale connections in months. The result: a scramble to build new substations, transmission lines, and peaker plants, often at breathtaking cost. And until now, the default regulatory model spread those costs across all customers, from the data center to the suburban household.

Inside Ohio’s Landmark Tariff

The PUCO’s order implements a data-center-specific tariff for AEP Ohio that upends the traditional model. Key provisions include:
- An obligation for very large new data-center customers to pay for at least 85% of their subscribed monthly capacity.
- A commitment term of up to 12 years, with a four-year ramp-up period.
- Exit fees for projects that cancel before the end of the term.
- Financial assurance requirements to protect the utility and existing ratepayers from stranded investments.

AEP framed the tariff as essential to prevent cost-shifting. If a data center demands a dedicated substation or a new transmission line, the logic goes, that data center—not the grandmother on a fixed income—should underwrite the asset. The ruling explicitly ties infrastructure risk to the party that causes it, a principle known as cost causation.

“This decision makes clear that large new loads must pay for the infrastructure they require,” said a PUCO spokesperson. “We are not willing to gamble with ratepayer money on speculative development.”

Why Ohio Matters Nationwide

Ohio’s move is one of the first comprehensive, regulator-approved frameworks for allocating data-center infrastructure costs. It sets a precedent that utilities and commissions from Virginia to Texas are already studying. If similar tariffs spread, the capital economics of data-center deals change overnight. Speculative projects—those built in hopes of finding tenants later—become far riskier. Real estate plays that counted on socialized grid upgrade costs suddenly face vastly higher upfront capital requirements.

The tariff also alters the siting calculus for hyperscalers. In the past, cheap land and tax incentives often drove location decisions. Now, the regulatory posture of the local utility and state commission becomes a first-order concern. Developers will likely favor jurisdictions with predictable, developer-friendly rules, potentially accelerating buildouts in some areas while slowing them in others.

Pushback from Big Tech and Developers

Major cloud providers and data-center developers didn’t let the tariff pass quietly. They argued in filings that the 85% minimum take obligation is rigid and anti-competitive, raises project costs, and could discourage investment and the jobs tax revenue that data centers bring. Tech firms pushed for more flexible, market-based approaches, such as variable pricing tied to actual usage or shorter-term contracts with renewal options.

“This tariff imposes a one-size-fits-all solution that ignores the diversity of data-center business models,” one industry comment letter stated. “It will chill investment in Ohio precisely when the state is poised to become a technology hub.”

But consumer advocates and other utilities backed the commission. Without strong cost-causation protections, they argued, residential and small-business customers would inevitably bear the brunt of massive, lumpy infrastructure investments. Petitions for reconsideration are expected, signaling that the legal and policy battles are far from over.

Virginia: The Canary in the Coal Mine

If Ohio is the first mover, Virginia is the epicenter. Northern Virginia hosts the world’s largest concentration of data centers, with more than 150 facilities and over 5 GW of load. A recent report from Virginia’s Joint Legislative Audit and Review Commission (JLARC) painted a stark picture: unconstrained growth could double the state’s power demand within a decade, overwhelming generation and transmission capacity. The report projected that under an unconstrained model, average monthly energy consumption could exceed 30,000 GWh by 2040.

Meeting that demand would require adding solar facilities at twice the 2024 annual rate and wind generation exceeding the potential of all secured offshore sites. Even building enough natural gas plants—a new 1.5 GW plant every two years for 15 years—would mean scrapping the Virginia Clean Economy Act’s fossil-fuel phaseout by 2050.

The cost to non-data-center customers? JLARC estimated that a typical Dominion Energy residential customer could see monthly bill increases of $14 to $37 by 2040, independent of inflation, to fund the $18 billion in needed infrastructure. The report recommended creating a separate data-center customer class, adjusting cost allocations, and requiring data centers to participate in demand response and renewable development.

“Unconstrained data center growth in Virginia will make it very difficult to scale the power generation and transmission infrastructure needed,” the JLARC report concluded.

The warning signs have already prompted action. Dominion Energy has started proposing new rates tailored to large-load customers, and other states are watching closely.

How Tech Companies Are Fighting Back

Facing new cost pressures and regulatory scrutiny, hyperscalers are deploying a range of strategies to secure power on their own terms:

Power Purchase Agreements (PPAs) and Direct Deals

Google, Microsoft, and Amazon have become the largest corporate buyers of clean energy, but now they’re moving beyond standard PPAs. Direct contracts with nuclear plants—including advanced small modular reactors and agreements to resurrect retired units—provide the firm, 24/7 baseload power that AI workloads crave. In some cases, tech firms have spawned subsidiaries to develop and operate generation assets, blurring the lines between consumer and producer.

Demand Response: The Grid-Friendly Pivot

In August 2025, Google announced formal demand-response agreements with Indiana Michigan Power and the Tennessee Valley Authority. The deals allow Google to temporarily shift or pause non-urgent machine-learning workloads during peak grid stress. It’s a pragmatic compromise: the company gets faster interconnection approvals, and the grid gets flexible load that can act as a shock absorber.

