Who pays the bill for surging power demand?

In the summer of 2025, residents of Washington in the United States got an unpleasant surprise when they saw their electricity bills.

The average monthly bill for a Pepco customer had risen by about $21, largely due to higher capacity costs of PJM, according to the District of Columbia Office of the People's Counsel.

A significant factor behind that increase was the rapid growth of data centers, a development most families had not anticipated would impact their electricity bills.

But with artificial intelligence driving large increases in electricity demand, the question being asked is: Who should pay for the grid infrastructure required to support this demand?

Modern data centers are industrial power consumers. Large facilities can require hundreds of megawatts, and some proposed campuses could need a gigawatt.

The Electric Power Research Institute estimates that by 2030, US data centers could consume between 9 and 17 percent of the country's electricity, up from about 4 to 5 percent currently.

While AI offers extraordinary economic possibilities, it also requires physical infrastructure, including generation, substations, transformers and transmission lines. These are long-term assets but also require substantial investments.

A long-standing principle suggests that when new demand necessitates new investment, the cost should be borne by the entity creating the demand. PJM, which serves 13 US states, offers a clear example.

Its capacity costs rose from $28.92 per megawatt-day in 2024-25 to $269.92 in 2025-26. They are projected to rise further to $329.17 in 2026-27 and to $333.44 in 2027-28.

While data centers are not the only reason, they contributed significantly to the cost hike, adding approximately $9.3 billion to the 2025-26 capacity auction.

A deeper problem is that grid investments must begin before planners know where the demand will come from.

Forecasts can lead to physical commitments: substations are expanded, transmission lines are reinforced and generation is contracted. If the anticipated demand does not materialize, the capital committed for expansion remains idle.

That is where engineering intersects with governance. The challenge is not only determining how much capacity to build but also deciding who bears the risk when forecasts are inaccurate.

The US is beginning to address this issue. In March this year, seven major technology companies — Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI — signed a White House Ratepayer Protection Pledge. They committed to finance the generation needed for their data centers, cover the costs of the necessary power-delivery infrastructure and pay even if the electricity is not ultimately used.

In June, the Federal Energy Regulatory Commission initiated proceedings with all six regional grid operators under its jurisdiction, directing them to justify their tariffs for connecting large loads.

On Sept 16, the US House of Representatives passed the Ratepayer Protection Act by a vote of 417-3. The legislation would require state regulators and nonregulated utilities to consider standards for how large data centers should pay for the electricity infrastructure serving them. China faces a similar challenge within a different electricity system. In 2025, China became the first country to consume over 10 trillion kilowatt-hours of electricity, with internet and related services growing by more than 30 percent.

State Grid plans to invest roughly 4 trillion yuan ($596 billion) from 2026 through 2030, about 40 percent more than during the previous five years, while China Southern Power Grid is expected to invest another trillion yuan.

Much of this investment serves broader goals beyond data centers, including renewable energy integration, electrification, electric vehicles, manufacturing, reliability and economic development. That underscores the importance of cost allocation.

A transmission corridor that benefits an entire region differs from a substation or line built primarily for one large customer. A sound policy should distinguish between these scenarios.

Four guiding principles can help in this endeavor.

First, treat very large loads as a distinct customer class.

Contracts should be long enough to recover investments made specifically to serve the demand, with provisions for deposits or minimum payments.

If a project is canceled, its infrastructure costs should not be transferred to other customers.

Second, distinguish dedicated infrastructure from shared infrastructure. Infrastructure built primarily to serve one customer should be paid for by that customer.

The costs of infrastructure that benefits the wider system can be shared under transparent rules.

Third, consider the value of flexibility. A data center that is able to reduce demand during periods of system stress offers value, while one requiring uninterrupted maximum power imposes a different demand.

Electricity prices and contracts should reflect these differences.

Fourth, ensure transparency in demand forecasting. Large-load forecasts drive billions of dollars of investment. Developers seeking substantial capacity should provide financial commitments to distinguish probable from speculative demand. Forecast assumptions should be transparent for regulators and planners to test.

Data centers need not be the villain of this story. They are vital infrastructure for science, commerce, communications, and AI. Properly planned, flexible data-center demand can also help power systems manage periods of stress.

The core issue is the bargain. Every power system — whether publicly owned, regulated, or market-driven — faces the same physical constraint. New demand must be supplied, delivered and balanced. These requirements incur costs, which must be appropriately allocated somewhere.

The enduring rule is: When new demand drives new investment, the cost should follow the demand.

The author is a professor emeritus of electrical and computer engineering at the University of Minnesota and a fellow of the Institute of Electrical and Electronics Engineers and the American Society of Mechanical Engineers.

The views don't necessarily reflect those of China Daily.

If you have a specific expertise, or would like to share your thought about our stories, then send us your writings at opinion@chinadaily.com.cn, and comment@chinadaily.com.cn.