Johannesburg, South Africa Correspondent

The easiest way to misunderstand the artificial intelligence electricity problem is to begin with the computer. Begin instead with the wire. That wire runs through a substation, a transmission network, generating stations, fuel deliveries, control rooms, regulatory decisions, and somebody’s balance sheet. If one of those fails, the most advanced computer in the world becomes a rather expensive way to produce heat. I say this from Johannesburg, where electricity has never been an abstraction and where a politician announcing megawatts has often been the first indication that somebody should look more closely at the actual megawatts. We have lived through the difference between installed capacity and available capacity. I would prefer that America learn from the experience without having to repeat it.

There is a genuine new demand problem. Lawrence Berkeley National Laboratory’s June 2026 update estimated that American data centers could account for 11.8 percent of national electricity consumption in 2030, within a modeled range of 9.5 to 15.3 percent. Those figures are estimates, not a meter reading from the future. The International Energy Agency’s earlier global outlook anticipated approximately 945 terawatt-hours of data-center consumption in 2030, about twice its 2024 level. Two useful observations follow. The electricity is substantial, and the uncertainty is substantial too. A responsible planner should neither declare the forecast an exact promise nor ignore it because its range is wide. The question is what happens if the load appears faster than the steel, transformers, gas supply, and operating discipline required to serve it.

When I worked around banking clients exposed to municipal utilities and energy markets, a recurring danger was confusing an investment announcement with an operating asset. A financing mandate, a construction contract, a planning approval, a generator order, and a plant delivering dependable power are different stages. Yet the public conversation repeatedly puts all five in the same column marked progress. The North American Electric Reliability Corporation has warned that the pace of new resources is lagging the rise in projected demand, and its 2025 long-term assessment identified data centers as a major source of that pressure. It did not say that a blackout tomorrow is inevitable. It identified the conditions under which risk rises, particularly if promised supply arrives late or an extreme-weather event exposes how thin reserves have become. That is a warning to be acted upon, not a political slogan.

The case of PJM makes the distinction unusually plain. Its July 2026 capacity auction failed to secure enough qualified resources to satisfy its stated reliability requirement; the shortfall was 6,831 megawatts. Capacity procurement is not the same as a prediction that lights will fail on a particular date. It is evidence that the forward market sees a tightening balance between resources and anticipated peak needs. On October 12, federal regulators approved a registry that should improve PJM’s view of large data-center loads, while rejecting its proposed emergency curtailment mechanism in the form presented. Better accounting is a necessary step, but a registry cannot spin a turbine or cool a transformer. To pretend otherwise is precisely the institutional habit that has done so much damage to electricity systems elsewhere.

There is also the problem of percentages. People hear that data centers are still a modest share of worldwide electricity and conclude that there is no serious issue. The global percentage is not the thing that trips a regional system. A server campus may require hundreds of megawatts concentrated at one location, and it may want that power most hours of the day. A nation may have adequate annual generation yet insufficient transmission to bring it to one rapidly developing county. A grid may have plenty of energy in spring and insufficient accredited capacity during a hot evening. Electricity is both a commodity and a synchronized service. Demand and supply must balance moment by moment. Anyone who ignores place, hour, and contingency while quoting an annual number is not describing the system the operator must actually keep alive.

I am encouraged that American utilities and technology firms are beginning to treat power as a primary input rather than an afterthought. Google and Constellation announced in October a long-term arrangement designed to unlock 890 megawatts of new nuclear capacity through improvements at existing reactors. American Electric Power also described a 702-megawatt fuel-cell project near Cheyenne under a twenty-year agreement for Google. These are more consequential than promises to purchase an accounting certificate after somebody else generated a unit of electricity. A firm signing a long-duration agreement can improve the investment case for new or upgraded assets. But the details matter: new firm generation, an existing plant’s output contracted to a different buyer, and a financial claim to clean power are not the same physical contribution to reliability.

The words ‘bring your own power’ sound appealing, and sometimes they describe a sensible bargain. The large customer underwrites a source of supply that would not otherwise have been built, pays the needed grid connection costs, and accepts responsibility for the consequences if the project is late or cancels. Yet the arrangement can become less attractive if a supposedly private plant continues to depend on public transmission for backup, system balancing, voltage support, and emergency reserves without paying fairly for those services. There is no such thing as an electrical island merely because a publicist has drawn a dotted line around a data center. Some facilities will genuinely operate isolated systems, but that is an engineering and fuel-security claim that must be demonstrated, not assumed.

I would also like Americans to examine the relationship between fuel supply and electrical supply with something approaching adult seriousness. Gas-fired generation can be built faster than many other firm alternatives, though turbines, pipeline capacity, permits, and interconnection studies impose their own limits. A gas plant that cannot receive fuel during a winter emergency is not the reliability solution its nameplate rating suggests. Nuclear stations provide valuable firm output, but a reactor project does not become available simply because a technology chief signs a memorandum of understanding. Wind and solar can expand generation economically, and batteries can shift energy and provide rapid system services, but they must be credited according to how they perform under the particular conditions being planned for. The answer is a portfolio with honest assumptions, not a theological argument about one perfect technology.

From South Africa I am especially wary of the idea that reliability can be purchased after years of neglect. Once a generating fleet, skilled workforce, or maintenance culture has deteriorated, the catch-up period is painful. The same applies to grid transformers, switchgear, skilled linemen, and protection engineers. Utilities have good reason to resist large customers demanding a connection date before network upgrades can be completed safely. They also have an obligation not to hide ordinary bureaucratic incompetence behind the word safety. The test is transparent engineering: show the studies, identify the constrained elements, present the options, put dates and costs on the record, and explain which risks remain under each alternative. The public deserves something more illuminating than either a corporate ribbon cutting or a regulator’s assurance that everything is under control.

Artificial intelligence can itself help operate an electrical system. Forecasting, inspection, demand prediction, anomaly detection, and maintenance scheduling offer genuine possibilities. But using AI to manage the grid does not excuse placing unrealistic AI loads on that grid. Better forecasting does not substitute for enough capacity. Predictive maintenance cannot replace a damaged transformer that has a long manufacturing lead time. Software may extract efficiencies from physical infrastructure; it does not abolish physics. This is particularly important when the industry describes efficiency improvements in chips and cooling. A more efficient computing task may cost less to perform and therefore be performed more often. Overall consumption can rise even while consumption per task falls. Efficiency is desirable, but it is not an automatic ceiling on total demand.

Cost allocation is perhaps the most serious political question. A new campus can benefit its host community with jobs, taxes, and investment. It can also create a demand for transmission and generation assets whose cost is spread through the general rate base. A residential customer should not finance a speculative industrial project merely because the utility expects the tax revenue to look impressive at the next press conference. Ohio’s special data-center tariff is an instructive response: the customer must make serious financial commitments, including minimum payments and provisions that discourage abandoning a reserved block of capacity. Such terms do not solve every problem. They make a project developer reveal how confident it really is in its plans. If a company wants an option on scarce grid capacity, it should not get that option without a meaningful price.

There is a dangerous temptation to make the cost argument entirely moralistic. One side says computing is wasteful while households need heat; another says anyone who questions a data-center buildout is hostile to technological progress. Neither position helps an engineer dispatch a grid. Hospitals, factories, rail networks, payment processors, and households all require reliable power, and computing services can support them. Still, not every proposed AI workload has equal value, and no infrastructure planner can assume limitless public willingness to underwrite it. The correct questions are ordinary ones: Will the customer pay the actual incremental cost? Will other users remain protected? Is the load flexible? Does the project improve or weaken the region’s capacity to respond to a crisis?

Flexibility deserves more attention than it currently receives. A facility that can reduce part of its nonessential computing during a short system emergency, schedule selected work to periods of abundant generation, or use properly controlled storage may be a better citizen of the grid than one insisting that every process run flat out every hour. Not all computer work can be interrupted; financial transactions, clinical systems, and real-time services have obligations that batch model training may not. The differentiation belongs in carefully negotiated contracts and verified operating tests. Emergency curtailment rules must be lawful, transparent, and predictable. Utilities should not be allowed to improvise an arbitrary hierarchy of users after conditions become dangerous. Contractual clarity before the crisis is cheaper than litigation afterward.

The electricity system is also being reshaped by geopolitics and supply chains. The same firms announcing data-center investments need copper, transformers, cooling equipment, power electronics, turbines, and skilled technicians. A delay in any component can jeopardize a planned opening. If every technology company and utility buys into the same aggressive schedule, quoted lead times may become collectively impossible. This is familiar to anyone who has watched a mining expansion or power-station refurbishment. A market can signal scarcity through higher prices, but it cannot manufacture a highly specialized transformer overnight. Planners need inventories, supplier diversification, and honest commissioning schedules. In an emergency they will discover that megawatts mentioned in separate investor presentations cannot be added together as if they were already dispatched.

I also suspect that America’s debate will miss the distributive issue if it looks only at the national average price of electricity. Large new projects cluster in places with available land, tax incentives, and favorable utility relationships. The people living beside the resulting substations and transmission corridors may not be the people receiving the largest economic benefit. Poorer households are more exposed to rate increases, while well-capitalized technology firms can negotiate bespoke arrangements. This is not a reason to deny permits automatically. It is a reason to insist on a public account of who pays for land, grid upgrades, backup services, road access, water, and environmental impacts. Local consent becomes more durable when costs and benefits are visible rather than hidden in a multipage tariff no one outside the industry can understand.

One lesson from load-shedding is that public patience can vanish with surprising speed when people believe a system is being managed for insiders. Power interruptions harm smaller enterprises disproportionately; a global cloud business can finance backup systems, while a neighborhood manufacturer may not be able to. That asymmetry should temper the enthusiasm of officials who promise the same electricity unit to new industrial developments without demonstrating how the extra demand will be accommodated. Reliability is an economic development policy in its own right. A regional government that attracts a prestigious data center but undermines electricity service for an existing industrial district has not necessarily grown its economy. It may simply have changed which corporate logos appear in the glossy brochure.

I would judge meaningful progress using five tests. First, has a promised generator or network upgrade actually entered service and performed under load? Second, are customers making enforceable long-term commitments proportionate to the investments made for them? Third, are system operators publishing stress-tested forecasts rather than multiplying every preliminary inquiry into guaranteed demand? Fourth, are there credible emergency operating arrangements, including contractual flexibility where feasible? Fifth, can ordinary customers see evidence that the project has not transferred disproportionate risk to their bills? These tests reward both private initiative and public competence. They also permit different regions to choose different fuel mixes according to what can be built, financed, and reliably operated.

There is a strategic opportunity here. If the United States builds a power system capable of accommodating new computing and manufacturing without sacrificing ordinary reliability, it will demonstrate an industrial advantage that is more durable than a lead in one generation of chips. It will also show that democracies can approve, build, and operate large infrastructure in a way that survives public scrutiny. If it fails, the expensive computers will be the least interesting part of the story. The bigger failure will be an inability to coordinate electricity, finance, regulation, and honest public accounting. South Africans understand too well what such failures cost. Americans still have the opportunity to avoid them. I would recommend taking the opportunity seriously before the lights flicker.

Long Island, New York Correspondent

The electricity business has acquired a new class of applicant who talks as though a connection to the grid were an entitlement. I have spent enough years with zoning files, assessment challenges, and the uncomfortable correspondence that follows a bad application to know the difference between an entitlement and a request. A data-center developer may have bought the land, announced the investment, and persuaded local officials to pose for pictures. None of that establishes an unconditional right to hundreds of megawatts whenever the applicant wants them. A public utility has legal duties to existing customers. The governing contracts and tariffs matter, and so does the record that leads to a regulator’s decision. For all the language about artificial intelligence remaking the world, the central questions here are delightfully old-fashioned: What was promised, by whom, on what authority, and who bears the cost if the promise cannot be kept?

I take the electricity forecasts seriously, with the caution that any reasonable lawyer’s office brings to projections. The latest Lawrence Berkeley National Laboratory update, issued in June 2026, describes scenarios in which data centers account for 9.5 to 15.3 percent of American electricity consumption by 2030, with a central estimate of 11.8 percent. Those numbers speak to a major new demand category. They are not evidence that every proposed data center will be completed or that all requested capacity will be used simultaneously. Both errors are expensive. Underplanning can put reliability at risk. Overplanning can saddle the public with transmission facilities built for customers who vanish when the investment cycle turns. A responsible regulatory framework must make both possibilities visible before the concrete is poured.

One development this week illustrates the legal difficulty. Federal regulators approved a large-load registry for PJM but declined to accept a proposed approach that would have allowed certain data centers to be curtailed during emergencies. Their reported objection concerned legal authority and an inadequate justification for selecting these facilities over other users. Some will interpret this as Washington putting paperwork ahead of reliability. I read it differently. A government can act urgently and still explain its legal authority and articulate a fair rule. When officials write a curtailment policy that may interrupt an enormous industrial customer’s operations, they should know what existing tariffs permit, what agreements promise, whether similar customers are treated consistently, and how an emergency decision will be reviewed. A court cannot manufacture electricity, but sloppy legal drafting can make scarcity harder to manage.

