In 40 of 41 local ballot contests across 10 U.S. states on Nov. 3, voters favored limiting data center development.
The measures included bans, moratoriums, rejected zoning proposals, and nonbinding votes. Their legal consequences vary, but they signal growing scrutiny of the electricity, land, and public resources required by large computing facilities.
Meanwhile, developers are competing to secure power capacity years before their facilities will operate.
The U.S. relies on utilities, regulators, and local governments to negotiate these commitments. China coordinates computing infrastructure and energy resources through national development programs.
View All Both face the same uncertainty: How much infrastructure should be built before future computing demand is known, and who pays when forecasts prove wrong?
Computing moves faster than the grid
The International Energy Agency projects global data center electricity consumption rising from approximately 485 terawatt-hours in 2025 to 950 TWh by 2030. Consumption at AI-focused facilities is expected to triple.
AI accelerator generations change within years. Substations, transformers, and transmission connections are designed to operate for decades. Utilities must therefore finance long-lived infrastructure against demand forecasts shaped by a much faster technology cycle.
The challenge also involves power characteristics. Training, inference, and facility-level power management create different load profiles. On-site systems can buffer fluctuations, but utilities still need to plan for the facilities they connect.
A 500-megawatt request may require substations, transmission upgrades, and additional generation capacity before the developer consumes its first megawatt.
If the project is delayed or canceled, those costs do not disappear.
Ohio’s answer: make developers commit
In July 2025, Ohio regulators approved a dedicated tariff for large data centers served by AEP Ohio.
Qualifying developers must enter agreements with a ramp-up period of up to four years followed by eight additional years. Minimum monthly billing demand reflects contracted capacity and historical usage, subject to the tariff’s 85% contracted-capacity limit.
Developers failing specified credit-rating and liquidity tests must provide qualifying guarantees or collateral equal to 50% of minimum contractual charges over the agreement. Cancelation or substantial delay before connection can also trigger reimbursement obligations.
By February 2026, AEP Ohio reported 5,642 MW of binding contracts under the new tariff, alongside 12,219 MW contracted previously. These are commitments, not operating demand.
The tariff reduces the risk that existing electricity customers will finance infrastructure for projects that never materialize.
It also changes the economics of competition.
A hyperscaler can support long-term electricity commitments through its balance sheet. A smaller developer may need project financing, customer contracts, and external guarantees before securing comparable capacity.
Electricity access could reinforce the advantages that large technology companies already enjoy in semiconductor procurement and capital markets.
Who receives the benefits, and who pays?
PJM Interconnection operates a wholesale electricity market across parts of 13 states and Washington, DC.
In July 2026, PJM’s independent market monitor, Monitoring Analytics, estimated that data center demand accounted for approximately $6.3 billion of $16.4 billion in capacity charges associated with the latest auction. Across four auctions, it attributed $29.4 billion of $63.6 billion in total capacity charges to that demand.
These are the market monitor’s estimates of the contribution of existing and forecast data center demand to wholesale capacity charges, not amounts billed directly to data center operators or households. Actual cost recovery depends on market arrangements and retail tariffs.
Local governments see a different balance sheet.
Fairfax County, Virginia, estimated approximately $110.6 million in data center-related general fund tax revenue for tax year 2026.
Those revenues benefit the county, while electricity infrastructure costs may extend across a wider network. The Fairfax and PJM figures cover different jurisdictions and do not establish a direct subsidy.
They illustrate a structural problem: The institutions approving projects and collecting tax revenues are not necessarily those allocating electricity infrastructure costs.
Regulation becomes a competitive variable
In June 2026, the Federal Energy Regulatory Commission issued orders requiring six regional grid operators to justify or propose reforms to rules for connecting large electricity users and allocating associated costs. The proceedings do not themselves impose new tariffs.
Developers must increasingly evaluate collateral requirements, connection charges, and regulatory uncertainty alongside land prices, fiber connectivity, and computing equipment.
There is no cost-free solution.
Weak commitments expose electricity customers to speculative projects. Strict requirements can favor companies with stronger balance sheets, potentially excluding viable competitors.
America’s decentralized system permits local scrutiny and commercial discipline, but coordination across utilities, markets, and regulators can be expensive and slow.
China is testing a different model.
China’s coordinated approach
China’s Eastern Data, Western Computing initiative uses nationally coordinated computing hubs and data center clusters to align infrastructure investment with regional electricity resources, land, and network connectivity.
Suitable workloads are encouraged toward resource-rich western regions, while latency-sensitive services remain closer to major economic centers.
That distinction matters technically. Training and inference have different requirements for latency, bandwidth, and data movement. Cheap electricity does not make every location suitable for every workload.
China can coordinate electricity infrastructure, industrial development, and state-owned investment more directly than the U.S. This may reduce barriers created by fragmented planning and approval processes.
But coordination does not guarantee productive demand.
Infrastructure built around development targets can remain underutilized. When commercial demand disappoints, losses may be absorbed by state-owned enterprises, lenders, or institutions supporting the projects.
America risks delaying viable capacity or concentrating development among financially stronger companies. China risks committing capital to capacity that customers cannot use economically.
The outcome cannot be judged by the number of announced projects or installed megawatts alone.
Four tests for 2030
The competition between these infrastructure models should be evaluated against four measures:
- Time to operation: How long does it take to turn an electricity request into a commercially operating computing facility?
- Delivered cost: What is the total cost of securing reliable power, including network investments, financing, and contractual commitments?
- Utilization: How much installed computing capacity generates productive workloads rather than remaining idle or constrained?
- Financial exposure: Who absorbs losses when demand forecasts fail?
Differences in electricity pricing, financing structures, and public reporting will complicate comparisons, but these questions reveal more than headline investment announcements.
By 2030, leadership in AI infrastructure may depend as much on electricity access and financial capacity as on processor performance.
For both the U.S. and China, the decisive measure will be how efficiently investment in electricity infrastructure becomes computing capacity that customers use.
[AI Data Centers Make Power, Cooling Critical to Scaling](https://www.eetimes.com/ai-data-centers-make-power-cooling-critical-to-scaling/)
[Amperesand Aims to Smooth AI Data Center Power Spikes](https://www.eetimes.com/amperesand-aims-to-smooth-ai-data-center-power-spikes/)
[Why Data Centers Must Become System‑Aware](https://www.eetimes.com/why-data-centers-must-become-system%E2%80%91aware/)