A Megawatt Needs Permission

Bottom line up front

The AI data-center race has moved beyond finding electricity. It now has to produce a public bargain.

Texas regulators approved two legs of a 765-kilovolt transmission backbone, a very-high-voltage system intended to move large amounts of power across long distances. ERCOT is redesigning how it studies large loads after requests reached 438 gigawatts, 89 percent of them tied to data centers. Utilities in Louisiana and Texas are arguing that large new customers can lower bills if those customers carry enough of the new grid cost.

Meanwhile, Columbia County, Georgia, tied a proposed Google data center to ratepayer protection, limits on noise and water use, a decommissioning plan and a possible property-tax benefit. That last benefit still depends on a November referendum and a final relief mechanism.

These are not separate stories. Together they show the next bottleneck.

The market counts power announcements as though a megawatt were finished inventory. It is not. A megawatt that cannot survive a rate case, a permit hearing or a neighborhood meeting is not usable capacity.

The next bottleneck is not only power. It is permission.

Texas approves a backbone, not capacity

The Public Utility Commission of Texas approved two 765-kilovolt transmission projects on August 28. One is expected to run roughly 235 to 270 miles from the Permian Basin toward North Texas and enter service in 2028. The second is expected to run roughly 160 miles and enter service in 2029.

Approval matters. Steel, transformers, conductors, engineering and construction can move from an abstract need into a defined project.

But the approval does not turn every load request into an operating data center. A transmission backbone still needs substations, distribution equipment, generation, interconnection work, site permits, customer commitments and time.

That distinction is visible in the ERCOT queue.

ERCOT said large-load requests had reached about 438 gigawatts by June, with data centers representing 89 percent. Texas does not have 438 gigawatts of near-term data-center demand. It has 438 gigawatts of requests, including projects that can overlap, change, move or never get built.

To put 438 gigawatts in perspective, ERCOT's grid set a preliminary record peak demand of 91.1 gigawatts on July 22. The request queue is 4.8 times that peak. It is also equal to 58 percent of the 759.2-gigawatt Lower 48 peak the US Energy Information Administration reported for July 2025. Those are comparisons of requested load with actual peak demand, not forecasts of what will be built. They show why even a small conversion rate would require a grid-scale response.

ERCOT's new Batch Zero process will study large projects together, allocate available capacity and identify the transmission needed to serve them. It also creates separate paths for projects that bring on-site generation or agree to curtail demand when the grid is tight.

The queue is not a forecast. It is a competition for permission to connect.

A megawatt has five owners

An announced data center may begin with a developer and a power target. Usable capacity has at least five economic owners.

The operator must commit capital and accept enforceable obligations if the project arrives late, shrinks or disappears.
The utility must decide what generation, transmission and distribution equipment the project requires.
The regulator must determine which costs belong to the large customer and which can enter the wider rate base.
The local government must decide what tax treatment, land-use rules and public benefits make the project acceptable.
The community must live with the noise, water use, emissions, construction traffic and visual footprint after the press release fades.

The fifth owner may not sign the power contract, but it can determine whether the first four keep theirs.

This is the public bargain: who pays for the infrastructure, who receives the benefit and who bears the operating cost.

Powered land without political durability is not finished infrastructure.

Who pays for the new grid?

Rate design is the set of rules that determines which customers pay which utility costs. It has become part of AI infrastructure.

The basic risk is simple. A utility may build generation and wires for a very large customer, only for that customer's project to arrive late or consume less power than promised. The infrastructure can then become a stranded asset, meaning an investment that is no longer fully used while its costs still have to be recovered.

The US Department of Energy identified fair cost allocation, stranded-asset risk and resource adequacy as central issues in electricity rate design for large loads. Virginia's legislative research agency reached a similar conclusion: existing rate structures allocate direct costs, but rapid data-center growth can still raise systemwide costs and expose other customers if demand does not materialize as planned.