PJM’s problem is tangible. Its July 2026 forward capacity auction procured fewer qualifying resources than its announced reliability requirement, missing by 6,831 megawatts. An auction result is not a scheduled blackout. It is a significant warning that the system is struggling to arrange enough future capacity on the terms offered. The next question is what contractual obligations connect the anticipated new demand to the generation or grid improvements required to serve it. If a large customer insists upon firm power and a firm deadline, it should be willing to commit to payments sufficient to support the facilities the utility must develop. It should not be able to reserve the capacity and walk away after the utility has borrowed money and obtained easements in anticipation of the promised load.

Ohio has begun to address that risk with a dedicated data-center tariff requiring significant financial commitments, including long-term contracts, minimum demand payments, collateral, and exit provisions. These requirements are not hostile to investment. They are a means of separating a serious project from an application submitted because a place in the interconnection queue might someday be valuable. The utility’s February 2026 report noted that some prospective projects did not sign binding contracts once financial commitments were required. That is useful information. It tells planners that not every request represents a financially committed customer. There is no principled reason why homeowners, shopkeepers, or smaller industrial users should underwrite capacity reservations that developers themselves would not pay to preserve.

I would begin with an explicit distinction between the electric-service agreement and the economic-development announcement. Local officials love to calculate construction jobs, tax assessments, and future prestige; I have read enough inflated projections in zoning files to recognize the genre. Such estimates may be offered in good faith, but they are not enforceable promises. The electric contract should address maximum demand, ramp-up schedules, metering, backup requirements, interruptions, minimum payments, connection costs, security for performance, and procedures when the project grows beyond its original plan. Where the system needs new substations or transmission lines, someone must state how the cost is allocated. Where power remains dependent on outside providers, a claim of self-supply should be examined rather than accepted at face value.

The obligation to existing customers is particularly important. Residential ratepayers generally cannot decide to take their business elsewhere when the utility makes an expensive mistake. Nor should they be expected to hire counsel and participate in every rate proceeding simply to prevent an avoidable transfer of risk. State commissions exist in part because utilities occupy a special position in the economy. Their decisions should make cost causation comprehensible to people who are not electrical engineers. This does not mean every cent of a network investment must be assigned to one corporation. Infrastructure often serves multiple customers and creates lasting regional benefits. It does mean that commissioners should identify those shared benefits honestly, calculate the incremental costs, and explain why a particular allocation is fair.

I also worry about the physical siting process. A developer may seek a quiet tract of land for its computing campus while the associated transformer yards, transmission corridors, and gas infrastructure cross other properties. Landowners are entitled to notice and a meaningful chance to challenge factual assumptions. Emergency generation raises noise and emissions questions; cooling facilities raise water and land-use questions; and tax abatements may reduce revenue that communities expected to receive from industrial growth. None of this justifies using zoning law to veto every productive enterprise. It calls for a disciplined record demonstrating why the project belongs in that location and what conditions will protect neighboring residents. A project that cannot withstand ordinary public scrutiny is poorly prepared for a twenty-year operating life.

The recent nuclear and fuel-cell agreements are constructive precisely because they begin to answer the supply question. Google and Constellation’s October agreement aims to unlock 890 megawatts of additional reliable nuclear output through investment in existing units. Another plan involves a 702-megawatt fuel-cell installation near Cheyenne and a long-term Google purchase commitment. Yet an announced agreement has to be translated into construction and operating documents. If the promised megawatts are delayed, what happens to the customer’s connection date? Does the utility supply interim power from its existing system? Is some load interruptible? Are the costs of procuring replacement capacity assigned to the corporation that promised to bring new supply? The answers ought to appear in contracts, not in a later press release expressing disappointment.

That analysis also applies to a developer’s claim to operate behind the meter. A building can have private generators and still depend heavily on the grid for frequency support, emergency backup, and power quality. If the utility must stand ready to supply the entire campus when the private generators fail, that capacity is a real service for which other customers should not quietly pay. Conversely, a site that can genuinely separate itself from the grid during a system emergency may provide a valuable benefit. The engineering and the contractual obligations should agree with each other. One of the more frustrating forms of regulation is a system in which the paperwork describes a resource that no longer corresponds to the physical facility.

There is an important due-process question in emergency curtailment. Most citizens would understand that a hospital cannot be treated exactly like a facility training an optional software model during a severe shortage. But a grid operator cannot make its priorities on the fly based on which businesses have the least popular public image. Some data centers host emergency communications, financial systems, medical records, and services that other industries cannot safely lose. Others carry workloads that could be postponed. A sensible tariff could distinguish those services by proven technical capabilities and agreed-upon obligations, not by calling every server equally essential or equally disposable. The system should define ahead of time what can be interrupted, for how long, with what notice, and under what compensation provisions.

I am not persuaded by arguments that technological efficiency will resolve every concern. More capable processors, more efficient cooling, and better scheduling may reduce the energy cost of a particular computing task. That is worthwhile. But a reduced unit cost frequently expands use. Utilities therefore need measured load profiles, not assurances that a future generation of chips will miraculously make a planned campus unnecessary. Similarly, energy certificates and annual renewable-energy purchases cannot establish that a facility will have adequate supply at the exact hours it runs. The distinction between annual energy and instantaneous capacity ought to be mandatory reading for policymakers asked to approve large tax breaks in the name of power security.

The regulatory process itself needs reform, provided reform means removing needless delay rather than dispensing with standards. A utility should be able to identify likely service dates without leaving applicants in limbo for years. Agencies should coordinate siting and transmission reviews instead of requiring the same factual record to be assembled repeatedly. A developer should receive clear instructions and an appeal process when applications are rejected or conditions imposed. At the same time, regulators must be willing to say no when a project would impose unacceptable risks on the system. A deadline that cannot be met safely is not progress. It is a deferred admission of failure.

I would go further and require periodic disclosures of committed versus speculative large loads. Much of the current public debate treats every gigawatt in a development queue as equivalent. It is not. A land option, a planning inquiry, a deposit, a completed interconnection study, a signed power contract, and a commissioned facility represent different levels of commitment. The public should be able to see those categories in aggregated form without revealing proprietary details. Such reports would help utilities avoid building too little, and they would protect customers from financing infrastructure for fantasies. Transparency is not an inconvenience to investors; credible numbers reduce the risk that an entire regional plan rests on inflated expectations.

Long Island has ample experience with electricity costs, infrastructure disputes, and the politics of keeping densely populated communities supplied. People here understand that power is both indispensable and expensive. They also understand that a rate increase can become unavoidable long before a customer learns how the underlying decision was made. That is why I distrust rosy comparisons between the tax revenue from a data center and the cost of its connection. They often omit the long-term consequences of network upgrades, generation contracts, stranded assets, and emergency procedures. A proper analysis should examine plausible low- and high-demand outcomes, including the case in which the expected technology boom stalls. An attractive investment under one optimistic forecast may be a burden under another.

Some commentators would prefer a sweeping prohibition on new data centers until the grid catches up. Others would have governments order utilities to connect them immediately in the name of national competition. Both approaches are too blunt. Large computing facilities can provide economic value and support services people genuinely need. New industrial demand can also justify valuable power infrastructure that might otherwise be delayed. But a right to compete in an industry is not a right to transfer one’s costs to neighbors. The proper mechanism is a disciplined bargain: measurable commitments by developers, transparent duties for utilities, independently reviewed reliability criteria, and fair treatment of established customers.

The most revealing test may be what happens after the announcement. Does the company post collateral? Does the promised plant deliver energy? Does the utility report construction costs clearly? Does a state commission revise its forecast when a project is canceled? Does the grid operator exercise its emergency plan before a crisis makes experimentation dangerous? These are mundane questions, but they are the ones that turn infrastructure policy into something that works. I would rather read an unglamorous service agreement that will survive twenty years than a magnificent speech about intelligence followed by a blackout notice. In the end the grid will operate according to physics; whether the cost and responsibility are assigned fairly is a question for law and public institutions. Those institutions should do their job before the lights make the argument for them.

Cedar Rapids, Iowa Correspondent

I used to close the books for a plant. The work teaches a person several habits that do not go out of fashion. A purchase order is not a delivery. A budget is not a result. An estimate is not an obligation, unless somebody has signed the right contract. And a machine that produces revenue for a decade should not be purchased on the assumption that one very good quarter will continue forever. These are simple propositions, which is probably why they disappear from large public discussions. We now have technology companies talking about electricity in the hundreds of megawatts and utilities discussing multibillion-dollar investment programs. My first reaction is not hostility to artificial intelligence. It is to ask for a cash-flow schedule, a credible load forecast, and the name of the party responsible if the business case falls apart.

The numbers are large enough to demand serious treatment. The June 2026 Lawrence Berkeley National Laboratory update put the possible data-center share of U.S. electricity consumption at 9.5 to 15.3 percent by 2030, with an estimate near 11.8 percent. The Electric Power Research Institute has published its own scenarios, suggesting a share of 9 to 17 percent. Both organizations make clear that forecasts depend on chip demand, server efficiency, utilization, and the pace at which facilities actually open. I would not average the estimates and declare that we have discovered an exact answer. I would ask what investments are justified under the low case, what must be reserved for the high case, and which decisions are irreversible once a utility borrows money against an anticipated stream of revenues.

There are two kinds of costs people like to mix together. Electricity generated and consumed this year carries fuel, operating, maintenance, and other expenses. A transmission line or substation is a capital asset that may sit on the utility’s books for decades. A company requesting service can create a need for both. Its electricity purchases may help finance them, if the company stays. But if it abandons a campus or uses half the promised power, the fixed network costs do not disappear. Somebody remains responsible for debt service and depreciation. Ratepayers who had no vote in the project may be asked to make up the difference. The proper question is whether the large customer has made commitments strong enough to carry the special costs created on its behalf.

That is why I like the idea behind Ohio’s data-center tariff. It requires developers seeking very large quantities of power to accept meaningful payment obligations rather than merely placing ambitious numbers in a planning queue. The company reports that applications converted into binding contracts changed the picture of demand. It is no surprise that speculative projects look less certain when someone asks for collateral or minimum charges. I have seen this in ordinary manufacturing: everybody wants capacity when a sales forecast is free; fewer want to pay a supplier to reserve production slots. A utility should not be embarrassed to discover that a portion of projected demand was speculative. It should be embarrassed if it issued debt before discovering that fact.

A data center can still be a good customer. Some facilities operate with substantial around-the-clock demand, supporting predictable revenues that help spread network costs over more kilowatt-hours. A long-term power-purchase agreement can give a generator confidence to invest. Local construction contractors may have years of work. Taxable property may strengthen a county budget if incentives do not consume the benefits. In a town that has watched employers close, I understand the attraction. But an economic-development brochure should never be substituted for a customer-specific cost-of-service study. When a council member says the project will lower everybody’s bills, I want to see the calculation, its assumptions, and a scenario in which the customer uses less power than promised.

A plant accountant knows the difference between a maintenance expense and an expansion. If an existing generator needs equipment to keep it operating safely, that is one decision. If it receives upgrades that create measurable additional output, there may be a genuine expansion of supply. Google’s October 2026 agreement with Constellation includes a plan for equipment and technology investments at existing nuclear units intended to provide 890 megawatts of incremental capacity. That is encouraging, provided the added capacity is real, metered, and available when needed. We should be careful not to count the same electricity twice: once as existing supply on the grid and again as new supply because a technology company bought a claim to it. On a ledger that would be called double counting. On a grid it can become more than an accounting error.

The other overlooked line item is time. An AI facility can be constructed in a couple of years, while new transmission corridors, heavy transformers, generating plants, and permits can take longer. The utility must start spending before it can sell all the promised energy. It may pay escalating prices for equipment because dozens of other companies have placed similar orders. Interest accrues during construction. Environmental and land-use proceedings can delay schedules. A company demanding a fixed connection date might be justified in complaining about bureaucratic delay, but none of these constraints vanishes because the data center has a valuable tenant. Financing terms need to recognize the difference between expected completion and proven delivery.

PJM’s capacity auction is an accountant’s warning light. In July 2026 it fell 6,831 megawatts short of its stated reliability requirement for a future delivery year. The result does not show that there will be a blackout on a particular afternoon. It shows that projected demand, available capacity, and the auction’s economic terms are not aligning sufficiently. The price can rise and still fail to attract the required quantity if investors cannot build resources in time or do not believe they will recover costs. The answer may require changes to permitting, contracts, market design, and resource planning. The dangerous response is to paper over the shortfall while announcing another development park, as though counting a customer were equivalent to securing a generator.

