Data Center Frontier Ghost Demand Poll Exposes a Grid Capacity Test

Data Center Frontier has opened a ghost demand poll as utilities face a conflict between record connection requests and projects that might never get built. The September 11 poll asks what evidence developers should provide before utilities reserve megawatts or plan expensive infrastructure. That question shifts the debate from how much electricity data centers want to whether their requests deserve a place in the queue.

The original poll does not announce a binding rule. It asks readers to judge which proof should distinguish committed developments from speculative filings. Yet its timing captures a policy dispute that has moved from industry conversations into formal regulatory proceedings.

Utilities must plan substations, transmission lines, generation, and equipment years before customers begin consuming electricity. Developers, meanwhile, often approach several utilities while comparing sites. A rational strategy for one developer can produce overlapping requests that make regional demand appear much larger than the eventual load.

The primary conflict is therefore credible development versus queue optionality. Utilities need enough evidence to protect grid capacity and other customers. Developers need flexibility before land, tenants, equipment, financing, and power arrangements are fully settled. A screening system that ignores either side will create a different failure.

The Data Center Frontier Ghost Demand Question Has Become a Planning Decision

A load request now affects more than one project because utilities can begin planning around capacity that has not secured a final customer.

The Data Center Frontier ghost demand poll centers on a specific action: deciding what proof should come before a utility reserves power. “Ghost” demand refers to requested electricity capacity associated with duplicative, early-stage, or non-viable developments. It does not mean every unbuilt project is deceptive.

A developer can have legitimate reasons to study several locations. Land negotiations can fail, local permits can stall, or a customer can select another market. Grid availability itself can determine which site moves forward, creating a circular problem for applicants.

The utility cannot treat every inquiry as firm demand, however. A large request can influence forecasts, engineering studies, capital plans, and regulatory filings. Several versions of the same underlying project can compound that influence across different service territories.

That is why the poll matters despite being informal. It identifies the unresolved gate between an inquiry and a capacity commitment. The industry's old processes were designed for fewer requests, smaller loads, and more predictable development schedules.

An Electric Power Research Institute survey illustrates the mismatch. It collected responses from 25 utilities between May and July 2024, including 22 in the United States. Sixty percent had received data center requests of at least 500 megawatts. Forty-eight percent had requests of at least 1,000 megawatts.

None of those respondents served an existing data center connection above 500 megawatts. That contrast does not prove the larger requests are false. It shows that proposed facilities have moved beyond the operating experience represented in the survey.

The same utility survey found that 75 percent still used a standard large-load request process for data centers. Only 4 percent used a specific data center model when analyzing requests. General processes are therefore absorbing projects with unusual scale, timing, and operating patterns.

Every responding utility reported aggregate requests above its existing connected data center load. Almost half faced requests exceeding 50 percent of present system peak demand. At that scale, an inaccurate assumption can alter an entire resource plan.

The immediate change is not a new technology. It is the point at which screening practices become consequential infrastructure policy. A utility that accepts a weak request too early can restrict viable projects or initiate unnecessary investment. One that demands finished-project certainty too soon can block real construction.

This tension leads directly to the people carrying the financial and operational risk. Developers submit the requests, but utilities, regulators, competing customers, and households must respond to their combined effect.

Utilities Are Being Pressured From Both Sides

Utilities must prepare for genuine AI growth without making every customer underwrite capacity reserved for projects that disappear.

Artificial intelligence has increased expectations for dense computing campuses, but utilities cannot build from headlines alone. Their forecasts support decisions about power plants, transmission corridors, substations, transformers, and customer rates. Those assets have long development lives and limited alternative uses.

Waiting for every uncertainty to disappear is not practical. A developer often needs some confidence about power before signing a tenant or ordering equipment. A tenant may need evidence of deliverable power before making a commitment. Utilities become part of the project's commercial validation loop.

Accepting every request at face value creates the opposite problem. A utility can plan around duplicate applications submitted to several regions. It can also count a campus's ultimate size before its first phase has financing, permits, or a credible construction schedule.

The EPRI findings show why utilities lack a common answer. The respondents reported no consensus on incorporating data center requests into load forecasts. Six did not include such requests in their forecasts, while others applied different screens and assumptions.