The strongest contracts attack that risk directly.

Entergy Louisiana says Meta will pay its full cost of service under a new agreement. Entergy projects the arrangement will produce $2 billion of incremental customer savings over 20 years, in addition to $650 million announced previously. CenterPoint Energy says up to 14 gigawatts of eligible or studied large-load projects could produce more than $5 billion of customer savings over a decade by paying more infrastructure charges and spreading fixed grid costs across a larger base.

Those figures are company projections, not realized savings. They depend on the loads arriving, staying connected and paying what the contracts require.

The principle is more important than the headline number.

Good rate design can convert private demand into shared infrastructure. Bad rate design can convert private demand into a public bill.

The local scoreboard is wrong

Data centers are frequently described as economic-development projects. That label is incomplete.

Virginia's Joint Legislative Audit and Review Commission found that the industry's economic benefits come mostly from capital investment and construction. A typical 250,000-square-foot facility has about 50 full-time workers, and roughly half may be contractors. The permanent job count is small relative to the land, power and equipment inside the building.

The durable local value can instead come through the tax base. In established Virginia markets, data centers accounted for 1 percent to 31 percent of total local tax revenue, depending on the locality.

That benefit is real, but only after incentives.

Washington's legislative research agency studied a sales-and-use-tax preference for data-center refurbishments. It found that the preference saved recipients $42.4 million over four years but did not cause new data centers to be built. The associated property value and property taxes increased, yet the agency could not establish how much of that growth the incentive caused.

The lesson is not that incentives fail. It is that gross investment is the wrong score.

The relevant measure is the durable local benefit after tax concessions, infrastructure obligations and operating costs.

The Columbia County compact

Columbia County offers a useful model because its agreement tries to make the public bargain explicit.

County officials said Google would pay 100 percent of the project's energy costs so existing ratepayers would not bear added infrastructure, transmission or capacity expense. The framework also includes sound limits, lighting controls, water protections, visual screening and a plan for the end of the facility's useful life.

Local reporting said the county development authority later approved a memorandum of understanding for at least $17 billion with Kinetic Infrastructure Solutions, an Alphabet subsidiary. The bond structure is described as nonrecourse to the county, meaning the private company rather than the county would be responsible for the debt.

The county also expects project revenue to support property-tax relief. That outcome is conditional. Voters still have to approve the relevant mechanism in November, and the final amount and design have not been established.

That qualification makes the case more instructive, not less.

A credible project separates what is already protected from what is still promised. Ratepayer protection can be written into the energy arrangement. Noise, water and decommissioning rules can be written into development terms. A tax benefit that still needs a vote should remain a contingent benefit until the vote occurs.

The cheapest site can become the most expensive if its power arrangement cannot survive public scrutiny.

A permit rewrites the bill of materials

Permission does not sit beside the bill of materials. It rewrites it.

xAI said in July that it had agreed to remove 69 mobile natural-gas turbines from its Memphis site by July 2027 as a permanent 1.2-gigawatt power plant comes online under a Clean Air Act permit. The company said the permanent configuration will use 41 turbines, with emissions controls, sound walls and silencers.

Those are company statements about a planned transition, not proof that the permanent plant is complete. The example still shows the physical consequence of a permit.

Mobile turbines can shorten the path to initial power. A permanent permitted plant requires different equipment, controls, construction, suppliers and monitoring. Sound rules change enclosure and site design. Water rules can change cooling architecture. Curtailment obligations can change backup power and storage. Decommissioning rules create an end-of-life liability before a rack is installed.

Regulation is not an external footnote to the infrastructure stack. It is a design input.

The five-gate public-bargain test

We would evaluate a large-load project through five gates.

Cost allocation: Does the large customer pay the generation, transmission, distribution and operating costs created for it?
Commitment: Are there minimum payments, collateral or other protections if the load arrives late, shrinks or leaves?
Local benefit: What tax revenue, infrastructure or direct public benefit remains after incentives and abatements?
Operating burden: How are noise, water, emissions, land use, reliability and decommissioning controlled?
Durability: Can the agreement survive regulatory review, elections, community scrutiny and a weaker demand environment?

