Congress has spent months talking about the electricity demands of artificial intelligence. On the House's September 14 floor list, that debate turns into legislation.

Four bills deal with different parts of the same problem: data centers are asking utilities for enormous amounts of power, often faster than new power plants, transmission lines and substations can be built. Someone has to pay for that infrastructure. Congress is trying to make sure the answer is not everyone else with an electric bill.

The scale helps explain the urgency. Lawrence Berkeley National Laboratory estimates data centers could consume 11.8 percent of all U.S. electricity by 2030. Its range runs from 9.5 to 15.3 percent.

That is a lot of new demand landing on a grid that was not built with the AI boom in mind.

H.R. 9340, the Ratepayer Protection Act, is the most direct of the four bills.

For a data center with peak demand of at least 100 megawatts at one site or campus, the bill would require a utility rate designed to recover the full incremental cost of generation, transmission and distribution upgrades needed to serve it. Before making those upgrades, the utility would also require financial assurances or contributions from the customer.

The second part matters as much as the first.

A utility may spend years planning and building around a giant load that has not arrived yet. If the customer cancels the project, scales it down or stops buying the expected power, the infrastructure does not disappear with it. The bill says the customer that caused the investment should remain responsible for the cost.

Congress is using the Public Utility Regulatory Policies Act, or PURPA, rather than writing one national electricity rate. State utility regulators and publicly owned utilities would have to consider the new standard. States that have already adopted or considered a comparable approach would not have to start over.

The Energy and Commerce Committee advanced H.R. 9340 52-0.

That vote tells you where the easy part of this debate is. Nobody on the committee volunteered household ratepayers to subsidize a data center.

H.R. 9332 deals with forecasting.

Utilities have to estimate future electricity demand years before all of that demand exists. Data centers make the job harder because a proposed facility can represent hundreds of megawatts in one request.

The Load Forecasting Enhancement Act would have the Federal Energy Regulatory Commission create regional boards with state regulators to study how utilities make those forecasts.

The bill gets unusually specific about what regulators should examine. That includes forecast accuracy, economic-development assumptions and whether a company requesting a large block of electricity has actually made a financial commitment to the utility.

That last question separates a serious project from a company shopping several sites at once.

Forecast too low and there may not be enough power when the facility opens. Forecast too high and utilities can build expensive infrastructure for customers that never show up.

Either mistake eventually lands somewhere on an electric bill.

Another bill asks whether utilities can get more capacity out of transmission projects they are already building.

H.R. 6633, the High-Capacity Grid Act, would direct the Federal Energy Regulatory Commission to establish a class of best-available transmission conductors and require them for certain new lines and major upgrades.

In plain English, when a utility is already replacing or building a transmission line, Congress wants it to consider wire capable of carrying more electricity instead of automatically rebuilding with older technology.

There are exceptions for emergencies, safety and reliability problems, as well as projects where the added cost does not produce a comparable improvement in capacity or efficiency.

That does not solve the country's transmission problem. It can make some of the lines that get built more useful.

H.R. 9339 goes in a different direction.

The Affordable Innovation for the Grid Act would tell the Energy Department to assess how artificial intelligence and high-performance computing could improve the capacity, reliability and operation of the bulk power system.

Among the questions: Can AI speed up the studies utilities use before connecting new power plants and large customers to the grid? Where is the technology already useful? What cybersecurity or operational risks come with putting more of it into grid operations?

The bill does not order utilities to turn grid management over to an algorithm. It orders the government to figure out where the technology actually helps and where it creates new problems.

The basic principle behind H.R. 9340 has already escaped the usual partisan trench warfare.

In March, the White House got seven technology companies to sign a voluntary Ratepayer Protection Pledge. They agreed to build, bring or buy the power their new data centers need, cover required delivery upgrades and pay for infrastructure built for them even if they ultimately use less electricity than expected.

The House bill tries to put a version of that idea into the utility-regulation process.

The disagreement is over how far Congress should go.

Rep. Frank Pallone, the top Democrat on Energy and Commerce, has called the electricity bills a useful start while arguing that Congress should also address water use, pollution and other local effects from rapid data-center construction. He has even raised a moratorium if stronger guardrails do not materialize.

Public-power utilities have a different concern. Tom Falcone, who represents large state and locally owned utilities, told the committee that local conditions matter. A 100-megawatt project in a part of the grid with spare capacity is a different engineering problem from the same project where the system is already constrained.

His members support making large customers pay the costs they cause. They do not want Washington replacing state and local authority over retail electricity rates with a single federal answer.

Those positions leave a fairly visible landing zone: make companies pay for infrastructure built specifically for them, demand better evidence before utilities build around speculative demand and leave room for state regulators to decide how those protections fit their own grids.

H.R. 9340 is written in roughly that shape.

The four bills should not be confused with a national AI-energy plan.

They do not decide whether new data centers run on natural gas, nuclear power, renewables or some combination. They do not build transmission lines. They do not guarantee that electricity prices fall.

They address a narrower problem that comes first.

When one customer suddenly needs the electricity of a small city, utilities need to know whether that demand is real, who will pay for the infrastructure and whether the grid can carry the additional power.

For years, those questions mostly lived inside utility commissions, power markets and corporate negotiations.

AI has made them congressional questions.

The House has H.R. 9340, H.R. 9332, H.R. 6633 and H.R. 9339 on its September 14 floor schedule. A fifth committee bill focused specifically on best practices for shielding households and small businesses from data-center costs, H.R. 6529, advanced through committee but is not on the current floor list.

If the four scheduled bills pass, the Senate still has to act.

The larger fight will continue either way. The United States wants more AI infrastructure, and AI infrastructure wants more electricity. Congress is beginning to put rules around the bill that comes with it.