I worry about public finance on two fronts. First, communities compete with each other by offering abatements, infrastructure grants, and favorable treatment to projects that might have been built somewhere in the region anyway. Second, the utility can recover costs through rates instead of the local tax budget, making the subsidy difficult for citizens to recognize. A county may proudly report a new employer while households across several counties pay higher electricity charges to support the necessary network. No single local official has to account for the full transaction. That is poor governance. If a project is worth subsidizing, the subsidy should be calculated openly and weighed against alternatives such as industrial-site improvements, schools, and repairs to aging water or road systems.

Artificial intelligence itself may be economically transformative. I do not dismiss productivity improvements because a few overexcited promoters are selling fantasies. Manufacturers can use machine-learning systems to improve inspections, schedules, and equipment utilization. Farmers already work with data-rich tools. Businesses rely on cloud services whose continuous operation matters. Nevertheless, the value of those uses is not the same as the profitability of every data-center operator. The market will sort winners from losers, sometimes after lenders have made unfortunate decisions. The public electricity system should be able to survive those failures. It should not depend on the assumption that every fashionable computing project will enjoy permanent high margins.

The financing of private generation deserves close reading too. Suppose a developer builds its own gas plant behind a data-center meter. The proposal sounds as though it eliminates a burden on other ratepayers. But if that plant needs grid backup when a turbine fails, then someone must reserve capacity for the emergency. If its gas supply requires a new pipeline, somebody must finance that. If the plant produces surplus electricity, the accounting depends on whether it can actually export power and whether the network can accept it. The arrangement may be excellent. It may also shift risks without admitting them. A balance sheet that excludes contingent obligations because they are inconvenient does not become sound by changing the project name from utility-scale to private.

One practical solution is to separate committed load into service classes with different prices and guarantees. A facility with critical real-time services could pay a premium for firm electricity and verified redundancy. A facility performing work that can be postponed could receive a discount for agreeing to curtail during defined emergencies, with penalties if it fails to comply. Such arrangements do not make all computational loads flexible, and the system operator must test the actual capabilities. But they create a financial incentive to use electricity when the system has it available. A manufacturer understands a similar bargain when a supplier offers a lower rate for nonpriority orders. The customer chooses the service level, and the price reflects the resources required to provide it.

There is a further danger in the way companies report efficiency. The electricity used for a single AI calculation may fall sharply as chips, models, and software improve. That does not mean total electricity demand will fall if cheap computation attracts far more customers. The plant analogy is straightforward. If a machine becomes twice as efficient and output triples, total energy use may still rise. Nor are all efficiency figures comparable. A server’s electrical consumption, a facility’s power usage effectiveness, the load from cooling, and the electricity drawn during peak grid hours answer different questions. I would be suspicious of any investor presentation that boasts about a percentage efficiency improvement but omits the absolute megawatt requirement for a fully occupied campus.

The workforce problem is another hidden liability. Utilities need people who can design and operate increasingly complex substations, install high-voltage equipment, and maintain generation plants. Those skills cannot be ordered by clicking on a software subscription. If a developer offers extraordinary wages for construction electricians during one peak building season, smaller manufacturers and local utilities may find it harder to fill vacancies. That is not a reason to prevent competition for workers. It is a cost that belongs in regional planning, especially where community colleges and apprenticeship programs need time to increase training capacity. A durable expansion pays attention to the labor market rather than assuming the available workforce will multiply by proclamation.

In an industrial region like eastern Iowa, the arrival of a large power customer could support investment that benefits everyone. It could create markets for machinery, electrical services, and maintenance contractors. It might even make it more economical to modernize a regional network that has been awaiting investment. I would welcome those possibilities. But the test is incremental benefit after incremental cost, not the number of construction workers photographed in reflective vests. We should compare what the system would have cost without the data center with what it will cost after the project, allocate the difference sensibly, and account for the risk that the business will not develop as planned. If the numbers remain attractive, build it.

I have watched industries become unfashionable and then necessary again. Power generation is receiving renewed attention partly because AI companies suddenly need it, but a manufacturing economy has always needed reliable electricity. It would be a mistake to build an electrical system optimized only for one class of computer customer at the expense of factories, farms, hospitals, and homes. The same transformer yards and transmission corridors should be evaluated for their broader usefulness. A generating investment that remains productive even if one data-center project fails is better than an asset whose value depends entirely on a single tenant’s ambition. Diversification is not glamorous, but accountants appreciate what it does in a bad year.

My conclusion is neither that the boom is a mirage nor that it is a miracle. Utilities and technology firms are making material progress where they sign binding power contracts, finance physical generation, upgrade existing plants, and make their demand forecasts auditable. They are not making enough progress when the public hears only about projects that have not cleared basic engineering or financing hurdles. My recommendation is to put every major proposal through an ordinary industrial investment test: credible demand, enforceable customer obligations, stress-tested construction costs, prudent financing, adequate reserves, and a sensible exit case. Then make the results public in plain language. Technology fortunes may turn quickly. Transmission bonds and household electric bills tend to remain long after the excitement has moved somewhere else.

Las Vegas, Nevada Correspondent

I manage people and vendors in a business where a power failure is not an interesting chart. It is a roomful of customers wondering what has happened, employees improvising, equipment shutting down, and supervisors calling every number they have. Las Vegas runs on the assumption that the lights will work, the air conditioning will work, the payment systems will work, and tomorrow’s shift can arrive and do it again. The fact that so much of our economy looks frivolous to people elsewhere does not make its physical dependencies any less serious. A new data-center campus has the same problem on a different scale. Somebody is selling a service that customers expect to be available, and that service exists only if a chain of people and machines performs properly. The AI electricity conversation would be better if it contained more operating plans and fewer victory announcements.

The numbers are no longer too small to dismiss. Berkeley Lab’s mid-2026 update placed the possible U.S. data-center share of electricity consumption between 9.5 and 15.3 percent by 2030. EPRI’s scenarios are also high and emphasize how concentrated development can become in particular states, including Nevada. I care less about the national average than about the neighborhood substation that must survive our heat. A computer campus may run equipment throughout the night when some other businesses are closed, but Nevada’s grid faces serious summer demands from cooling and other uses. The operational question is not simply how many kilowatt-hours a building buys in a year. It is whether enough electricity can reach that building during a brutal afternoon without threatening service for people who have no option to shut down their homes or businesses.

People sometimes talk as if data centers can be located anywhere there is cheap land. That is the first mistake. Land is only one line in the decision. Operators need access to transmission, workable interconnection dates, cooling arrangements, skilled contractors, security, and often a network of fiber routes. Water conditions matter in a dry region, as do the choices among air cooling, liquid cooling, recirculation, and other designs. Those systems have different electrical and water demands; no single cooling statistic describes every center. A developer can advertise an efficient design, but the real-world result depends on heat, equipment density, and operations. Our city has decades of experience managing buildings in conditions that expose fanciful engineering. A desert summer is unimpressed by a promising rendering.

One reason the debate is confused is the use of the word ‘power’ for several different products. A megawatt of nameplate capacity is a rating. A megawatt-hour is energy over time. A long-term contract may secure financial rights to output but not necessarily a new cable to the right location. Battery storage can deliver power quickly but does not create energy out of nothing; it needs to be charged. Solar production can be substantial in Nevada, but the hours of highest output are not identical to every hour of business demand. Natural gas and nuclear generators can provide valuable firm supply, but they have fuel, maintenance, outage, and construction schedules to manage. An operator needs to know which of these resources will be available when the system is strained, not just how impressive the combined total sounds at an investor presentation.

I am encouraged by technology firms signing long-duration energy deals. A utility or generator is more willing to invest if a creditworthy buyer guarantees a market for its output. Google’s agreement with Constellation, announced this month, is meant to support added nuclear output at existing plants. American Electric Power has announced an ambitious fuel-cell project near Cheyenne. Each effort addresses part of the real issue: power infrastructure requires commitment ahead of delivery. But anyone who has managed a large vendor contract knows that signing and fulfillment are separate events. An equipment delivery can slip, labor can be scarce, and commissioning may uncover problems. The relevant document is the contingency plan. What happens to the data center if electricity arrives six months late, and what happens to the grid if the campus opens earlier than its dedicated supply?

A data center is also not one uniform block of demand. It has life-safety systems, cooling, communications, backup supplies, and computing tasks with different priorities. Some activities may be capable of shifting to another time or location. Others support services whose interruption would cause immediate damage. From an operations standpoint, this suggests an obvious bargain: pay more for truly firm service; receive a lower price for a portion of load that can be curtailed on agreed terms; test the curtailment procedure; and document what remains noninterruptible. What I reject is the notion that a firm should receive a discount for theoretical flexibility while refusing to reduce load when the system is under strain. We have all encountered vendors whose promises disappear when the emergency arrives. The grid cannot afford that style of contract.

There is a worthwhile comparison to running a large hospitality facility. We do not assume that the emergency generator can power every piece of equipment just because it can light the evacuation signs. We decide which services are critical, which can be paused, and which must be abandoned to keep people safe. Staff rehearse procedures. Fuel arrangements are verified. If a vendor claims a system has redundancy, someone asks what common failure could disable both paths. Large computing operators and utilities should have the same discipline. A computer cluster with backup generators may still depend on a common transformer yard, fuel supplier, or cooling loop. Redundancy is only meaningful after the common points of failure have been identified and tested.

I suspect that the most immediate bottleneck is the mismatch between construction calendars. A tech developer may think in eighteen- or twenty-four-month increments. A transmission project can require a much longer sequence of engineering, permitting, materials, and construction. Skilled crews are not infinitely available, and some electrical equipment has long lead times. The resulting impatience can create pressure to authorize temporary arrangements, such as diesel generators or gas turbines, that were never intended to support an industrial district permanently. A short-term bridge may be justified in special circumstances. It must be costed, regulated, and time-limited honestly. We have enough experience with temporary measures that become permanent once the ribbon has been cut.

Las Vegas is familiar with extraordinary capital investment and difficult financial cycles. Visitors see impressive new buildings; operators see the debt payments that follow. A data-center buildout can create a similar problem for utility planners. The grid upgrades may be financed over decades, but an AI company’s expected growth can change in a quarter. Sometimes a canceled project frees capacity that another buyer can use. Sometimes a substation has been built in the wrong place for that to happen quickly. It is fair to require deposits, long-term minimum payments, and sensible termination provisions when a customer’s requests force expensive infrastructure. That is not discrimination against technology. It is ordinary risk management of the kind a smaller business is expected to practice every day.

There is another cost that frequently gets buried in municipal enthusiasm: who uses the scarce construction capacity while these projects are going up? Electricians, mechanical contractors, trucking operators, steel suppliers, and permitting personnel are already busy. If a large company pays a premium to accelerate its campus, local hotels, shops, apartment buildings, and public facilities may face higher prices or slower service. The labor market will adjust, but not overnight. Public officials should weigh these effects when offering special incentives or promising extraordinary approval speeds. It is possible to welcome a data center and still insist that the town’s water facility and emergency services receive the professional attention they need.

Water is especially worth discussing in the West, though it must not be reduced to a frightening but meaningless number. A cooling design using evaporative water has a different consumption profile from a largely air-cooled installation. The choice may affect electricity demand, equipment cost, and operational efficiency in complicated ways. A city must understand whether a proposed project uses potable supplies, reclaimed water, or other sources, and how demand changes during extreme heat. Some cooling technologies save water at the price of increased electrical load, which matters when the reason for choosing the site was supposedly low-cost power. A serious permit application should identify these trade-offs without turning them into a marketing contest over whose technology deserves the greenest label.

Reliability has an ethical dimension that is easy to miss because electricity sounds like a technical commodity. In Nevada heat, a prolonged outage can be dangerous, particularly for older residents, people with illnesses, and households unable to afford backup solutions. A highly financed computing campus may have redundant supplies while neighboring businesses have no such protection. If a region’s planning decisions improve the campus’s resilience at the expense of the public system’s operating margins, that is a problem. I am not asking anyone to rank artificial intelligence as worthless. I am asking utilities to preserve the basic service duty that existed before the investment boom arrived.

Of course, AI can bring benefits to hospitality and other services. Better demand forecasts, staffing systems, property maintenance, and customer assistance are not fiction. A more capable technology sector may become one of the foundations of the American economy. I would be foolish to insist that no new electricity should be allocated to it. But the sector’s economic contribution is still subject to cost and supply constraints. A technology that makes a company more productive should ideally help pay for the infrastructure it requires. The assertion that an industry is important does not answer whether its most speculative projects are entitled to discounted grid connections, emergency backup, and tax relief at public expense.

