Byron Donalds Proposes Off-Grid Mandate for AI Data Centers

Byron Donalds has proposed forcing data centers above 5 megawatts off public power and water systems, a conflict now spreading through Google News coverage. His Protecting Ratepayers Act would reach both new and existing facilities. Operators would have only 180 days after enactment to comply.

The proposal goes far beyond asking technology companies to pay their share of infrastructure costs. It says covered data centers must obtain every unit of electricity, backup power, and water from sources separate from public systems.

That creates the central tension. Donalds wants continued AI infrastructure investment without making households subsidize it. Yet the bill attempts to protect public utilities by removing large data centers from them entirely, including facilities already operating.

The measure arrives during Donalds’ campaign for Florida governor and a national backlash against AI infrastructure. Communities want protection from higher bills, water competition, noise, and pollution. Developers need reliable power, dependable cooling, and predictable rules.

The Bill Draws a Hard Line at 5 Megawatts

The Protecting Ratepayers Act treats off-grid operation as a legal requirement, not a voluntary corporate promise.

Donalds announced the federal proposal on July 20, 2026. His office described it as the twenty-seventh bill he had introduced during the 119th Congress.

The published bill text covers private entities that own, operate, or maintain a data center. It also covers companies planning to operate one within five years.

A covered data center is defined through existing federal law, with an added power-demand threshold of at least 5 megawatts. That threshold captures facilities much smaller than the hyperscale campuses driving most public debate.

For comparison, Florida’s recently enacted data center law uses a 50-megawatt threshold. It calculates that level from a facility’s anticipated monthly peak load over a 15-minute interval.

Donalds’ federal proposal is therefore broader in one important respect. It does not limit the mandate to the largest AI campuses or facilities specifically built for machine learning.

The text distinguishes between new and existing data centers. A new facility is one that has not begun operating when the act takes effect. An existing facility is already operational on that date.

Beginning 180 days after enactment, a covered company could not build, own, operate, or maintain either type unless it met two conditions.

First, all energy must come from a captive power plant, on-site generation, or another source separate from the electric grid. The language explicitly includes backup energy.

Second, all water, including backup water, must come from an on-site source or another supply separate from the public water system.

The proposal also says President Donald Trump’s Ratepayer Protection Pledge would receive the full force and effect of law. That pledge emerged from negotiations intended to stop data center infrastructure costs from reaching residential bills.

Donalds’ official announcement frames the measure as a way to preserve American technology investment while protecting households. He argues that developers will accept responsibility for the resources their facilities require.

This is not simply a requirement to reimburse utilities. A developer could not satisfy the current text by paying a public utility for a dedicated substation or signing a higher-cost tariff.

The facility would need supplies physically and legally separate from the public grid and water system. That distinction makes the proposal more consequential than many state ratepayer protections.

The public draft also leaves an important procedural question. Its cover still shows a blank House bill number and an unspecified committee referral.

Donalds’ office says he introduced the measure, but the posted text does not show a numbered congressional vehicle. Readers following the story through Google News should distinguish the announced proposal from an enacted or committee-approved bill.

Google News Coverage Reveals a Bigger Political Shift

The headline is not merely that one Florida Republican introduced an AI data center bill. It is that support for AI expansion now comes with resource conditions.

Donalds remains supportive of building data centers in the United States. He opposes a statewide Florida moratorium and argues that rejecting investment would send an unfavorable message to businesses.

His position is conditional expansion. Developers can build, but they must not raise household utility bills or compete with communities for public water.

That stance reflects political pressure from several directions. Environmental groups have questioned water consumption, fossil generation, and local pollution. Conservative voters have focused on affordability, property rights, and public subsidies.

The Associated Press account reported that more than 20 Florida counties and municipalities had rejected major projects or adopted temporary bans. Local officials commonly cited utility costs, environmental effects, and disruption.

Republican strategist Matt Gorman told the AP that technology companies had fallen behind the political change. In his assessment, both progressive climate advocates and conservative populists had turned data centers into a target.

That combination matters because it weakens the industry’s familiar partisan defenses. A developer cannot assume that promises of economic growth will answer concerns about power bills and aquifers.

Donalds also faces competition in the Florida governor’s race. Republican candidate James Fishback has made opposition to data center development a central issue. Former Congressman David Jolly, a leading Democratic candidate, has called for a statewide moratorium.

Donalds is positioning his proposal between those approaches. He rejects a statewide halt while supporting stronger operating conditions.

At a July 24 campaign event, Donalds said Florida needed balance. According to local coverage, he warned that a moratorium could discourage capital investment.

This creates a more revealing opponent than Republican versus Democrat. The real contest is unrestricted access to shared infrastructure versus development that internalizes its resource costs.

