While tech titans race to dominate AI benchmarks, their ambitions run on electricity, and lots of it. Take xAI’s Colossus supercomputer in Tennessee. It’s drawing 150 MW from the grid (enough to power 124,000 homes). Before its full grid connection, xAI bridged the consumption gap by operating 20 unpermitted gas turbines at double their permitted capacity; a reality only discovered by civil society-commissioned flyovers. Another independent satellite analysis then revealed a 79% spike in peak pollutants near the facility. The Colossus example isn’t an isolated case of messy siting. It is part of a broader U.S. trend: opaque data center buildouts leaving Americans with unregulated externalities.
Meanwhile, households are footing the bill. In Virginia, residents near the world’s largest data center cluster will have to pay an extra $276 a year by 2030, and consumers across just seven PJM states have already paid $4.3 billion to fund new transmission infrastructure for the industry.
The real kicker? With no standardized federal or state energy-use reporting requirements, data center operators can choose what they disclose to the public, lawmakers, and utilities.
Communities deserve to know whether new data centers will drive up costs, strain infrastructure, or crowd out other economic developments.
Ratepayer Protections Meet the White House
These cost increases are getting the White House’s attention. This month, President Trump signed a “Ratepayer Protection Pledge” with tech executives to foot the electricity bill for their buildouts. This alignment between industry and the White House signals a shared recognition that data centers are now one of the fastest-growing drivers of electricity demand in the United States.
But these commitments dodge a fundamental problem: we don’t actually know how much, or how, AI data centers consume. The data center stack’s load behavior remains a black box. Without standardized reporting, policymakers and communities are left guessing, unable to manage what they can’t measure.
In most jurisdictions, no one outside the companies themselves can see the full breakdown of how data center electricity is actually used. Grid operators and utilities can track the total load at a given facility. But they can’t see what’s driving it, including computing, cooling, and IT overhead, or how those demands will grow. This is a problem because data center workloads can ramp up or down in seconds, making their behavior difficult to track in real time and to plan for when connecting them to the grid. For example, without knowing how much power is going to computing versus cooling, utilities can’t model how the facility will react to external factors like a heatwave (which spikes cooling demand) or a new AI model release (which spikes computing demand).
Right now, grid operators and planners lack sufficient data to understand the different load behaviors within AI data centers. As a result, these operators may underschedule or overschedule power delivery. If they underschedule and there isn’t enough power to meet a sudden AI surge, they risk a blackout. If they overschedule, they waste money and fuel on power plants that aren’t required.
Blindspots to Blackouts
Many of these load risks don’t appear as line items on a bill, but they are evident in how strained the grid becomes when large, volatile loads are added.
As power grids are pushed to their limits, the consequences of planning with incomplete data are no longer abstract. In 2024, 60 data centers in Virginia suddenly went offline. Grid operators and the local utility had to take emergency actions to prevent wider outages. Texas is now facing a similar moment. As a surge of new data center projects is rapidly increasing the state’s electricity demand, local leaders are grappling with how to maintain grid reliability without clear, standardized data on these facilities’ load profiles.
Together, these cases point to a national problem: without standardized, public reporting on AI data center electricity use and load behavior, the institutions responsible for managing America’s power system are forced to react to crises rather than prevent them.
A Policy Proposal: Start with the Basics
Fixing this system requires an important first step: standardized measurements.
Congress should mandate the Energy Information Administration (EIA) to establish a standardized reporting methodology for data centers. That means reporting the full consumption stack:
IT and related equipment
Power sources
Electricity delivery and backup systems (e.g., generators)
Overhead energy components (e.g., cooling)
Facility size
Miscellaneous loads like lighting and security
To protect proprietary data, a third-party anonymization process would disaggregate sensitive data before publication. EIA would then produce standardized, anonymized statistical products that the North American Electric Reliability Corporation (NERC) could use in reliability assessments and the national labs could use for forecasting.
The payoff extends beyond transparency. When data center operators are required to measure and report their consumption in detail, they have a direct incentive to optimize it. And when grid planners finally have accurate data on what data centers actually consume, and how that consumption behaves, they can stop guessing and start planning.
Measure to Manage
Transparency can yield three additional benefits.
Determining Fair Share
As tech companies promise to pay their “fair share,” reporting would help determine what that share really is, and what the risks are if the AI boom cycle turns to bust.
Curbing Misleading Claims
Transparency would curb AI companies from promoting selective or misleading consumption claims. Today, companies can point to renewable energy purchases or efficiency improvements without revealing the full scale of their power use. A standardized reporting framework would create a common yardstick, enabling consumers to assess whether corporate climate pledges align with real-world impacts.
Incentivizing Efficiency
Transparency would create real incentives for efficiency. When energy use is visible, it becomes something companies compete to reduce. Just as fuel economy standards pushed automakers to innovate, disclosure would reward AI developers who design more efficient models and data centers.
Congress Must Act
From South Memphis to Texas and Virginia, the pattern is the same: communities are being asked to live with the costs and risks of AI infrastructure that regulators still can’t fully track.
Congress has the authority to change this. By requiring standardized, first-party reporting of AI data center energy use, lawmakers can take the first step toward responsible planning and power reliability.