The October 2026 debate around PJM is a warning to everyone. Its capacity auction fell short of the stated requirement, and regulators rejected a proposed emergency curtailment mechanism while accepting a system to gather better information about large data-center loads. I understand the frustration of operators who want a clear plan. The answer is to write a lawful, predictable plan now and negotiate commitments while customers still have choices. Waiting until demand exceeds supply to decide whose equipment shuts off would be reckless. So would promising uninterrupted service at prices too low to finance the resources needed to provide it. Markets do not eliminate the need for standards; they work better when the standards are clear before commitments are made.

I would evaluate an AI data-center proposal much as I evaluate a major operations contract. Is the customer creditworthy? Is the timetable realistic? Which dependencies could cause a shutdown? What happens if an equipment supplier fails? Is the emergency plan practiced rather than printed? Are there measurable service levels and penalties? Are the costs recoverable even under an unfavorable demand scenario? None of these questions is glamorous. Their answers determine whether a fine building becomes a reliable business. Utilities and policymakers ought to ask the same questions instead of using the technology’s novelty as a reason to waive the ordinary rules.

There is also a distinction between the consequences of a normal business failure and the consequences of a utility planning failure. A hotel vendor can go bankrupt, and everyone involved may lose money; the next operator does not automatically have to finance the entire failed venture. A utility has obligations to serve a territory, and an abandoned investment can remain in its rate base for a very long time. A power-plant or transmission decision therefore needs protections that would not be necessary for an ordinary private building. Regulators should not set impossible demands that prevent useful projects from happening, but they should require enough deposits and committed revenues that the ordinary customer does not become the involuntary guarantor of a technology speculation. This is especially important where the developer has created multiple prospective sites and may eventually build only one of them. The utility needs to know whether it is being asked to build capacity for an actual customer or hold a costly option for an uncertain business plan.

The habit I would like to see become normal is an annual public comparison between what each region forecast and what it actually delivered. How much data-center demand appeared, how much power was added, how many network upgrades were commissioned, and how often did operators request emergency support? If a forecast was wrong, why was it wrong? Was the problem overly ambitious developers, delayed equipment, or an unrealistic permit schedule? There is no shame in revising a forecast on the basis of evidence. There is plenty of shame in allowing the same bad assumptions to support new borrowing every year because nobody wants to interrupt a good news cycle.

If this expansion is handled well, the country will gain more than computing horsepower. It will have upgraded generating stations, stronger transmission networks, better load forecasting, new skilled trades, and an electrical system able to support many industries. If it is handled poorly, the public will remember the higher bills and outages long after the names of today’s fashionable AI projects have changed. I have spent too many late nights helping people keep complex facilities running to confuse an impressive opening with durable performance. The lights are part of the product. They have to work every night, and somebody has to pay honestly for keeping them on.

Knoxville, Tennessee Correspondent

When somebody proposes a major public-works project in my part of Tennessee, I begin with the drawing that nobody wants to show at the groundbreaking: the plan for how the pieces connect. The new road might be magnificent but useless if it ends at an old intersection that cannot carry the traffic. A wastewater facility may have excellent treatment equipment and still fail because the pipe delivering sewage is undersized. Electricity presents the same problem at a larger and less forgiving scale. A generation announcement does not automatically supply a data center. Power must travel through equipment with physical limits, meet stability requirements, and remain available when another component fails. The AI boom has made those mundane design constraints a national economic question, but it has not rewritten the underlying engineering.

The forecasts deserve attention. Berkeley Lab’s 2026 update estimates that data centers might consume between 9.5 and 15.3 percent of U.S. electricity in 2030. NERC’s long-term reliability assessment identifies fast-growing large loads as a significant source of stress for regional systems, in part because resource development has trouble matching projected demand. These figures should not become theatrical props. They describe scenarios that require planning under uncertainty. A designer does not respond to an uncertain future flood by pretending no flood will occur, nor by constructing the most expensive imaginable dam without further analysis. We define the plausible range, evaluate consequences, and design a system that meets an explicit standard with an acceptable margin for error.

Electricity planners have several margins to consider. A region requires enough energy over the course of the year, enough generating capacity at the hours of highest demand, and enough transmission and distribution capability to deliver that power. It must account for generators under maintenance, extreme temperatures, unforeseen equipment failures, and the behavior of large customers during disturbances. Different resources contribute differently. A solar facility may produce enormous amounts of annual energy but less during a dark winter peak. A gas-fired generator may be highly useful yet depend on a pipeline operating through the same weather event. Batteries can provide rapid services and move energy across hours, but their stored energy is finite. The engineering task is to evaluate the entire portfolio, not to compare one technology’s most flattering characteristic with another’s weakest point.

A 300-megawatt data center is not merely an industrial building with a large bill. Its demand can exceed that of many existing facilities, and it may require a dedicated substation or transmission upgrade. Its internal electronics can behave differently from ordinary motors and resistive loads, especially during sudden disconnection or restoration. Computer loads, power converters, backup systems, and control software have to operate consistently with the bulk power system. NERC’s work on computational-load standards, advanced in September 2026, recognizes that issue. I welcome the effort, provided standards are written clearly and tested against actual equipment behavior. A new load is not reliable merely because the owner has installed redundant computers; the way the facility interacts with voltage and frequency matters to every other user connected to the network.

Consider what happens when a utility studies an application. The developer gives an estimated peak demand and desired service date. Engineers examine existing lines, transformers, protection systems, and generation adequacy. They model the effects of faults and outages. They identify improvements needed to connect the customer without degrading service for others. If the developer submits multiple inquiries in different jurisdictions to keep its options open, the total apparent demand may exceed the capacity the developer ever intends to build. Yet dismissing all proposed demand as speculative is equally dangerous. Planning requires transparent stages: a preliminary inquiry, an advanced study, a signed commitment, construction, and commissioning. Each stage should increase confidence in the forecast and trigger an appropriate level of infrastructure investment.

This is one reason Ohio’s special data-center tariff is interesting. It makes large applicants put money behind reserved capacity. The utility’s February filing showed that requiring binding commitments helped identify which projects were sufficiently serious to proceed. That does not eliminate uncertainty, and a contract cannot guarantee that a company stays solvent. It does create a better information system than treating every speculative inquiry as a firm load. In civil engineering we would not design a bridge for five different proposed developments and then neglect to ask whether those developments represent competing plans for the same investment. Utilities need a disciplined way of filtering requests so that a pipeline of ideas does not become a fictitious pipeline of electrical demand.

The physical bottlenecks differ by region. In some places generation is scarce; in others a transmission line or transformer is the limiting factor. A site might have a nearby high-voltage corridor yet lack the available capacity or switching arrangements to connect safely. The existence of a line on a map does not make service instantaneous. Permits and right-of-way acquisitions add complexity, especially where a project crosses farms or neighborhoods. A developer can choose a rural location partly because the land is inexpensive, then discover that the most expensive part of the project is getting sufficient electricity to it. These considerations need to be settled before local government grants incentives premised on an opening date nobody has engineered.

I am not opposed to accelerating permits. I have worked in systems where procedural duplication delayed sensible construction without improving public safety. When the same engineering issue is reviewed by three agencies under slightly different forms, consolidating the process can be beneficial. But there is a difference between removing needless repetition and erasing substantive review. A transmission project affects landowners and ecological resources. A gas plant requires fuel and emissions analysis. A substation must protect workers and neighboring properties. Regulators should set clear decision deadlines while preserving the studies that make the decision defensible. An approval that collapses during construction is not fast. It is merely late in a more expensive way.

The demand for ‘bring your own power’ comes from a sensible instinct. If a new industrial customer needs an extraordinary amount of supply, asking it to finance additional generation can protect existing customers and accelerate construction. Google’s agreement with Constellation to finance upgrades that provide 890 megawatts of incremental nuclear output is an encouraging example of attaching money to physical capability. But firm electrical supply is a system service. Even a private generator may require an interconnection, backup arrangements, and grid support. If its output drops suddenly, the remainder of the network must accommodate the imbalance unless the associated computing load is disconnected at the same time. The technical and contractual conditions must be designed together. Behind the meter is a billing relationship, not a waiver of electrical physics.

An emergency policy is another place where engineering and law meet. PJM’s July 2026 capacity auction produced a 6,831-megawatt shortfall against its reliability requirement. In October, federal regulators accepted a registry for tracking data-center load while rejecting a particular curtailment proposal. The registry should improve information about when facilities will operate and how they can behave. But a registry is not a replacement for a workable operating plan. I would prefer data-center contracts that explicitly differentiate critical from noncritical computing and describe what portions of load can be reduced during a system emergency. Control systems should be tested so that curtailment does not create a larger reliability problem through an uncontrolled shutdown.

That last point is important. A sudden reduction in a massive electronic load is not automatically harmless; system operators need to understand how the network responds. Likewise, a wave of simultaneous restarts after an outage can stress equipment that has already endured abnormal conditions. A facility should have a sequence for restoring cooling, backup arrangements, and servers. If hundreds of megawatts reconnect abruptly, frequency and voltage excursions may affect other customers. Reliability standards should therefore address behavior during disturbances as well as ordinary peak demand. It is a less visible aspect of the AI energy debate, but it is exactly the kind of detail that distinguishes a robust system from one operating too close to its limits.

Cooling and water systems are equally practical matters. In eastern Tennessee, where water is more familiar than in the Southwest, it may be tempting to ignore the environmental cost of cooling. That would be foolish. Industrial water demand can affect treatment systems, withdrawals, and receiving waters even in relatively wet regions. Different cooling designs trade water consumption against energy requirements, equipment expense, and local conditions. Reusing treated water may be feasible in some locations but require additional pipelines and treatment elsewhere. The right decision is site-specific. An environmental review should discuss both water and electricity, because a system advertised as water-saving can sometimes increase the electrical load that brought the developer to the region in the first place.

Some people expect more efficient chips to make these concerns temporary. I hope technological progress reduces energy required per unit of useful work. We should measure and reward it. But a cheaper computation can attract more users, and demand forecasts must consider that response. Engineers also have to look at the complete facility, not merely the advertised processor. Power conversion, cooling pumps, fans, networking equipment, redundancy, and battery systems consume electricity. Performance claims should include realistic operating conditions and explain whether a facility is expected to ramp up gradually or reach its maximum demand shortly after opening. A shiny efficiency number from a laboratory does not establish the load profile of a completed campus.

Utilities need practical construction capacity as well as financing. Transformer manufacturing, switchgear, high-voltage cable, skilled electricians, and control engineers are limiting resources. New power plants may require turbine deliveries and pipeline work. If many developers assume that the same suppliers will meet accelerated schedules, the timetable becomes impossible in aggregate. The result may be rising prices, delayed projects, and stranded investments. A competent regional plan should identify these supply-chain constraints and publish an honest sequence of upgrades. I would rather see a realistic six-year program that gets the job done than a two-year announcement followed by four years of excuses.

Local governments have responsibilities too. The promised property-tax base may be substantial, but incentives, roads, water infrastructure, and public safety costs reduce the net benefit. A jurisdiction may gain revenue while the utility’s wider service territory absorbs electrical costs. Local officials should therefore coordinate with state commissions and regional grid operators before treating a campus as a sure economic victory. Public meetings should discuss the trade-offs in plain terms rather than reducing the controversy to a contest between technological progress and environmental obstruction. People living near new substations are entitled to understand why those facilities are necessary, how they will be financed, and how reliability will be protected.

Our region has substantial experience with major power infrastructure through the Tennessee Valley Authority and its interconnected network. That history does not give us immunity from future bottlenecks. It teaches that the benefits of a robust power system are broader than any one customer. Dependable electricity supports manufacturing, hospitals, universities, logistics, and homes. If the AI boom prompts upgrades that strengthen the region for decades, that will be an important public gain. If policymakers let one highly visible industrial class monopolize planning attention, they may weaken the broader foundation of economic development. Electricity systems should be designed for resilience under multiple plausible futures, not for one fashionable industry’s most optimistic forecast.

It would help to evaluate projects using a consistent set of engineering milestones that the public can understand. A site becomes more credible as it moves from a preliminary load inquiry to a studied connection, executed service agreement, completed right-of-way approvals, committed generating resources, and commissioned equipment. Those are not interchangeable milestones. Officials should show which have been met rather than putting all projects into one inventory of promised megawatts. That distinction allows construction schedules to be synchronized sensibly and gives regulators an early warning when work on one piece is falling behind the rest.

If I were advising a county on a large proposed campus, I would ask whether the surrounding infrastructure is being designed for a useful life beyond the initial tenant. Can a transmission improvement support other industrial customers if demand changes? Can access roads accommodate emergency vehicles and future employers? Are the water and cooling systems compatible with the region’s long-term plans? What arrangements remain after a developer sells its project? The more of those questions answered up front, the less likely local taxpayers are to inherit an expensive, highly specialized facility nobody can readily reuse.