Technology companies have often presented new generation contracts, renewable procurement, and water-efficiency programs as voluntary sustainability work. Political leaders increasingly treat those measures as conditions for permission to operate.

The shift changes the questions developers must answer during site selection.

A project can no longer rely only on available land, tax treatment, fiber connections, and proximity to customers. It must also show who pays for generation, how electricity reaches the servers, and where cooling water originates.

Communities are asking whether the public receives enough value for the infrastructure burden. Data centers can produce substantial property-tax revenue, construction employment, and demand for supporting services. Their permanent on-site job counts can remain modest compared with their power and land requirements.

This debate is reaching voters because utility costs are immediate and visible. An abstract promise about future AI leadership carries less weight when residents believe their monthly bills will finance private computing capacity.

Google News results now place campaign reporting, utility analysis, environmental concerns, and technical infrastructure stories beside each other. Together, they show that the social license for AI expansion has become a material development constraint.

Travis Fisher of the Cato Institute described that pressure directly. He told the AP that technology companies feared losing their social license to operate and being excluded from communities.

For developers, this is no longer only a public-relations problem. Opposition can delay zoning, change tariffs, impose water conditions, or stop a project before construction begins.

Paying for Infrastructure Is Not the Same as Leaving It

Donalds’ proposal turns a cost-allocation debate into a physical separation mandate, and that change carries major engineering consequences.

Most ratepayer-protection measures follow a cost-causer principle. The customer creating a new expense pays for the needed generation, transmission, distribution, and water infrastructure.

A utility can implement that principle through a specialized tariff. A tariff is a regulated schedule defining rates, contract terms, minimum payments, and other service obligations.

Those rules can require a large customer to guarantee revenue for years. They can also prevent households from inheriting costs if a proposed data center cancels its project or uses less electricity than forecast.

The Protecting Ratepayers Act takes another route. Its current text does not authorize a specialized grid connection after the operator pays every associated cost. It says the power source must remain separate from the grid.

That requirement removes access to one of the grid’s main benefits: pooled reliability. A grid combines many generators and transmission paths, allowing operators to respond when equipment fails or demand changes.

An isolated data center would need enough generation to meet normal demand and withstand outages. It would also need fuel security, maintenance reserves, voltage control, and equipment capable of matching supply with computing load.

Battery storage can bridge short interruptions and manage rapid changes. It does not automatically replace the sustained output needed during a prolonged generator failure or fuel disruption.

A captive natural gas plant could provide steady output, but it would require pipelines, emissions permits, maintenance, and backup arrangements. On-site solar would need substantial land and storage to support continuous operation.

Nuclear power offers high-capacity output, but new facilities require long development timelines and extensive oversight. Small modular reactor projects remain subject to licensing, manufacturing, financing, and construction risks.

Geothermal resources suit only certain locations. Fuel cells and other distributed systems can support selected projects, but their economics and supply chains vary.

The bill does not select among these technologies. Its neutrality gives developers flexibility, but it also leaves carbon emissions and local air pollution outside the central requirement.

A facility could comply with the power language through dedicated fossil generation. Protecting residential bills would not necessarily protect nearby communities from emissions, noise, or fuel infrastructure.

The scale of the challenge is growing. Lawrence Berkeley National Laboratory estimated that data centers consumed 4.4% of United States electricity in 2023.

Its energy usage report projected that the share could reach between 6.7% and 12% by 2028. The range reflects uncertainty around AI adoption, hardware demand, and efficiency.

The U.S. Energy Information Administration has also identified data centers as a major source of load growth after years of nearly flat national demand. More demand can support investment, but mismatched construction timelines create risk.

Servers can be installed faster than large power plants and transmission lines can receive approval and enter service. That timing gap can raise wholesale prices or increase reliance on existing fossil generators.

Off-grid generation appears to remove the new load from utility planning. In practice, implementation details determine whether separation is complete.

A data center might still need transmission access to move electricity from a remote private plant. A plant described as dedicated might remain electrically connected to a regional system.

The draft does not define how far “separate” extends across those arrangements. It also does not explain whether a privately owned transmission line crossing public rights of way would satisfy the mandate.

The same uncertainty applies to emergency support. A facility cannot use public power even for backup under the published language. That restriction could make data center operators build redundant generation that remains idle most of the year.

Idle capacity increases construction costs and may reduce overall resource efficiency. A shared grid can often achieve reliability with less total reserve capacity because customers do not experience peak demand or failures simultaneously.

The proposal therefore creates a real tradeoff. Physical separation can make cost responsibility easy to identify, but shared infrastructure can deliver electricity more efficiently.

A narrower rule could require full cost recovery while preserving grid access. Donalds’ language instead prioritizes a bright boundary between public systems and private computing demand.