“This is a scalable tool that benefits both the grid and our operations,” Google’s blog post explained. “We’re turning what could be a conflict into a collaboration.”

Other cloud providers are likely to follow suit, integrating AI workload scheduling with real-time grid signals. It’s a form of operational load management that reduces the need for immediate capital-intensive transmission upgrades.

The Economics of “Who Pays”

The Ohio tariff and the Virginia report both force a reckoning with cost causation. In traditional rate regulation, a utility recovers its investments from all customers in a broad class. But when a single mega-load triggers a specific substation or transmission line, that model breaks down. Spreading the cost across millions of households makes little sense; it’s equivalent to asking the neighborhood to pay for a factory’s private road.

Take-or-pay contracts like Ohio’s 85% minimum effectively turn the data center into a long-term anchor tenant, underwriting the asset that serves it. This reduces the risk of stranded costs if the developer fails to materialize or cancels. But it also raises the bar for project financing: developers must now factor in a multi-decade energy commitment that can sink a marginal project.

For residential ratepayers, the alternative is grim. If utilities are forced to build ahead of demand and then socialize the costs, everyone’s bills rise. JLARC’s $14–$37 monthly increase estimate for Virginia is not theoretical; it’s based on $18 billion in needed upgrades. And those costs compound when multiple large loads cluster in a single service territory.

Market Power and the Public Interest

The tech industry’s aggressive energy procurement raises deeper questions about market fairness. Exclusive, long-term contracts with merchant generators can edge out other buyers and compress capacity margins. Some critics warn that big-tech subsidiaries selling into wholesale markets could distort prices, leaving smaller customers to soak up system costs. These concerns have caught the attention of federal regulators and regional transmission organizations.

“We’re seeing the emergence of a two-tier energy market,” said one energy policy analyst. “Hyperscalers lock in cheap, firm power behind the meter or through bespoke deals, while everyone else competes for what’s left.”

However, defenders point out that tech capital is keeping aging nuclear plants alive and accelerating renewable investment. The challenge is crafting transparent rules that harness that investment without sacrificing competitive fairness.

What IT Planners and Enterprises Should Do Now

For IT decision-makers, the energy landscape is no longer a given—it’s a variable that can affect cloud costs, service availability, and colocation contracts. Practical steps include:

  • Re-evaluate site selection: Jurisdictions with clear, stable tariff rules and interconnection processes will be less risky. Expect developers to flock to friendly territory, potentially shifting the geography of cloud availability zones.
  • Negotiate energy clauses: In colocation and hyperscale agreements, insist on transparency around demand-response participation, pass-through of grid upgrade costs, and variability clauses tied to energy market conditions.
  • Prepare for fluctuating cloud costs: If a major cloud provider responds to grid strain by throttling non-urgent workloads or relocating compute, your AI training jobs may see unpredictable completion times or surcharges. Build redundancy and resource-aware scheduling into your architecture.
  • Engage with your cloud provider’s sustainability team: Understand their demand-response posture and contractual commitments. Some may pass on cost savings or offer credits in exchange for flexible workload placement.

Policy Guardrails for a Fair AI Energy Future

The Ohio and Virginia examples underscore the need for coordinated, transparent policy. Key recommendations include:

  • Codify cost causation: Regulators should require large new loads to pay for the dedicated infrastructure they trigger, with phased-in commitments that match actual consumption.
  • Mandate contract transparency: Long-term power purchase agreements, especially those with utility-scale generation assets, must be filed publicly so market monitors can assess systemic impacts.
  • Expand demand-response programs for AI: Standardize protocols for shifting ML workloads and compensate participants for the grid value they provide. This turns flexibility into a system resource.
  • Accelerate transmission planning: Federal and state agencies should fast-track permitting in high-growth corridors, using advanced forecasting that accounts for AI-driven demand.
  • Use staged approvals: Condition data-center project permits on demonstrated load build-out, preventing utilities from building for phantom demand.

The Road Ahead

The AI-driven data-center expansion has exposed a fundamental tension between Silicon Valley’s speed and the deliberate pace of public-interest grid planning. Ohio’s tariff is a bold experiment: it forces big tech to pay for the infrastructure their AI ambitions require. Virginia’s report is a warning that the old model—socialize the costs and let the market rip—is unsustainable.

The industry’s embrace of demand response, direct generation deals, and behind-the-meter power shows a pragmatic recognition that the grid must evolve. But without clear rules and transparent cost allocation, communities risk footing the bill for a speculative future. The path forward requires a pragmatic compromise: incentivize investment, preserve reliability, and ensure the costs and benefits of the AI era are shared fairly. The next few years will determine whether the U.S. energy system can keep pace with the machines—or get left behind.