My position is conditional but not timid. Build new data centers where the engineering, customer commitments, and public benefit justify them. Support genuine new generation, including firm resources suited to regional needs. Expand transmission capacity and simplify redundant permitting. Require large computational loads to disclose the operating characteristics that matter for grid stability, and arrange flexibility through lawful, enforceable agreements. Most importantly, report progress in commissioned megawatts, available reserves, and demonstrated performance rather than in contracts and announcements alone. The engineering drawings eventually meet the ground. On that day, no one will care whether the press release was inspiring. They will care whether the system works.

Wheeling, West Virginia Correspondent

The country has found a new industry that promises to change everything, and the first thing it wants is electricity. That makes me smile, though not entirely with amusement. For years we were told that the future would be built from knowledge and software, not from the dirty work of digging, generating, welding, and maintaining. Now the companies selling the future are scrambling for turbines, substations, copper, and the people who know how to keep a power station operating. It turns out that bits need atoms, and that the smartest machine cannot negotiate with a dead transmission line. I am not inclined to turn this into a feud between technology workers and industrial towns. I would simply like the advocates of the next revolution to acknowledge the workers and communities they have suddenly rediscovered.

The projected electricity use is serious. The June 2026 Berkeley Lab report placed data centers’ potential share of national electricity consumption in 2030 between 9.5 and 15.3 percent. The International Energy Agency’s global outlook anticipates a substantial increase as well. That does not mean every projected facility will materialize. Anyone who has watched an industrial boom knows that proposed plants outnumber operating plants. But the forecasts are large enough that utilities must prepare. The trouble begins when planning for genuine growth becomes a license to charge ordinary customers for speculative projects, or when governments use the excitement of a new industry to forget the existing workers and industries whose reliability has already been under strain.

In the Ohio Valley we do not talk about electric service only as a convenience. It is what keeps a machine shop open, preserves a shift, and makes it possible to recruit a manufacturer who has a choice of locations. A factory does not care how elegantly a utility explains a power interruption if that interruption destroys a production run. When an older plant closes, the official reasons tend to be summarized in a paragraph: global competition, modernization, changing markets. The effect on a community lasts decades. If new data centers are permitted to consume scarce capacity before dependable new supplies are secured, the region may discover that the glamorous customer has displaced the boring one that actually employed a lot of people. We should not repeat that mistake simply because the new customer’s executives know how to use better vocabulary.

There is genuine potential for shared prosperity. Increased demand can justify keeping a viable generating station open, modernizing transmission, and training new electrical workers. Long-term contracts between technology companies and power suppliers may help finance projects that otherwise would remain on paper. Google’s recent agreement with Constellation to support upgrades adding 890 megawatts of nuclear capacity is the kind of proposal worth studying. So are plans to restart existing nuclear stations, provided independent safety requirements remain intact. The relevant question is whether these ventures create reliable additional output, not just claim an existing plant’s electricity under a new buyer’s name. A legal agreement can change ownership of megawatt-hours without changing the number available on a bitter winter evening.

A less romantic subject is the workforce. You do not build a generating station with an announcement. You need electricians, pipefitters, welders, machinists, instrument technicians, civil crews, inspectors, and people who can troubleshoot systems that never behave exactly as they did on the drawing. When national policymakers talk about energy independence, they often skip this part because it cannot be solved with one executive order. Skilled trades require apprenticeships, experience, decent wages, and employers willing to invest in people beyond one construction cycle. A data-center boom that pulls workers from existing industrial maintenance crews without expanding training could simply move scarcity around. It might produce impressive paychecks today while leaving vital systems short-staffed tomorrow.

I am also skeptical of any plan that treats energy workers as interchangeable symbols in a political debate. A coal employee, a gas-field worker, a nuclear technician, and a transmission-line worker have different skills and work under different hazards. Transitions take time. If a region closes dispatchable plants while assuming a new resource will appear exactly on schedule, the people exposed to the risk are not limited to employees of the retiring facility. Everyone needing dependable power is exposed. On the other hand, keeping an unsafe or uneconomic plant running indefinitely just to preserve a familiar job title is not sound policy. The answer is a realistic schedule, proper worker transition support, investment in genuine alternatives, and an honest explanation of what the system needs at each stage.

The supply debate is becoming less hypothetical. PJM’s July 2026 capacity auction fell short of its forward reliability requirement by 6,831 megawatts. A shortage in a forward market is not proof of immediate blackouts, but it is evidence that anticipated demand is growing faster than the system is obtaining qualifying resources. Federal regulators have also been wrestling with the problem of how large data centers should be tracked and curtailed during emergencies. Those are not minor administrative matters. A poorly designed curtailment rule could harm customers unfairly. No curtailment capability at all could put everyone at risk. The objective should be a transparent and technically workable set of priorities before a cold snap or heat wave makes those priorities unavoidable.

I would distinguish between uses of computing, even if the industry dislikes the implication that not every calculation is sacred. A facility storing medical records, processing emergency communications, or managing essential financial operations has different interruption consequences from an operation training a model whose work can resume tomorrow. There may be engineering reasons why some supposedly deferrable calculations cannot be interrupted instantly without damage; that is for operators to explain and demonstrate. The point is not for politicians to pick favorite computer programs. It is to structure contracts so that a customer pays for the degree of firmness it needs and receives appropriate compensation when it offers genuine flexibility. A society can value innovation while recognizing that human needs are not identical to every commercial demand.

We should examine what ‘self-powered’ means too. If a company builds a private generator but expects the grid to replace its entire output whenever the unit trips, it has not entirely freed itself from the public system. The standby service has value. The utility may have to reserve generation, transmission, and operating flexibility for it. If the private plant runs on natural gas, pipeline reliability matters. If it relies on batteries, the duration of stored energy matters. None of these facts makes private supply a bad idea. They make the details important. A corporation wealthy enough to build a computing campus should be able to pay for the services it actually uses, including the services it hopes never to need.

The other recurring promise is that advanced chips will become efficient enough to solve the load problem. More efficient equipment is welcome. I have seen factories lower electricity use per item produced through better motors, controls, and process design. But it is possible to reduce energy per unit while increasing total output so rapidly that total consumption still rises. The new AI industry is attempting to make computing useful in more areas of life. If it succeeds, demand for the service may expand rather than contract. There is no contradiction there. The sensible course is to support efficiency while building realistic supply, not to pretend one removes the need for the other.

Electricity costs also have to be discussed without assuming every rate increase is caused by data centers. Fuel prices, maintenance backlogs, grid modernization, weather damage, and existing market structures already affect bills. It would be dishonest to blame the newest industry for every dollar of cost pressure. It would be equally dishonest to pretend a new load never raises capacity or transmission costs. Regulators ought to publish careful studies separating the direct cost of serving a campus from the wider benefits or burdens. They should require long-term commitments where large investments depend on one customer. Citizens deserve more than an industry talking point from either side.

The incentives offered by local governments deserve scrutiny. I understand why mayors and county executives compete for investment; I have seen what happens when a town receives none. But the benefits of a data center differ from those of a labor-intensive manufacturing plant. During construction, employment can be extensive. Once operating, the facility may employ comparatively fewer people than its enormous electricity demand suggests. The property-tax base may be attractive, depending on abatements and the way equipment is assessed. Before giving away revenue, officials should calculate permanent jobs, wages, local purchasing, infrastructure costs, and the fiscal consequences if the technology depreciates faster than expected. It is possible for a city to gain a landmark investment and still get a poor bargain.

A question almost nobody asks loudly enough is what happens if the AI investment rush reverses. I do not pretend to know when or whether it will. Some technologies change society and also produce periods of financial overbuilding. Railroads were transformative, and railroad securities still ruined investors. Telecommunications created an indispensable infrastructure while bankrupting companies during periods of excess construction. If utilities build for the most aggressive forecast and the demand fails to appear, ordinary customers should not inherit the entire burden. Binding contracts, deposits, and fair exit obligations matter because they preserve the distinction between private commercial risk and a publicly regulated monopoly’s obligation to serve.

There is also an environmental conversation that should not be reduced to slogans. Natural gas can provide needed dispatchable power, but its infrastructure and emissions have costs. Nuclear energy supplies firm low-carbon output, but operating safety, fuel handling, and financing must be taken seriously. Wind, solar, and storage can make substantial contributions, but their performance during critical hours must be evaluated honestly. Communities living near power plants, pipelines, transmission corridors, and cooling systems deserve meaningful review, not a lecture about how opposition threatens progress. Industrial regions know what it means to live beside infrastructure whose benefits flow elsewhere. There is nothing anti-technology about requiring the people who profit to acknowledge the people who bear the consequences.

A region can gain more from this moment if it treats the demand as an opportunity to rebuild its whole industrial base. Improved transmission could attract advanced manufacturing. Expanded apprenticeships could support local utilities and maintenance firms. Retrofitting existing generating stations might preserve valuable assets and shorten lead times. Better supply-chain planning could encourage domestic production of power equipment. Those benefits should be part of the analysis, but they must be earned through actual projects. A press release cannot provide the stability needed for a small manufacturer to hire another shift. Commissioned equipment and reliable service can.

The national debate also needs some memory. Industrial communities have repeatedly been promised that a new economy would make past sacrifices worthwhile. Sometimes the promised jobs came, and often they did not. That history explains why people grow suspicious when a powerful company requests tax incentives, preferential infrastructure, or exceptions from rules others must follow. It does not mean they reject innovation. It means they expect the company to keep its commitments. Local workers are not obstacles in a national strategy. They are part of the system that makes the strategy physically possible. If technology firms want to be seen as partners in economic renewal, there is a straightforward path: invest, train, pay, and remain accountable after the cameras leave.

My preference is a broad energy expansion that treats reliability as a common asset. Approve useful generation and transmission, preserve safety standards, speed up needless procedural delays, and build training pipelines for the trades. Make massive new customers pay their attributable costs and demonstrate their ability to stand behind long-term obligations. Require realistic planning for outages and emergencies. Above all, do not tell a family already struggling with bills that it must accept worse service so a highly profitable company can open a project before its own power supply is ready. Economic development should not become a contest in which the strongest customer gets the current and everyone else gets the explanation.

The AI boom may prove to be an extraordinary chapter in American technological history. I hope it does. But I have watched enough economic miracles come and go to know that their foundations are built by ordinary people doing difficult work. The electric grid is one such foundation. Strengthening it would benefit computer companies, older factories, and households alike. Weakening it in the name of speed would leave the country with expensive buildings and a familiar excuse. There is no virtue in choosing nostalgia over progress. There is also no wisdom in asking the old industrial towns to pay for another miracle they have been told they should simply trust.

Dublin, Ireland Correspondent

I am reading the American debate about data centres with a certain weary recognition. People speak of a possible future in which one industry absorbs a striking share of electricity, local communities disagree about who benefits, and regulators struggle to connect new facilities without degrading service to everyone else. In Ireland we do not have to imagine the outline of that problem. Our Central Statistics Office reported in July that data centres consumed 23 percent of the country’s metered electricity in 2025. That figure was 5 percent in 2015. It is not a forecast made by a consultant selling either fear or optimism. It is a measure of what happened. It does not prove that data centres caused every difficulty in our energy system, but it does make clear that the physical footprint of digital services is no longer a marginal subject for an environmental committee.

I work in compliance for a software business, so I have no interest in pretending that cloud infrastructure is some optional luxury we can do without. European and American businesses rely on it for payments, records, collaboration, cybersecurity, and the very services that make smaller companies competitive. Ireland’s position as a technology hub has provided jobs and investment, and foreign technology firms are woven into our economy. I also know that an impressive digital business may employ relatively few people at a given campus compared with the electricity it uses. Economic benefit is real, but it must be measured correctly. The question is not whether data centres are good or bad. It is whether each new facility is being connected on terms that leave the electricity system stronger or at least no weaker.

The scale of American plans is formidable. Berkeley Lab’s June 2026 update estimated that data centres might account for 9.5 to 15.3 percent of U.S. electricity use by 2030. The International Energy Agency expects global consumption to rise considerably, though it acknowledges uncertainty arising from the adoption of AI, hardware efficiency, and the speed of energy development. The American grid is vast, and an American percentage cannot simply be compared with Ireland’s as if the two systems had identical geography or resources. But our experience should make one point plain: the change can be fast enough to outpace infrastructure decisions whose approval processes were designed for a different era. A country can welcome technology investment and then discover that the pace of electric load growth has created a problem it did not price into the bargain.

Ireland’s electricity regulator, the Commission for Regulation of Utilities, published a new connection policy for data centres in December 2025 after consultation about how additional connections should be handled. That process was not an admission that digital infrastructure has no value. It was an attempt to reconcile its economic role with a constrained power system. A prospective customer must understand that a service connection depends on engineering and security-of-supply conditions, not merely on a signed property lease. Americans facing similar questions should study the regulatory choices made here, including the trade-offs and controversies, rather than treating a pause or a conditional connection as a purely ideological act.

