Energy

Most US Data Center Power Requests Will Never Be Built

A Wood Mackenzie analysis reported August 12, 2026 finds grid operators will commit to only 28 percent of the 1,066 gigawatts requested for US data centers. Three more analysts reached the same range independently - and only some utilities have already priced in the gap.

EnergyBy Servola Tech Desk2026-08-156 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Most US Data Center Power Requests Will Never Be Built

Key takeaways

Wood Mackenzie's analysis, first reported by Bloomberg on August 12, 2026 in a piece bylined Gabriel Levin and Emily Forgash, found that US grid operators are likely to commit to only about 28 percent, or 298 gigawatts, of the 1,066 gigawatts now requested for data center interconnection nationwide, leaving roughly 768 gigawatts, 72 percent of the total, as what the industry calls phantom load. A separate summary of the same Wood Mackenzie work put firm service commitments at about 12 percent of requests, with a further 17 percent rated likely to proceed, a different cut of the same data that lands in the same range.
Two more independent analyses reached a similar conclusion the same week using different methods. Bernstein analyst Madison Rezaei's report, titled Data Center Pipeline Probabilities: Separating the credible developers from dudes with PowerPoints, scored only 33 percent, 135 of 492 gigawatts in its own dataset, as credible. Rystad Energy's Reid Ramdathsingh separately rated about 50 percent of PJM's queue as credible against only 14 percent of Texas's, and Rapidan Energy Group's Glenn Schwartz put the nationwide buildable share at 20 to 30 percent, saying grid operators don't know which ones are real and which ones aren't.
The phantom load is already showing up in utility decisions, not just in projections. Exelon, which serves the stretch of the PJM grid running from Illinois to Delaware, cut its own data center interconnection pipeline by roughly 40 percent in July 2026, keeping an 11 gigawatt queue it judges real. In Texas, where 474 gigawatts of requests, about 90 percent of it from data centers, dwarf the state's own 91 gigawatt all-time peak demand, Governor Greg Abbott ordered ERCOT and the Public Utility Commission on August 3 to audit every pending data center project before further approvals proceed, citing an estimated 13 billion dollars in at-risk revenue.
The gap between utilities is the part this week's coverage did not lead with. Dominion Energy and Appalachian Power in Virginia already require developers to pay for 60 to 80 percent of contracted demand whether they use it or not, and Rappahannock Electric Cooperative requires collateral plus payment for 100 percent of contracted load, policies in force since 2025 that pre-priced exactly this phantom-load problem. Utilities without an equivalent policy are still processing speculative requests at face value, which means a signed interconnection agreement is worth a different amount depending entirely on which utility signed it.

Four Analysts, One Number, Different Methods

Wood Mackenzie's finding, as reported by Bloomberg on August 12, 2026, is specific rather than directional: of 1,066 gigawatts of data center interconnection requests sitting in US grid operator queues, only about 28 percent, 298 gigawatts, is likely to be committed to actual service. Wood Mackenzie's Global Head of Grid Edge, Ben Hertz-Shargel, told Bloomberg that much of the planned capacity belongs to new developers with a small number of massive, speculative projects, disproportionately targeting the South and Southwest. A parallel summary of the same analysis, published the same day, broke the number down further: about 12 percent of requests already carry firm service commitments, with another 17 percent rated likely to proceed, together landing close to Wood Mackenzie's headline 28 percent from the opposite direction. PJM, which serves 67 million people from Illinois to Virginia, has itself committed to serve roughly twice as much new large-load demand as its planned generation can support, the clearest sign that even the operator managing the queue cannot yet tell which requests are real.

The Wood Mackenzie number did not stand alone for long. Bernstein analyst Madison Rezaei published a report the same week, titled Data Center Pipeline Probabilities: Separating the credible developers from dudes with PowerPoints, that scored only 33 percent of a 492 gigawatt dataset, 135 gigawatts, as credible using its own criteria. Rystad Energy's Reid Ramdathsingh reached a regional version of the same finding, rating roughly half of PJM's queue as credible against only 14 percent of Texas's, a gap that lines up with Texas's own 474 gigawatts of requests, about 90 percent of it from data centers and more than five times the state's 91 gigawatt all-time peak demand. Rapidan Energy Group's Glenn Schwartz, working independently, put the nationwide buildable share at 20 to 30 percent. Four firms, using four different methods on four different datasets, converged on the same order of magnitude: somewhere between a fifth and a third of what has been requested will actually get built.