Water Is the Harder Off-Grid Problem

Electricity can be generated on-site, but a data center cannot manufacture a dependable water supply without obtaining water from somewhere.

The bill allows an on-site water source or another source separate from the public water system. It does not define which supplies qualify or how regulators should treat groundwater.

A private well is separate from municipal pipes. It can still draw from the same aquifer that supplies residents, farms, wetlands, and public utilities.

That distinction matters in Florida, where groundwater management already involves regional permits and competing uses. Moving a project away from a city water connection does not guarantee that local water resources face less pressure.

Reclaimed water offers another option. It generally refers to treated wastewater reused for cooling, irrigation, or industrial applications instead of being discharged.

A data center could contract for reclaimed water from a private or public treatment system. However, the federal draft’s separation requirement raises questions if that supply originates from publicly owned infrastructure.

Closed-loop cooling can reduce ongoing withdrawals by recirculating water or coolant. It still requires an initial fill, treatment, and replacement for losses. The surrounding heat must also leave the system.

Dry cooling uses air instead of evaporating large volumes of water. It can require more electricity, especially during hot weather when cooling equipment works hardest.

That produces a water-energy tradeoff. Reducing direct water consumption can increase the facility’s power requirement, which enlarges the on-site generation challenge.

Florida’s existing approach is more detailed and more flexible. Governor Ron DeSantis approved Chapter 2026-65 on May 7, and most provisions took effect July 1.

The Florida data center law covers facilities with an anticipated monthly peak load of at least 50 megawatts. It allows water management districts or the Department of Environmental Protection to reject harmful uses.

A permit must be issued when the applicant establishes that the use is reasonable and beneficial, does not interfere with existing legal uses, and serves the public interest.

The law requires reclaimed water when an appropriate source is available, accessible, permitted, and technically feasible. That structure recognizes that water options depend on local infrastructure and environmental conditions.

Applications seeking at least 100,000 gallons per day must disclose all sources, volumes, and losses. They must also include conservation plans addressing cooling-water recycling, leaks, efficient fixtures, and employee awareness.

Florida requires a hearing before approval of a large-scale data center’s water permit. It also preserves local zoning authority and orders an independent study due by July 1, 2027.

The state law and Donalds’ federal proposal share a political objective but use different mechanisms.

Florida requires developers to bear service costs and navigate resource permitting. Donalds would prohibit covered facilities from using public systems at all.

The difference will matter if Congress considers the proposal. A national mandate must work across regions with different climates, aquifers, utility ownership models, and cooling options.

Water availability in the Pacific Northwest does not resemble water availability in Arizona. A single separation rule could produce different environmental outcomes in each place.

A developer might select a private groundwater source to achieve formal compliance. That choice could increase local competition even when a public reclaimed-water system would have caused less harm.

The draft does not establish water-use reporting, conservation targets, recycling requirements, or standards for proving that a private source is sustainable. It regulates the source’s ownership relationship, not the total environmental effect.

That omission does not make the ratepayer concern invalid. It shows why off-grid water is more complicated than a slogan suggests.

A strong policy must separate three questions: who supplies the water, who pays for the infrastructure, and whether the withdrawal is environmentally sustainable. Donalds’ bill answers the first question most clearly.

The Draft Leaves Enforcement and Existing Facilities Unresolved

The bill’s strongest promise is also its greatest uncertainty because the public text states a sweeping mandate without an enforcement system.

The three-page draft does not name an administering federal agency. It provides no civil penalties, compliance filings, audit process, or appeal mechanism.

It does not specify whether the Department of Energy, Environmental Protection Agency, Federal Energy Regulatory Commission, or another body would determine compliance.

The text also lacks an explicit waiver for emergencies, grid-reliability events, national security needs, or technical impossibility. It contains no staged timeline based on facility size.

Those gaps are especially significant for existing data centers. A facility designed around utility power and municipal water cannot always replace both systems within 180 days.

Generation equipment requires land, engineering, permits, interconnection decisions, fuel contracts, and construction. Private water infrastructure may require wells, treatment equipment, storage, permits, and cooling-system changes.

Some urban facilities lack enough adjacent property for those additions. Colocation centers also host equipment belonging to many customers, complicating decisions about investment and shutdown risk.

If the mandate applied exactly as written, an operator unable to comply would apparently lose the legal ability to own, operate, or maintain the facility. The draft does not explain how regulators would manage that outcome.

The 5-megawatt threshold increases the implementation challenge. It reaches beyond the largest campuses that can finance dedicated power plants.

Smaller enterprise, cloud, and colocation sites may lack the scale needed to support independent generation and water systems economically. The rule could encourage consolidation into larger campuses with greater financing capacity.

It might also change geography. Developers could favor remote sites with private generation potential, available groundwater, or nearby reclaimed supplies.