The lesson I would emphasize is transparency. The European Union has introduced energy-performance reporting requirements for larger data centres, along with a common database intended to track energy and environmental metrics. The initiative is imperfect and, like most European regulatory projects, at risk of creating paperwork that grows faster than practical insight. Yet measurement is necessary when policymakers propose to offer incentives, modify grid plans, or impose environmental conditions. Annual electricity use, maximum demand, cooling needs, renewable-energy claims, and site-specific effects should be recorded in a comparable form. Reports do not generate megawatts, but a regulator cannot responsibly allocate costs or impose standards when each applicant is allowed to define its own numbers.

I am particularly suspicious of claims that a company is entirely powered by clean energy based on an annual purchase arrangement. Such contracts can be valuable because they support investment in renewable generation. They do not, by themselves, establish that a facility runs on a particular supply at every hour. Wind generation varies; solar output is different at midnight; and a connected grid must keep supply and demand balanced continuously. If a company wants to claim that its electricity service is carbon-free around the clock, it should substantiate the claim with hourly matching or a similarly rigorous method. Otherwise it should describe the actual contractual arrangement without allowing a reader to infer more than the evidence supports. The distinction is environmental, commercial, and regulatory, not a semantic quarrel.

The same problem arises with ‘bring your own power.’ A private generator may contribute new electricity and relieve pressure on the public system. It may also remain dependent on grid standby arrangements, gas networks, or other common infrastructure. The cost of those services must be understood. If a large customer has its own equipment but expects other users to finance the reserve needed when that equipment stops, the headline about self-sufficiency is incomplete. Likewise, an investor should not assume that an arrangement labeled behind the meter frees it from ordinary environmental or reliability requirements. The legal label and the physical function need to correspond. A regulatory framework that looks only at who owns a generator rather than how a system behaves will not withstand the next serious disturbance.

The American decisions this October are revealing. Federal regulators approved better tracking of large data-centre loads in the PJM region, but rejected a proposed curtailment mechanism in its presented form. At the same time, large power-supply agreements with established generators and utilities are moving forward. I see progress in the willingness to negotiate long-term physical supply rather than rely exclusively on image management. Yet the regulatory dispute shows how much work remains. Emergency powers need a lawful basis and clear conditions. A customer must know when its service can be interrupted and what compensation or obligations apply. Other customers should know why that facility is being treated differently. Improvising a hierarchy in an emergency would be unwise, but that is an argument for designing fair rules, not for avoiding them.

European policymakers have their own temptation toward comforting abstractions. A target for renewable-energy deployment may be worthwhile, but it does not create transmission capacity in the exact location or hour a data centre needs electricity. A proposal to triple regional computing capacity by a future date does not explain who will build the electrical infrastructure. Energy efficiency requirements may lower use per computation, but demand can continue growing as services become cheaper and more widely adopted. The European Union’s ambitions for digital sovereignty, climate performance, and industrial competitiveness cannot all be satisfied by writing three separate strategies that never interact. They need a coordinated physical plan whose constraints are admitted rather than hidden.

From an Irish perspective, geography adds a political complication. Data centres have concentrated near attractive network and commercial locations. Their impact is not dispersed evenly across the country. A national total can conceal serious local connection constraints; one region may have spare capacity while another requires major upgrades. An applicant’s preference for a familiar cluster is understandable because connectivity and technical services are already present there, but public authorities must consider whether expansion in that location remains sensible. The proper response may sometimes be to develop new sites rather than burden an already constrained one. Geography is a real planning input, not a matter of prejudice against one county.

It is also worth asking what proportion of a data centre’s workload can be moved. Software businesses know that some tasks are tied to latency, jurisdiction, confidentiality, or service commitments. Others may be scheduled where electricity is cheaper or cleaner at a particular hour. A responsible policy should create incentives to identify this difference rather than impose an unrealistic blanket assumption that all computation can be interrupted. Flexibility can reduce pressure during emergencies and improve use of existing renewable supply, but only if the operator’s equipment, contracts, and customer obligations permit it. A regulator should seek verified capabilities and measurable commitments. Good policy avoids both the fantasy of perfectly movable computation and the claim that no computer task could ever be postponed.

There is a data-governance aspect here that Americans may underappreciate. Large connected computing facilities now play a central role in public and private services. Security, operational continuity, and incident reporting are therefore relevant to energy reliability as well as cybersecurity. A failure of cooling can knock out essential digital services even where an electrical system remains stable. Conversely, an abrupt power interruption can cause operational consequences extending far beyond the facility’s walls. Policymakers should distinguish ordinary batch training from infrastructure supporting hospitals or public agencies, and require credible continuity planning. It would be ironic if we spent heavily to secure digital systems against malicious attacks while giving little attention to the ordinary electrical dependencies that can disable them.

My work also makes me attentive to the difference between formal compliance and effective compliance. It is easy to create an elaborate reporting framework that nobody uses to make decisions. The test of Europe’s database should be whether officials can identify misleading claims, compare actual performance, and make better planning decisions. Data quality matters. If operators report incompatible boundaries, different cooling assumptions, or estimated rather than metered electricity without marking the difference, the resulting ranking will not deserve confidence. Similarly, regulators should not reward firms merely for completing forms. Better performance in energy efficiency, load flexibility, water management, and reliable operation should be the actual goal.

Cost fairness is another principle that crosses the Atlantic. Large technology companies are capable of making financially enforceable commitments. They should do so when a grid operator must build expensive assets to meet their demand. Ohio’s adoption of long-term minimum charges, security arrangements, and exit provisions offers a useful American example. The precise terms will differ by legal system, but the reasoning is familiar: a customer that asks the public infrastructure network to invest specifically for it should not retain a cheap option to disappear afterward. This matters most when forecasts include a large number of announced projects, some of which may never reach commercial operation. The financing risk should not be silently assigned to households or smaller firms.

There is a related question of democratic consent. Communities may welcome the taxes and investment associated with a new campus while questioning water use, new power lines, or the distribution of benefits. Describing opponents as anti-innovation does not resolve those concerns. Equally, opponents should not assume that every data centre is a climate disaster or that cloud services can be conjured into existence without physical facilities. A mature process requires project-specific disclosure, opportunity for informed objections, and an explanation of why the proposed use of scarce resources is justified. Ireland’s debates have been heated precisely because the choices are real. They deserve more than slogans.

The energy transition increases both urgency and complexity. Electrifying transport and industry adds demand, while decarbonising generation changes the portfolio used to meet it. AI companies are entering this system at a moment when it is already being rebuilt. They can contribute financially to new low-carbon capacity, storage, and transmission. They can also aggravate bottlenecks if their development schedules outrun public infrastructure. Gas generation may fill some gaps; nuclear may be attractive in jurisdictions able to use it; wind, solar, storage, and interconnection will remain crucial in others. No single technology supplies an exemption from careful planning. Countries should select mixes according to resources and realistic delivery schedules rather than merely according to the politics of a favored fuel.

I am encouraged by companies making longer-term commitments and by regulators collecting better data. I am less impressed by announcements that combine several future projects into one enormous figure without explaining dates, contingencies, or actual system benefits. A government should publish how much capacity is operating, how much has binding financing, how much has permits, and how much is an aspiration. Those categories matter enormously. A power agreement awaiting plant construction is not equivalent to a generator delivering electricity today. Nor should a forecast of electricity consumption be confused with a demonstrated public benefit from the computing it will support. One is a demand estimate; the other requires economic evidence.

Ireland has already demonstrated that the digital economy can become a major electricity consumer in a remarkably short time. We have also shown that government must sometimes revisit its connection policies to preserve confidence in the system. Americans and Europeans should take the lesson without drawing a simplistic conclusion that technology investment must stop. The more useful answer is to connect investment to measurable obligations: honest load forecasts, enforceable financing, transparent energy claims, clear emergency rules, and public accounting of costs. If the industry can meet those conditions, it has a stronger claim to be an engine of prosperity. If it cannot, describing its products as intelligent does not make its infrastructure demands reasonable. The electricity network is not a slogan. It is the thing that must work after the conference ends.

Bismarck, North Dakota Correspondent

A person looking at North Dakota from an airplane might imagine we have solved the energy question. There is land, there are fuels, there are wind turbines, there are power plants, and in places there are more miles of transmission line than houses. I work around energy-sector compliance and community relations, and I have learned that plenty of resources on a map do not equal electricity at the location and hour someone needs it. The difference is infrastructure. It is the process of taking an available resource, turning it into dependable power, and delivering that power while respecting landowners and the people who already depend on the system. Artificial intelligence companies discovering the Plains as a potential home for data centers are welcome to visit. They should bring accurate engineering studies and serious financial commitments with them.

I am not dismissing their demand. Berkeley Lab’s June 2026 update estimates that data centers may require between 9.5 and 15.3 percent of American electricity in 2030. EPRI’s scenarios show that some states could have much higher concentrations than the national average. NERC warns that large loads are emerging rapidly and placing pressure on the pace at which generation and transmission can be brought online. These reports describe a real planning challenge. But there is a great deal of variation across the country. The circumstances of a constrained Mid-Atlantic grid are not identical to those of a northern Plains utility. Our resources, weather, customer base, and transmission connections differ. A sound national response gives states room to use the sources that work for them while holding everyone to genuine reliability standards.

My first concern is weather. A data center may be a fine customer in normal conditions, but infrastructure needs to work on days when normal conditions have packed up and left. We face severe cold, blizzards, ice, and occasional summer heat. A generating source’s dependable contribution must be evaluated during those periods. Natural gas-fired power is valuable, but pipeline deliverability and equipment performance in cold weather matter. Coal and nuclear plants provide characteristics the grid may need, but they also have maintenance and outage risks. Wind production can be very useful, yet the system cannot assume its maximum output at every critical hour. Storage helps with short-duration flexibility and some operating services, but battery duration remains a physical constraint. The mix should reflect reliability evidence rather than the political satisfaction of choosing one fuel.

A new data-center campus might offer an attractive long-term market for regional generation, including new natural gas, renewable, or other projects. That could help finance upgrades and support local employment. It might also strain transmission facilities intended for a different pattern of demand. A company that wants a few hundred megawatts cannot simply point to a power plant within driving distance and assume the electricity is reserved for its project. There are existing customers, wholesale arrangements, line ratings, and system-contingency studies to consider. A new connection can require transformers, switchyards, protection settings, and line improvements. The most important part of an agreement is sometimes the paragraph specifying what must be built before full service begins.

I watched with interest the October announcement that American Electric Power plans a 702-megawatt fuel-cell project near Cheyenne under a long-term arrangement to supply Google. Wyoming is not North Dakota, but the regional lesson is similar. Technology firms are beginning to make material commitments to physical energy infrastructure in the interior West rather than merely securing real estate and waiting for a utility to deliver. That is useful progress. It does not prove the project will be completed on its target schedule, or that every fuel-cell installation will perform identically under the conditions of a northern winter. Large projects need independent engineering, commissioning, service support, and clear arrangements for failures. I would much rather see a company sign a credible twenty-year commitment than request a speculative reservation and leave the risk with everybody else.

I also worry about the way ‘private power’ is advertised. If a data center builds its own generators, it may reduce the additional burden on the grid. Yet a private installation may depend on a pipeline that serves other customers, a grid connection for backup, and the same skilled maintenance contractors used by the utility. During an emergency, what happens if both the plant and its backup face a common fuel or weather problem? Does the campus disconnect promptly, or does it draw emergency capacity that the rest of the region assumed would remain available? These questions belong in interconnection agreements and operating procedures. The words behind the meter are meaningful for billing and ownership, but they do not eliminate the physical dependencies.

Electricity generation has environmental consequences, and rural communities often encounter them more directly than city residents who use the finished product. Transmission lines cross farms and grazing land. Pipelines and access roads disturb property. Water withdrawals and discharge arrangements require care. Construction traffic affects county roads that were never designed for constant heavy deliveries. Such projects can be entirely justified while still imposing burdens that deserve compensation and thoughtful design. Developers should consult landowners early, disclose routes and alternatives, and make commitments they can keep. Nothing breeds local resistance faster than presenting a nearly finished plan and treating residents as an inconvenient footnote to a national economic strategy.

The same is true of promises about jobs and taxes. A large campus creates construction employment, sometimes on a considerable scale. The permanent workforce may be much smaller relative to the capital invested. That does not make the project worthless; a high-value facility can support a community’s tax base and local contracting. But government officials should distinguish temporary construction jobs from long-term positions, and direct local tax revenue from utility-system costs borne over a wider territory. If the project receives property-tax exemptions, road improvements, or discounted service, those concessions must be included in the calculation. A well-off technology customer should not be entitled to a subsidy merely because officials appreciate being mentioned in a national business publication.