Forecasting also involves the shape of consumption, not only its maximum. Twenty-six percent of surveyed utilities had experienced operational effects from connected data centers involving increased load ramp rates. A ramp rate measures how quickly electricity use rises or falls over time.

AI computing can complicate that profile because large clusters can shift workloads or change utilization faster than conventional industrial operations. Still, a proposed nameplate capacity does not reveal when each phase will start. It also does not show how often the facility will approach that maximum.

These uncertainties pressure public utility commissions. Regulators must decide whether proposed investments are prudent and who should bear the risk. If a dedicated asset becomes stranded, the cost cannot simply vanish.

The risk extends to other businesses waiting for electricity. A viable factory, housing development, or data center can lose time when engineering teams study speculative projects first. Equipment and personnel committed to weak applications are unavailable for stronger ones.

Legitimate developers also suffer. Former Meta energy strategy director Peter Freed described the market as a struggle to isolate signal from noise. Technology companies can receive the same proposed site through several intermediaries, forcing their energy teams to identify duplicate opportunities.

Reporting on phantom requests documented one utility customer where nearly 30 percent of applications submitted during 2024 were canceled. Public queue data cited in that reporting showed 162 withdrawals among 521 recorded requests since 1996. The dataset was incomplete and should not be treated as a national cancellation rate.

That limitation is important. The United States lacks a standardized, transparent load queue comparable with many generation interconnection systems. No national dataset can reliably label every request as active, duplicated, withdrawn, or operational.

Utilities are therefore being asked to make defensible decisions using fragmented information. The forced response is a staged process that separates preliminary interest from increasingly firm commitments. That process must also keep early exploration possible.

Credible Development and Queue Optionality Cannot Share One Standard

The central policy choice is whether scarce grid capacity follows stated ambition or verified project progress.

Developers value optionality because data center siting is sequential and uncertain. They compare land, fiber, taxes, permits, water, power, construction labor, and customer demand. A viable project can examine several utility territories before selecting one.

Utilities operate under different constraints. They must study cumulative effects and prepare assets before a customer energizes its servers. Their decisions become less reliable when each application represents the full buildout of a project that is also shopping elsewhere.

The practical answer is not a single pass-or-fail test. It is a sequence of evidence gates tied to the consequences granted at each stage. An early inquiry should receive information without controlling the same capacity as a contracted project.

At the first gate, the developer should identify the project entity, site control, requested capacity, initial phase, and expected ramp schedule. Site control can include ownership, an option, or another documented right to develop the property. It proves access, not eventual construction.

The developer should also disclose related applications for the same project or customer. Confidentiality can make that requirement difficult, especially when tenants remain unnamed. Utilities can still request identifiers that allow duplicate detection without publishing commercial details.

A later gate should require permit progress, completed engineering work, and a credible interconnection design. These materials show that the applicant has spent time and resources beyond submitting a broad megawatt estimate. They also give the utility better data for network studies.

Tenant evidence deserves careful treatment. A signed end-user commitment offers strong proof, but requiring one too early can freeze the development sequence. A utility could accept confidential documentation, minimum capacity commitments, or another form of independently reviewable demand support.

Financial security is the clearest economic filter. Deposits, study payments, or withdrawal obligations make speculative applications costly. Yet the amount must reflect the capacity being reserved and the project's stage. A flat fee could favor large incumbents while excluding credible smaller developers.

Britain is testing a more explicit version of this approach. Ofgem proposed a data center commitment fee that would remain secured after a connection offer is accepted. The fee would not be returned when a project fails to progress under the applicable rules.

Ofgem also proposed project milestones covering a credible end user, long-lead electrical equipment, and financial and technical capability. Its connection reforms would let network operators evaluate whether a development remains on track and impose penalties when it does not.

The British proposal is significant because it combines economic commitment with evidence of progress. Neither mechanism is sufficient alone. A well-funded speculator can afford a deposit, while a genuine developer can encounter a reasonable delay.

Milestones make the queue dynamic. Projects retain their position by demonstrating progress rather than merely entering early. Capacity can return to the system when evidence weakens or deadlines pass without an approved explanation.