The first four gates describe a deal. The fifth determines whether it lasts.

This is a better way to compare two nominally identical sites. One may have cheaper power on paper. The other may have a contract that protects ratepayers, a completed transmission plan, enforceable operating rules and a visible local benefit.

The second site can reach service first because fewer assumptions remain political.

Where the return migrates

If permission becomes scarce, value can migrate toward the companies that turn an approved arrangement into operating capacity.

That includes grid equipment, high-voltage engineering and construction, on-site generation, storage, power-delivery electronics, cooling and the systems that monitor and control them.

Where demand can show up inside Otto's current coverage universe:

Generation and on-site power: CEG, VST, NEE, CMI and BE.
Grid hardware and construction: ABB and PWR.
Storage and power delivery: FLNC and MPWR.
Cooling and site systems: VRT and MOD.

This map is illustrative. Universe membership is not a rating, and an approved project is not the same as an order, revenue or profit for any supplier.

The strategic advantage belongs to the project that clears institutional gates without transferring an open-ended risk to everyone else. The supplier advantage belongs to the layer that remains necessary after the final agreement changes the site design.

What would strengthen, weaken or break the thesis

Strengthen: more utilities create enforceable large-load tariffs; regulators require meaningful customer commitments; transmission approvals convert into procurement; communities accept projects with visible benefits; and operators disclose the path from requested power to service.

Weaken: permitting and rate cases remain slow but do not change project selection, contract economics or equipment design; hyperscalers consistently absorb every incremental cost without delaying construction; or communities accept projects without additional protections.

Break: announced megawatts convert to operating capacity at similar rates regardless of tariffs, permits and local agreements, or the cost of permission proves too small to influence location, schedule and supplier selection.

The AI infrastructure market has spent years asking where the power will come from. That question is incomplete.

Power needs a wire. A wire needs approval. Approval needs a durable answer to who pays and who benefits.

Stop counting announced megawatts. Count permitted, ratepayer-protected and community-accepted capacity.

That is what we read this week.

Otto Analytics

The Weekly Read returns next week.

Download this issue (PDF)

Past issues

No. 11 · September 3, 2026 · Who Owns the Rack When Everyone Builds a Chip? Read → No. 10 · August 23, 2026 · When Robots Become Participants Read → No. 09 · August 14, 2026 · When GPUs Become Collateral Read → No. 08 · August 7, 2026 · $733 Billion Is Not an Order Book Read → No. 07 · July 31, 2026 · When Does Space Become the Cheapest Place to Compute? Read → No. 06 · July 24, 2026 · Open Models, Closed Economics: Who Actually Pays for “Free” AI? Read → No. 05 · July 19, 2026 · The Cheaper Intelligence Gets, the More Infrastructure It May Need Read → No. 04 · July 16, 2026 · The Machine Is Only the First Sale Read → No. 03 · July 10, 2026 · The Second Door Read → No. 02 · July 3, 2026 · The Purchase Order Read → No. 01 · July 2, 2026 · The Robot Is Not the Product Read →

Sources and assumptions

ERCOT queue figures describe requests, not projects expected to enter service. ERCOT's 2026 peak-demand record remains preliminary pending final settlement, and the Lower 48 comparison uses EIA's reported July 2025 coincident peak. Entergy and CenterPoint savings figures are company projections and depend on future loads, rate treatment and contract performance. Columbia County's energy-cost protections and operating conditions are county-disclosed; the proposed property-tax relief remains conditional on a referendum and final design. The xAI turbine transition is company-described and incomplete. Virginia and Washington findings are government program analyses, not forecasts for every market. The five-gate public-bargain test and the replacement denominator are Otto Analytics frameworks.