Rural communities also have existing demands for electricity that deserve respect. Agricultural processing, fertilizer production, manufacturers, grain-handling facilities, hospitals, schools, and households all depend on reliable supply. A single data center could alter a utility’s load profile enough to change its planning assumptions. Some investment might benefit everyone by justifying upgrades to transmission or generating resources. Other investments might serve only the new campus. The allocation should follow actual benefits and cost causation as closely as practical. It is not fair to hand every cost to the developer if a network expansion also strengthens reliability for thousands of existing customers. Neither is it fair to bury a specialized service cost in ordinary rates because the customer is politically fashionable.

The risks of mistaken forecasting are particularly important in places with smaller customer bases. If a utility builds expensive facilities for a single speculative project and that customer withdraws, there may be few other buyers capable of using the capacity. An underused substation in a growing metropolitan area may eventually find customers; one built in a remote location may not. Long-term minimum payment obligations, collateral, and exit fees are therefore sensible tools. Ohio’s data-center tariff offers an example of asking developers to put financial substance behind extraordinary power requests. The precise terms should reflect regional conditions, but the principle belongs everywhere. Communities should welcome enterprise without taking unpriced options on the balance sheets of private firms.

There is also a legitimate national-security dimension. America has strong reasons to support domestic computing capacity, advanced manufacturing, and reliable energy infrastructure. The international competition surrounding AI will not be won through software alone. The country will need dependable electricity, equipment, trained workers, and secure supply chains. However, national security should not be a blanket justification for bypassing ordinary engineering and environmental requirements. A data center connected to an unreliable grid is not a strategic asset simply because it hosts advanced chips. In a crisis, resilience would matter far more than the number of investment announcements issued beforehand.

I am encouraged that NERC has been working on specific standards for computational loads, including how large electronic facilities interact with the bulk power system. Their operating behavior can matter during faults and other disturbances. Planners need to know whether a large facility will drop off suddenly, how it will restore load, and how backup arrangements affect the system. I have spent enough time in compliance work to know that standards can become cumbersome, but the answer is to make them relevant, testable, and clear. A checklist that verifies the facility’s actual operating behavior is worthwhile. A mountain of vague documentation that nobody uses during an emergency is not.

Technology companies also have opportunities to offer flexibility. Some AI training tasks could be shifted to other times or locations, reducing pressure during a regional peak. Storage and on-site generation may help if the facility’s controls and contracts permit safe operation. But policymakers should not assume that every data center is the same. Essential digital services need firmer arrangements than interruptible batch work. Nor should firms be paid for flexibility that exists only in a presentation. The utility must test the capability and know precisely how quickly the load can be reduced, for how long, and with what effects when it returns. A resource that will not respond in a crisis is not a resource the system can count on.

The clean-energy discussion needs some restraint as well. It is possible to support new wind, solar, and storage while recognizing that dispatchable capacity remains necessary for many system conditions. It is possible to support gas generation while demanding careful assessment of pipeline reliability and emissions. It is possible to favor nuclear energy while acknowledging long construction and regulatory timelines. I would rather have an energy policy built around what each resource actually contributes than a marketing competition over whose megawatt receives the most applause. Large computing customers can help diversify the supply mix if they fund real projects. They can also make the situation worse if they insist on a connection date unsupported by the local system.

A final concern is the physical equipment supply chain. New lines require conductors and structures. Substations need transformers, switchgear, and sophisticated controls. Generators require components that sometimes have long manufacturing lead times. A project can have ample funding and available land yet remain stalled because a crucial item will not arrive in time. Northern regions also have limited construction seasons for some activities and need crews capable of working in difficult conditions. Scheduling must account for these facts. I would be suspicious of a business plan assuming that every vendor, regulator, and skilled contractor will provide the most optimistic possible delivery date. That is rarely how large energy projects behave.

If a computing campus wishes to settle here, I would welcome the economic opportunity while insisting on four things. First, show credible load projections and binding customer commitments. Second, identify the actual generating and network facilities necessary for reliable service under adverse weather conditions. Third, disclose and fairly allocate costs, including backup and possible project cancellation. Fourth, work with landowners and local governments early enough that their input can affect the design. These are not hostile demands. They are what a serious developer should want because they reduce the likelihood of delays, litigation, and public distrust once construction begins.

I would give rural emergency managers a place at the table earlier than they usually receive one. Their knowledge of winter road closures, fuel deliveries, emergency shelter requirements, and restoration priorities is not glamorous, but it can expose assumptions that will fail under severe weather. A proposed campus may have a perfectly adequate electrical connection on paper and still create problems if its backup fuel deliveries require roads that are regularly closed during storms. A utility may need access to a remote substation when visibility makes travel difficult. These are ordinary realities for the Plains, and a reliable project plans around them rather than labeling them exceptional after a failure occurs.

Better regional cooperation could also improve outcomes. Utilities, cooperatives, transmission organizations, state commissions, and developers often hold different pieces of the information needed for a sensible decision. Sharing credible load data and project milestones while respecting commercial confidentiality would help prevent both underbuilding and speculative overbuilding. A national economy can benefit from rural generation and computing investment without stripping rural customers of the protections they have long relied on.

The Plains can contribute to the next phase of America’s technology economy. We have fuels, land, transmission opportunities, and people accustomed to working with difficult physical infrastructure. Those advantages are real, but they are not automatic. Energy still has to be generated, transported, controlled, and paid for. Utilities and technology firms are beginning to make meaningful progress when they recognize those obligations in contracts and investments. They are falling behind where forecasts are treated as commitments or deadlines are promised without engineered supply. The future may depend on artificial intelligence, but the electric system will continue to depend on sound decisions made by human beings. I would start with those decisions and work outward.

Savannah, Georgia Correspondent

A port runs on schedules that sound solid until they meet water, weather, crews, trucks, inspections, and equipment that refuses to cooperate. The electronic system can show a container as available while the actual container sits behind three stacks and the driver has an appointment somewhere else. That is not an argument against software. It is a reminder that a record of intended movement is not the same as movement. When I listen to technology executives describe the power they expect for new artificial intelligence data centers, I hear the same confusion on a grand scale. They have land, customers, construction plans, and an anticipated date for switching on the servers. The electricity system has its own schedule, which depends on transformers, generators, substations, technicians, and permits. The two schedules are not automatically compatible because somebody has put them in the same presentation.

The expansion is substantial. Berkeley Lab’s June 2026 report estimates that data centers might account for 9.5 to 15.3 percent of U.S. electricity consumption by 2030. IEA expects global data-center demand to rise sharply as well. These are estimates, and the credible reports say so. I get nervous when people take a forecast range, choose one end, and speak as though the future has already occurred. But neither should we dismiss the risk because a few announced campuses may never be built. Logistics planning is always about uncertainty under time pressure. You need enough capacity to handle plausible demand, an honest record of commitments, and contingencies for when equipment arrives late or not at all.

The Charleston and Savannah port regions understand how a bottleneck migrates. Expand a terminal but fail to improve road access and trucks pile up. Build warehouse space without adequate labor and containers linger. Add cranes without the maintenance staff to operate them and the nominal capacity remains on paper. In the electricity system, a similar chain runs from generating facilities to transmission lines, substations, distribution equipment, and the customer. A data center may stand near a major corridor without being able to connect at the requested load. The transformer needed for its service may have a long lead time. A generator may be ready before the transmission path. A utility may have enough annual energy but inadequate capacity during a regional peak. Progress must be measured at the weakest necessary link.

That means the current race is not just for electrical energy but also for industrial equipment. High-voltage transformers, copper, switchgear, power electronics, turbines, cooling systems, and specialized construction labor all have supply chains. Those chains compete with ordinary utility maintenance, factories, ports, and public infrastructure for many of the same components. A company can pay a premium for delivery priority, but dozens of companies cannot all be first in the same manufacturing queue. Port operators learned long ago that optimistic arrival schedules become useless if nobody checks capacity throughout the network. Utilities and data-center developers should jointly test equipment lead times against supplier commitments rather than assume every future plant will receive its most important parts on demand.

The geography of power matters to logistics in another way. Industrial investment clusters in attractive locations, which puts simultaneous demands on land, roads, water, and electric service. A port-adjacent district may be a logical place for warehouses and manufacturers. A different region might be better suited for a server campus because it has greater spare electrical capacity or better cooling conditions. Local officials should not assume that every large investment belongs in every jurisdiction merely because they would like the tax revenue. Some projects create relatively few permanent jobs per megawatt of demand; others offer broad employment but use far less electricity. The economic analysis needs to compare opportunities honestly instead of awarding every project the same generic score for growth.

I am encouraged by recent signs that the technology industry is taking physical supply more seriously. Google’s arrangement with Constellation to underwrite improvements that provide 890 megawatts of additional nuclear output is more meaningful than a general promise to buy green electricity someday. AEP’s announced fuel-cell project near Cheyenne similarly couples a planned physical resource to a long-term buyer. These projects remain subject to engineering, equipment deliveries, commissioning, and operational performance. The financing commitment matters; so does the contingency schedule. If the promised plant starts late, does the data-center operator reduce its opening load, buy temporary supply, or ask the existing grid to carry the difference? The answer cannot be discovered responsibly after the computers are already installed.

There is a habit in large projects of treating the most visible facility as the main project and everything supporting it as somebody else’s problem. A port terminal can make that mistake with roads. A data-center developer can make it with substations. The local government can help by coordinating reviews early, but a county’s zoning approval cannot commit a regional grid to deliver power on a date the utility has not accepted. Nor should a utility conceal avoidable delays behind technical jargon. Successful execution requires a master schedule shared among the developer, grid operator, major equipment suppliers, and regulators. Critical dependencies should be identified openly, and the costs of missed milestones should be assigned in advance.

The reliability issue is distinct from ordinary construction delay. NERC has warned that data-center growth and other large loads are creating resource-adequacy pressure in North America. PJM’s July 2026 capacity auction fell 6,831 megawatts short of its stated requirement for a future delivery period. Again, the shortfall is not a time and place for a scheduled blackout. It is a warning that the resources being secured are not keeping pace with the forecasted need. In an operating business, such a warning ought to trigger a serious review of alternatives. It should not be answered with the promise that a supplier is enthusiastic or that the next software update will solve the problem.

I would put more attention into electricity service contracts. A customer requesting hundreds of megawatts should identify its realistic schedule of increasing demand, the level of service it requires, and what it will pay if it abandons the project. Ohio’s dedicated tariff for large data centers is one attempt to require long-term payment commitments and discourage speculative reservations. In logistics we routinely deal with the difference between reserving capacity and actually using it; nobody can run a terminal if customers place extravagant reservations without consequence. Utilities should apply the same common sense, adjusted for the public nature of the grid. Binding financial commitments help separate credible industrial demand from a developer keeping open five options it will ultimately use only once.

The opposite side of that bargain is a utility’s duty to keep its promises. If a company has provided accurate load information and funded the required upgrades, it deserves clear milestones, fair procedures, and timely decisions. It should not have to navigate a different answer from every agency or discover that a necessary permit was overlooked after equipment has been ordered. Government can make the system better by coordinating approvals, publishing requirements, and moving decisively when the engineering is sound. Cutting corners would be counterproductive. A transmission line that becomes trapped in litigation or a generating station that fails commissioning is not a demonstration of speed.

Some AI workloads may be flexible enough to help the system. A developer could schedule selected computing work when electricity is abundant, shift certain tasks between regions, or reduce agreed-upon load during emergencies. This requires reliable controls and careful distinction among services. Real-time customer systems, communications, and other critical functions cannot be treated as interchangeable with batch training. From a logistics perspective, the analogy is familiar: some cargo must move immediately, while other shipments can accept a different delivery window for a lower rate. A successful system prices that difference and communicates it clearly. Pretending all cargo has the same urgency creates needless congestion; pretending every computing task is equally time-sensitive may do the same to the electric grid.

There is a catch. If thousands of servers reduce power use and then restart together, the transition itself can cause problems. Controls and backup systems need to be coordinated with the utility, and operators must understand their demand profile during disturbances. NERC’s recent work on reliability requirements for computational loads is therefore worth the industry’s attention. The aim should be rules that reflect actual behavior rather than paperwork completed once and forgotten. Ports depend on safe procedures when heavy equipment starts, stops, and changes operating modes. The interconnected grid has even less tolerance for casual experimentation. New facilities should demonstrate that they can respond predictably under stress.

Water and cooling enter the picture differently from one region to another. In the humid Southeast, industrial cooling design faces different conditions than in Arizona, but it still requires electrical power, water management, and maintenance. Some systems use more water to reduce electrical cooling costs; others consume less water while requiring more electricity. The better answer depends on site conditions, available water sources, and the value of power at constrained hours. Officials should not let a developer display one favorable efficiency measure and imply that the overall resource footprint is negligible. A good industrial permit compares reasonable alternatives and identifies the practical consequences of each.