The model also exposes the tradeoff. Proof requirements reduce weak applications, but they can turn utilities into judges of commercial viability. That role becomes sensitive when officials must compare confidential tenants, financing structures, or strategic importance.

Utilities should therefore define evidence categories publicly and apply them consistently. Regulators should review exceptions and penalties. Developers need a route to cure minor failures without keeping a dormant position indefinitely.

The best standard is progressive commitment. Each additional right, including detailed studies or reserved network capacity, should require stronger proof and greater financial responsibility. This structure makes optionality possible without treating it as equal to construction readiness.

Data Center Load Forecasting Needs Probability, Not a Binary Label

Screening improves the queue, but utilities still need forecasts that recognize uncertainty instead of calling every project real or fake.

“Ghost” is a useful warning, but it can oversimplify a development pipeline. Projects exist along a spectrum, from an exploratory inquiry to an energized building. Forecasts should reflect that progression.

A utility can assign probability weights based on objective milestones. Site control, permits, financing, tenant commitments, equipment orders, and completed studies each strengthen the likelihood of service. Missed deadlines, duplicate requests, or unresolved interconnection requirements weaken it.

Those weights should also account for timing. A project expected to ramp across five years should not enter the forecast as its ultimate load on the first service date. Forecasts need separate values for initial energization, contracted demand, expected expansion, and optional buildout.

Scenario analysis provides another safeguard. Regulators can examine a base case, a credible high-growth case, and a delayed or lower-growth case. Decisions that perform reasonably across those scenarios carry less stranded-cost risk.

The World Resources Institute has highlighted Texas as one emerging example. ERCOT developed an adjusted large-load forecast that discounts utility projections using observed delays, ramp schedules, and the share of expected load already operating. That forecasting method moves beyond accepting submitted capacity as a single guaranteed outcome.

Probability weighting cannot replace project diligence. A utility still needs accurate technical information to study local constraints. It can, however, prevent a portfolio of uncertain projects from becoming one falsely precise number.

The International Energy Agency has identified a wide gap between connection queues and expected delivery across several markets. Its 2026 electricity analysis estimated that only around 20 percent of reported U.S. data center connection requests materialize in the short to medium term.

The same analysis cited 44 gigawatts of requests in Australia, against an estimate that 8 gigawatts would enter service. Brazil had more than 26 gigawatts of requests by November 2025, while 6 gigawatts were under review or at advanced stages.

These comparisons do not establish a universal discount rate. Markets use different definitions, timelines, and reporting systems. Applying 20 percent to every utility's queue would replace one crude assumption with another.

The numbers instead show why data center load forecasting must preserve ranges. A request represents information about developer interest, not a guaranteed operating load. Its predictive value improves as evidence accumulates.

Utilities should publish their methodology even when project details remain confidential. Regulators need to see which milestones change a project's probability and how delays affect timing. Other customers need confidence that large investments do not rest on opaque assumptions.

Forecasts should also distinguish system planning from project reservation. A utility might include a probability-weighted share of early projects in a regional scenario. It should not grant each one an equal claim on the same constrained substation.

This distinction reduces the temptation to classify projects publicly as real or fake. A development can remain plausible in a planning scenario while receiving no firm capacity reservation. It gains stronger treatment only after meeting the next evidence gate.

The Data Center Frontier ghost demand debate is therefore partly a data governance problem. Utilities need consistent identifiers, status definitions, milestone dates, and withdrawal records. Without those basics, sophisticated forecasting models will still process unreliable inputs.

Tougher Screens Can Create New Forms of Gatekeeping

Any rule designed to remove ghost demand can also protect incumbents, expose confidential plans, or reject projects whose schedules change for legitimate reasons.

The strongest case for commitment fees is behavioral. Developers submit fewer marginal requests when each application carries a meaningful obligation. The fee also allocates some development risk to the party seeking capacity.

The skeptical case starts with market access. Large hyperscalers and established operators can secure deposits more easily than new developers. A high financial threshold can make balance-sheet size a substitute for project quality.

Tenant disclosure presents another problem. Operators may be bound by confidentiality agreements, while prospective tenants can require proof of power before signing. A rule demanding a named customer at the opening stage can make a legitimate project impossible to validate.

