Bring Your Own Generation Comes With Its Own Risks

Last night, between the trumpeting of fossil fuels and the lengthy honors awarded to both veterans and hockey players, President Trump devoted a portion of his State of the Union address to announcing a “ratepayer protection pledge,” under which big tech companies pay for their own power plants for data centers — a show of how central energy prices are becoming to today’s affordability debate.

Electricity in the United States is rapidly becoming expensive and unreliable. Vast swaths of the United States are at elevated risk of outages. January’s winter storms wiped out power for millions of Americans from Louisiana to Brooklyn. In 2025, utilities requested a record $31 billion in rate increases from captive customers. Gas and electricity prices are the two highest drivers of inflation.

The main driver of these new stressors on the grid: the expected $6.7 trillion to be deployed in data centers by 2030.

Policymakers at all levels of governments are coalescing on a strategy for dealing with rising data center demand that mirrors Trump’s ratepayer protection pledge: “bring your own generation,” or BYOG. Bipartisan bills introduced in Washington by Senators Chris Van Hollen, and Josh Hawley and Richard Blumenthal; and by Representatives Rob Menendez and Greg Casar, among others, would require hyperscalers like Meta, OpenAI, and Microsoft to pay for their own power plants and grid upgrades in order to plug in. Michigan, Oregon, Florida, Washington, Georgia, Illinois, and Delaware are all at various stages of enacting BYOG legislation for data centers.

BYOG would create something like a regulatory sandbox for data centers, insulating utilities and ratepayers from the risks of data center demand. But while efforts at consumer protection are important, these policies do not grapple with the scale of data center deployment.

A sandbox won’t withstand a tidal wave. Over the next five years, the equivalent of 17 to 32 New York Cities’ worth of electricity demand is expected to be added to the grid, more than half of which will come from data centers. This incredibly wide estimate means that generators risk overbuilding.

Amidst all this uncertainty, BYOG does not address who pays for new capacity in the event the AI bubble bursts and energy infrastructure is left stranded. Neither does BYOG address the drastically mismatched lifetimes of the chips powering AI (one to three years) and power plants (25 to 30 years). The Federal Energy Regulatory Commission expects 22 New York Cities’ worth of generation to be added to the grid by 2028. Who pays for all of this generation in a decade if even 5% of projected data center demand disappears?

AI is a promising technology, but that does not prevent it from being overvalued. Policymakers must consider the risks when data centers eventually disconnect from the grid, not just when they interconnect. This means ensuring that ratepayers and taxpayers are not left footing the bill for stranded energy infrastructure if data centers disconnect prematurely.

Rather than cordoning off data centers from the rest of the electricity market, policymakers should take a stronger hand in planning these deployments for social and economic benefit. Colocating datacenters with energy-intensive industries and requiring long-term commitments from hyperscalers are more efficient solutions that would also make new data centers more politically palatable.

Public sentiment has turned overwhelmingly against data center development. These vast facilities create relatively few jobs beyond their construction, but colocated with the manufacture of energy-intensive products like aluminum, steel, or fertilizer, suddenly they’re supporting employment. Colocation will also help diversify economic growth. Data center investment was responsible for a whopping 92% of GDP growth in the first half of 2025, creating a potentially dangerous dependency on continued expansion.

There are also simple legal guardrails that can provide a first line of defense against stranded costs. One is requiring long-term power purchase agreements between hyperscalers and generators. Thirteen bipartisan governors and the Trump administration recently urged the country’s largest grid operator, PJM Interconnection, to require 15-year generation contracts for hyperscalers. Notably, Van Hollen’s bill would only require states to “consider” the extension of “minimum utility contract lengths,” while the Hawley/Blumenthal and Menendez/Casar bills make no mention of contract length or stranded costs.

Hyperscalers can also curtail usage during peak demand, a policy that has seen bipartisan support in Texas. A now-famous study from Duke University last year found that if data centers were to curtail 1% of their usage during peak hours, they could avoid installing 126 gigawatts of new generation — that’s 21 New York Cities’ worth. Lawmakers have since taken to the idea. Several states are considering mandating so-called “demand response” programs, and Representatives Alexandria Ocasio-Cortez and Kathy Castor inserted a federal study on demand response into the appropriations bill Trump signed in January.

Regardless of how it’s done, ratepayers should not pay full freight for the tidal wave of infrastructure coming online, and most utility balance sheets should not be exposed to that risk. BYOG’s flaws have more to do with what it leaves out — namely that the planning of significant parts of our economy and electric system is left to tech companies, and little thought is given to the long-term ramifications of overbuilding. Rather than deal reactively with the nasty politics of a bailout, policymakers should make muscular interventions now to reduce risks for ratepayers and taxpayers.

Energy markets are not free markets. For the past century they have been heavily regulated at the state, regional, and federal level. Any discomfort with planning (or “statutory tools”) must be set aside if policymakers are going to efficiently manage the growth of data centers.