Another consideration is the ability of utility construction to coexist with the rest of a growing region. Port operators need functioning roads during construction; hospitals need dependable electrical service; public utilities need workers to maintain existing assets. A technology project that absorbs scarce contractors and equipment may indirectly delay unrelated developments. Those costs are difficult to quantify but very real. Conversely, a large project might support training and industrial investment that benefits the whole region. Local officials should plan rather than merely react. Coordinating apprenticeship programs, road improvements, procurement schedules, and construction windows can prevent growth from becoming a series of avoidable emergencies.

There is an international competitive question beneath the local details. China, Europe, and the United States are all trying to develop digital capability supported by electricity and industrial supply chains. America has advantages in technology, capital, and markets, but those advantages do not automatically produce transmission transformers or skilled line crews. The country cannot protect its strategic position by treating its industrial base as a supporting character. The technology firms that want new computing capacity are now partly in the power-infrastructure business whether they enjoy that fact or not. If they finance new generation and strengthen supply chains, they can contribute to a more resilient national economy. If they merely absorb the best connections, they can increase risk for everyone else.

I also distrust the idea that a single national verdict will capture the result. One region may make smart investments that improve reliability and lower average costs. Another may grant costly incentives, build infrastructure for speculative demand, and see rates rise without much local employment. The difference will be in execution, contract terms, and the existing system’s condition. Data from actual operating projects should be published so that communities can compare them. The size of an investment announcement is not a sufficient metric. We should want to know the incremental generation delivered, transmission improvements completed, demand actually observed, service interruptions, and costs borne by different customer classes.

For my part, I want the planners to draw the entire route. Show where the electricity will be generated, how it will reach the site, which upgrades have to occur first, which equipment is on order, who pays for delay, and what happens if the customer consumes less than predicted. Show the contingency when a generator trips or a heat wave raises demand across the region. Then explain how the project benefits the community after accounting for incentives and utility costs. If those answers are persuasive, I will be pleased to welcome a new data center. The digital economy is not unreal. It simply depends on physical deliveries like every other economy.

A functioning port is a daily demonstration that promises and logistics are different disciplines. A ship arriving does not mean its cargo has reached the customer. Likewise, a corporate power agreement does not mean a megawatt has reached a server rack. Utilities and technology firms are making progress when they finance real projects and work through interconnection constraints honestly. They are still creating bottlenecks when they count tentative commitments as completed infrastructure. We can solve these problems, but only by giving the unglamorous schedule the respect it deserves. The grid has a delivery schedule, too, and it cannot be revised by changing the date on a slide.

Phoenix, Arizona Correspondent

I work in hospital operations, which means I am not impressed by a technology that functions beautifully when conditions are perfect and fails when the building is busy. We plan for emergencies precisely because patients do not schedule their problems around available staffing, air-conditioning capacity, or supply deliveries. Phoenix teaches the same lesson about electricity. Extreme heat makes reliable cooling more than a convenience, and a prolonged outage can put vulnerable people at risk. When I hear proposals for enormous AI data centers, I do not begin by asking whether the computers are exciting. I ask what the utility must build, whether our hospitals and neighborhoods will remain protected during a heat emergency, and who is responsible if a promised resource does not come online. That is the ordinary public-safety standard we should apply to extraordinary new loads.

No serious reader can ignore the demand projections. Lawrence Berkeley National Laboratory’s June 2026 update puts the potential data-center share of U.S. electricity use in 2030 between 9.5 and 15.3 percent, with a central estimate of 11.8 percent. EPRI also anticipates substantial growth and identifies Arizona among states that could experience significant concentration of data-center electricity demand. Those are scenarios, not a precise invoice for 2030. They warrant planning because the consequences of an underestimated peak can be severe. Electricity demand is especially consequential in Arizona, where summer conditions already place a large burden on cooling systems. What looks manageable in annual averages may become uncomfortable or dangerous during the hours when the grid has the least room for error.

I believe some of the new computing capacity will produce useful medical technology. Advanced analysis may help interpret diagnostic images, organize records, support research, and improve the scheduling and maintenance tasks that make a hospital function. Better data processing may also support weather forecasting and grid management. I have no interest in declaring all AI frivolous. But technology’s possible benefits cannot be used to evade its immediate infrastructure costs. An algorithm that improves medical care in theory is no consolation to a patient in a building whose cooling systems are stressed because the local network was expanded without adequate reserves. A strong technology sector and a strong public electricity system should reinforce each other rather than become competitors for the last available margin of reliability.

Hospitals spend a considerable amount of effort defining critical loads and backup arrangements. Emergency generators are intended to keep necessary functions operating, not necessarily to supply an entire modern medical campus exactly as under normal conditions. Fuel supplies, maintenance, transfer switches, and cooling requirements all have to work. The presence of a backup system does not make utility reliability irrelevant. The same is true for data centers, which may have redundant power supplies and generators. Their large operations can still depend on a common grid, pipeline, cooling system, or equipment supplier. A proposed campus should explain the physical relationship among these facilities and the public network rather than treating the words redundant and resilient as magic protections.

PJM’s difficulties provide a national warning even for states operating under different arrangements. Its July 2026 capacity auction missed the future reliability requirement by 6,831 megawatts, while regulators have been debating how to collect better information about large data-center loads and establish fair emergency curtailment rules. An auction shortfall does not mean a blackout is scheduled. It means that available and committed resources are not keeping pace with the stated forward need under the existing procurement process. Utilities should not ignore that evidence, and technology companies should not expect a special exemption from responsible system planning. Planning for growth while preserving reserves is not bureaucratic obstruction. It is the service an interconnected utility is supposed to provide.

The most immediate question is who pays for the expansion. A giant campus can require upgrades to substations, transmission lines, and generation. If its operator signs an enforceable long-term agreement to finance those costs, the project may benefit the region. If it makes speculative requests and then cancels after construction begins, other customers may inherit expenses for infrastructure they did not request. Hospitals cannot simply stop purchasing electricity when rates become uncomfortable. Families living on fixed incomes have even less bargaining power. State regulators need to protect them through transparent cost allocation and credible financial security from unusually large applicants. Ohio’s data-center tariff, with minimum commitments and exit obligations, is a useful example of making private demand visible in private financial risk rather than transferring it to everyone else.

The public-benefit calculation has to be realistic. A local government may celebrate an enormous capital investment and count temporary construction positions as if they were permanent employees. Yet once a large server facility is complete, its ongoing employment can be comparatively modest relative to electricity consumed. This does not mean a data center lacks economic value; taxes, contracts, and the services it enables can be substantial. It does mean that officials should evaluate permanent jobs, wages, tax incentives, infrastructure spending, and the effect on utility rates before announcing that the community has won. Hospital administrators live with budgets that have to balance after the ribbon-cutting. Economic-development officials should be held to a similar standard.

Cooling deserves particular attention here. A high-density computing facility produces heat that must be removed. Different systems make different trades between electricity and water consumption. Evaporative systems may use significant water; other methods may conserve water while consuming more electrical power. The right engineering decision depends on the climate, water source, treatment needs, and the hours of constrained electricity supply. In the Valley, water is already a subject of difficult long-term planning. A project should identify whether it will use reclaimed or potable supplies, how its cooling demand changes during exceptional heat, and what happens during restrictions. That is not an environmental slogan. It is ordinary resource planning, and it has direct consequences for the communities sharing those systems.

I would also insist on an honest account of the new power being promised. Technology firms have announced large arrangements for nuclear capacity, fuel cells, and other generation. Google’s October partnership with Constellation is meant to support 890 megawatts of incremental nuclear output at existing plants, while a separate fuel-cell project near Cheyenne is being developed under a long-term arrangement. Those commitments are signs of maturity. They help the industry move beyond simply purchasing available power and assuming more will somehow appear. But none of these announcements should be counted as dependable supply before the equipment is installed, tested, and available. Nor should a company claim to have solved reliability by buying energy certificates that do not correspond to power delivered at the time its facility is consuming it.

The rate of construction creates problems that a hospital operations manager recognizes immediately. When many facilities expand at once, they compete for the same contractors, electrical components, and equipment specialists. Hospitals themselves need transformers, backup systems, generators, and cooling equipment. If their projects are delayed because every supplier has promised an impossible schedule to the booming technology market, the impact is not merely an extra line in a budget. It can postpone services or require expensive temporary workarounds. I do not propose reserving all industrial capacity for healthcare. I propose recognizing that large national construction surges can create bottlenecks across essential sectors, and planning workforce training and supplier investments accordingly.

Could data centers reduce demand when the grid is under stress? Some can, at least for selected functions. Certain computational tasks might be scheduled for periods of abundant power or moved between regions. Other services support real-time activity and cannot safely be interrupted. A fair market could allow operators to buy firm service for critical functions and accept lower rates for truly interruptible workloads. But a utility must know whether the promised reduction will happen, how fast, and whether the facility’s restart creates additional stress. In healthcare we do not accept an emergency preparedness plan solely because somebody signed a form. We run drills. Energy operators should test computational-load flexibility with equal seriousness before counting it as part of their safety margin.

An emergency curtailment policy also requires ethical judgment. Society cannot simply announce that all data centers are unimportant when many host services for clinics, emergency communications, and payment systems. Yet it would be perverse if households in dangerous heat or a medical facility faced heightened interruption risk while a nonessential computing task ran uninterrupted because its owner had negotiated a more powerful political position. The practical way forward is to establish service categories and operating agreements in advance, based on actual criticality and technical capability. Decisions should be lawful and auditable. Waiting until a crisis to determine who receives electricity is both cruel and inefficient.

There is another overlooked aspect: system restoration. When electricity fails, the order in which loads return can matter as much as the initial outage. Hospitals have protocols for restoring essential systems and avoiding damage from sudden changes. Massive computing campuses must develop similar procedures. If several facilities simultaneously reconnect large loads, the system may experience a secondary problem at precisely the moment it needs stability. NERC’s recent work on computational-load reliability requirements is therefore important. It acknowledges that these facilities are not merely passive consumers. Their power electronics and automated controls can affect the behavior of the grid, particularly during disturbances. Safe integration requires collaboration among technology operators, utilities, and system engineers.

Some advocates describe energy efficiency as the inevitable solution. Better chips, model design, cooling, and operational controls are all welcome. The electricity needed for a given task may fall substantially. But making a service cheaper can expand the number of tasks customers want performed, so the total power requirement may keep rising. Hospitals have experienced a version of this with medical technology: more efficient equipment can expand the range of services provided rather than reduce overall resource use. Utilities should therefore combine efficiency expectations with conservative system planning. No hospital would build emergency capacity solely on the hope that next year’s equipment becomes wonderfully efficient; neither should a public grid.

The debate should also recognize how electricity costs compound across households. Higher rates raise the expense of cooling a home, but they also increase costs at grocery stores, clinics, schools, and employers. Those businesses eventually recover some expenses through prices, reduced investment, or thinner margins. A prosperous household may absorb the change. A family dealing with a medical bill or fixed income has fewer options. That is why cost allocation for data-center infrastructure is a public-interest issue, not merely a quarrel among utilities and large corporate buyers. The benefits of technological progress should not be financed through obligations that fall disproportionately on people who cannot negotiate the terms.

None of this requires a blanket halt to construction. A well-designed campus that finances additional reliable power, uses appropriate cooling systems, provides measurable economic benefits, and accepts clear obligations may be a worthwhile neighbor. New demand can help justify grid upgrades and generation that serve the region for years. Artificial intelligence may improve the management of the electric system itself, helping utilities forecast demand, spot faults, and schedule maintenance. I favor pursuing these benefits with care. But I reject the assumption that every project bearing an AI label deserves accelerated service regardless of the infrastructure’s ability to supply it. High technology should be expected to pass a rigorous practical test rather than granted a sentimental exemption from it.

Arizona officials and utilities should publish accessible information about the major loads they are planning to serve, the corresponding generation and transmission projects, the completion dates, and the rate implications. They should distinguish tentative inquiries from financially committed construction and tested operating facilities. They should explain how projects perform under extreme-heat scenarios, not just average annual conditions. Where a company claims to supply its own power, the public should understand its need for backup and the associated costs. Such disclosure may be less appealing than a groundbreaking ceremony, but it gives residents a reason to trust that growth has been planned rather than wished into existence.

At the end of the day, electricity is an essential service before it is an industrial input. That does not mean hospitals and households are the only legitimate customers. It means planners must preserve the system on which all customers depend. Utilities and technology firms are making meaningful progress where they move from vague promises to financing real generation and making enforceable commitments. The missing piece is often public accountability for the full reliability and cost consequences. I welcome better medical AI and a stronger technology economy. I will not welcome a development strategy that treats the margin keeping a hospital cool in July as excess capacity available to the highest bidder. Progress deserves that basic boundary, and so do the people who cannot opt out of needing electricity.