Equipment procurement can also become circular. Long-lead switchgear or transformers demonstrate commitment, but developers need technical specifications before placing orders. A utility cannot reasonably demand equipment tied to an interconnection design it has not finished studying.

Milestones must therefore match what the developer can control. A permitting delay caused by an agency differs from repeated inactivity. A supply-chain delay with documented orders differs from an applicant that has not started procurement.

Standardization helps, but local grids are not identical. A rigid national checklist can miss regional constraints, ownership structures, or development practices. Utilities still need limited discretion, accompanied by written reasons and regulatory oversight.

Another risk involves strategic prioritization. Governments might move favored data centers ahead because of promised jobs, AI policy goals, or national security claims. Such decisions can subordinate queue readiness to political judgments.

Britain's proposal makes that tension visible. Its broader program includes curating viable projects, planning for strategically important demand, and accelerating connections. Those goals can align, but they do not always identify the same winner.

Ofgem reported around 73 gigawatts of data center demand in the relevant queue. Its July 29 announcement said total demand connection applications had more than tripled within one year. The regulator argued that speculative projects can delay viable schemes and complicate investment planning.

The regulatory statement framed stronger requirements as consumer protection. That is defensible, but the final rules must reveal how fees, milestones, exceptions, and strategic status interact.

The United States faces an additional fragmentation problem. Utilities and commissions can develop different applications, deposits, study rules, and reporting practices. Developers may continue filing broadly because each territory provides different information and timelines.

Former FERC commissioner Allison Clements and Peter Freed have argued for a standardized large-load interconnection process. Their standardization proposal compares today's load problem with earlier congestion in generation interconnection queues.

That historical reference offers both support and warning. Generation queues became crowded with projects that had uncertain prospects, prompting deposits, readiness requirements, and withdrawal penalties. Reform took years, and stricter gates did not eliminate regional bottlenecks.

A load-side framework should borrow the useful mechanics without assuming that generators and customers are identical. A data center consumes electricity, may arrive in phases, and can sometimes adjust operations. Its service obligations and reliability effects differ from those of a proposed power plant.

No screening test can promise perfect classification. Some qualified projects will fail, while some early applications will become major campuses. The defensible objective is better allocation under uncertainty, not the elimination of every forecasting error.

Three Signals Will Show Whether Ghost Demand Rules Work

The next test is whether formal reforms produce cleaner queues without slowing credible construction or transferring new risks to ordinary customers.

The first signal is Ofgem's final treatment of commitment fees and milestones after its consultation closes on September 16, 2026. The important details include fee security, refund conditions, evidence deadlines, and remedies for delay.

A final rule that binds capacity rights to measurable progress would strengthen the case for progressive commitment. Broad exemptions or weak withdrawal consequences would preserve much of the existing optionality. Excessive fees or premature tenant requirements would instead strengthen concerns about incumbent advantage.

The second signal is adoption of transparent probability-based forecasts by U.S. utilities and regulators. Watch for filings that separate contracted initial load from expected expansion and optional campus capacity. Scenario ranges should replace one headline number drawn from the queue.

That change would strengthen the argument that screening and forecasting solve different parts of the problem. Continued reliance on unadjusted requests would weaken confidence in proposed infrastructure spending, even if utilities introduce larger deposits.

The third signal is measurable queue behavior. Regulators should track duplicate applications, withdrawals, time between milestones, study completion, energization dates, and the share of forecast load that becomes operational.

Falling application totals alone would not prove success. Strong projects might also be withdrawing under poorly designed rules. A better result would combine fewer stale requests, faster studies, credible project advancement, and reduced cost exposure for unrelated customers.

Data center developers should watch these signals because power access increasingly shapes where computing capacity can be built. Enterprise technology buyers should care because connection delays affect the availability and location of future cloud and AI infrastructure.

The Data Center Frontier ghost demand poll has surfaced the right question, but no single document can prove a project is real. Utilities need staged evidence, enforceable financial commitments, probability-weighted forecasts, and transparent review.

The final standard should make developers earn stronger capacity rights as their projects mature. It should also protect legitimate site exploration and confidential negotiations. Readers should now ask whether each proposed reform improves both queue accuracy and project delivery, rather than rewarding whichever side submits the largest number first.