That movement would support Donalds’ preference for locating large facilities away from population centers. It could also place infrastructure near rural communities with limited regulatory staff and emergency resources.

The bill’s treatment of ownership raises further questions. A private entity planning to own, operate, or maintain a data center within five years becomes a covered entity.

Yet the operating prohibition attaches to the data center’s supplies. The text does not explain reporting obligations for a company whose plans remain preliminary or confidential.

It also does not distinguish AI computing from ordinary digital services. The public discussion emphasizes AI, but the statutory definition covers qualifying data centers generally.

Hospitals, financial institutions, telecommunications providers, cloud platforms, government contractors, and other organizations rely on data center capacity. Their facilities could fall within the same rule if privately owned and above the threshold.

That broad application requires a careful accounting of downstream effects. Higher infrastructure costs could reach cloud customers through service contracts even if they never appear on residential utility bills.

Developers might cancel marginal projects, extend construction schedules, or seek locations outside the United States. They might also accelerate investment in dedicated generation and low-water cooling.

The proposal’s supporters can reasonably argue that private costs belong in the price of AI services. If a computing facility cannot operate without shifting expenses to captive utility customers, its apparent economics are incomplete.

Critics can reasonably answer that total physical separation sacrifices the efficiencies of shared systems. They may favor tariffs, minimum contracts, collateral requirements, and infrastructure contributions instead.

Evidence on historical rate effects is also contested. Growing demand can spread fixed utility costs across more sales when sufficient capacity already exists. Rapid demand growth can raise prices when generation and transmission become constrained.

Both effects can occur under different market conditions. A national rule based on one simple causal story risks overlooking those differences.

Donalds’ claim that the measure will protect households is plausible at the cost-allocation level. It has not been demonstrated under the draft’s full operational requirements.

The next legislative version needs to clarify enforcement, exemptions, definitions, transition periods, and the relationship with state utility regulation. Without those elements, its practical effect remains uncertain.

Three Signals Will Show Whether the Proposal Changes AI Infrastructure

The bill’s importance now depends on whether its uncompromising language survives contact with Congress, utilities, and data center engineering.

The first signal is formal congressional status. Readers should watch for an assigned bill number, committee referral, hearing, or revised text.

A numbered bill with bipartisan cosponsors would strengthen the case that off-grid mandates are becoming a serious federal option. Continued reliance on an unnumbered discussion draft would weaken that conclusion.

Committee jurisdiction will also shape the measure. Energy regulators will focus on grid reliability and federal authority. Environmental lawmakers will examine water withdrawals, emissions, and permitting.

The second signal is whether Congress preserves the existing-facility mandate and 180-day deadline. Those provisions turn the proposal from a planning rule into an immediate restructuring order.

If lawmakers replace physical separation with full-cost tariffs, the measure would move toward Florida’s model. That change could improve feasibility while narrowing Donalds’ original boundary.

Watch the 5-megawatt threshold as well. Raising it would concentrate the policy on hyperscale projects. Keeping it would expose a much broader portion of the data center market.

The third signal is how developers respond before Congress acts. Companies can announce dedicated generation, long-term power contracts, reclaimed-water systems, or closed-loop cooling without waiting for a federal mandate.

Those investments would support Donalds’ argument that developers can absorb the responsibility. Requests for waivers, public subsidies, or conventional utility access would expose the limits of that claim.

Local decisions offer another useful measure. Moratoriums, zoning rejections, and permitting delays show whether corporate commitments are restoring public trust.

Florida will provide an early policy test. Its new tariff and water-permitting requirements can reveal whether full cost responsibility protects households without stopping investment.

The state’s independent study, due in 2027, falls outside the immediate three-month window. Preliminary utility filings and local hearings can still show how the framework operates.

Developers and enterprise buyers should track these decisions because infrastructure rules eventually affect computing availability, location, reliability, and contract terms. Cheap model access depends on physical systems that many software teams rarely examine.

Knowledge workers following the debate can use a structured AI workflow to separate enacted rules from campaign promises and draft legislation. That distinction is particularly important when Google News headlines compress several different Florida and federal measures into one narrative.

Donalds has identified a problem that neither technology companies nor policymakers can ignore. AI expansion requires large, continuous flows of electricity and cooling capacity, while communities want protection from costs and scarcity.

His answer is unusually direct: separate private computing infrastructure from public utility systems. The simplicity makes the proposal politically legible, but it does not make implementation simple.

The decisive question is whether lawmakers preserve that physical separation or convert it into a requirement that developers pay every attributable cost. Watch the next bill text, the first committee action, and the first major developer response. Those signals will show whether this Google News story marks a durable infrastructure rule or a forceful opening position in a longer ratepayer fight.