The Two Utility Tiers Already Living Inside the Same Queue

That convergence is already forcing real decisions, not just research notes. Exelon, which covers the Illinois-to-Delaware stretch of the PJM grid, cut its own data center interconnection pipeline by roughly 40 percent in July 2026, keeping an 11 gigawatt queue it considers credible and discarding the rest without waiting for a regulator to force the question. In Texas, Governor Greg Abbott ordered ERCOT and the Public Utility Commission on August 3 to audit every pending data center project, pausing new approvals until the review is complete and citing an estimated 13 billion dollars in revenue at risk if the wrong projects get prioritized. Hut 8 Corp's SVP of Energy, Brad Richter, described the mood among developers now hearing utilities say they don't have it anymore: many territories are, in his words, closed for business. Halcyon co-founder Alex Klaessig put it more plainly: everyone's trying to figure out the rules of the road, and Cloverleaf Infrastructure's Brian Janous has floated a first-ready, first-served allocation model as one candidate for what those rules should be.

What this week's coverage understated is that some utilities never waited for a rule change to write their own. Dominion Energy and Appalachian Power in Virginia have required, under policies in force since 2025, that data center developers pay for 60 to 80 percent of contracted demand regardless of whether they draw it, and Rappahannock Electric Cooperative goes further, requiring collateral on top of payment for the full contracted load. Camus Energy CEO Astrid Atkinson has estimated interconnection requests run five to ten times the number of data centers that actually get built, and former Texas Public Utility Commissioner Karl Rabago summarized the underlying economics in one line: when it's cheaper to buy a queue position than not to use it, developers will buy. Utilities that priced that incentive out of existence in 2025 are now sitting on a queue that already screens itself. Utilities that did not are the ones doing 40 percent culls in July and emergency audits in August.

What a Signed Interconnection Agreement Is Actually Worth

The number every outlet led with this week, 72 percent phantom, is a national average, and averages hide the variable that actually matters to anyone holding a queue position today: which utility is on the other side of the agreement. In a Dominion, Appalachian Power, or Rappahannock-style territory, a signed interconnection agreement already reflects a developer who has put real money behind it, 60 to 100 percent of contracted demand plus collateral, so the queue is smaller and closer to the buildable 28 percent by construction. In a territory that has not adopted an equivalent policy, a signed agreement is one paper commitment among many, indistinguishable on paper from the phantom load around it, until the utility performs its own cull on its own schedule, the way Exelon did with no advance warning in July or Texas is now doing under an August 3 gubernatorial order. A developer, co-location buyer, or infrastructure investor evaluating a queue position in 2026 should ask one question before treating any signed agreement as bankable: has this utility already converted the phantom-load risk into an upfront financial requirement, or is it still processing speculative paper at face value.

The same asymmetry follows the money in the other direction. Interconnection studies and early transmission planning cost real capital regardless of whether the underlying project is ever built, and in a hardened territory that cost is substantially covered by the deposits and minimum payments developers already put down. In a territory still processing requests at face value, the planning and study costs sunk into the share of the queue that eventually gets withdrawn, on Wood Mackenzie's math most of it, have nowhere else to land except a future rate case spread across the same regional customers this whole debate is supposedly protecting. The states moving fastest to hardened, collateral-backed interconnection rules are not just filtering out phantom demand for the sake of tidier queues; they are also the ones keeping ratepayers out of a bill nobody has agreed who should pay.

Frequently asked questions

What did Wood Mackenzie's analysis, reported by Bloomberg, actually find?

That of 1,066 gigawatts of US data center interconnection requests, only about 28 percent, 298 gigawatts, is likely to be committed to actual service by grid operators, leaving roughly 768 gigawatts as phantom load that will not be built.

Do other analysts agree with that estimate?

Yes. Bernstein's Madison Rezaei scored only 33 percent of a separate 492 gigawatt dataset as credible, Rystad Energy's Reid Ramdathsingh rated about 50 percent of PJM's queue credible against 14 percent of Texas's, and Rapidan Energy Group's Glenn Schwartz put the nationwide buildable share at 20 to 30 percent, all independently landing near Wood Mackenzie's number.

Are any utilities already filtering out phantom demand?

Yes. Exelon cut its own PJM-territory data center pipeline by roughly 40 percent in July 2026. Dominion Energy, Appalachian Power, and Rappahannock Electric Cooperative in Virginia have required developers since 2025 to pay for 60 to 100 percent of contracted demand plus collateral. Texas Governor Greg Abbott ordered a full ERCOT and Public Utility Commission audit of pending projects on August 3, 2026.

What does this mean for someone actually trying to build or invest in data center capacity?

A signed interconnection agreement is only as reliable as the utility that signed it. In territories with hardened, collateral-backed intake policies, a queue position already reflects real financial commitment. In territories still processing requests at face value, a queue position carries no more certainty than the phantom load around it until the utility performs its own cull, potentially without warning.

Every outlet covering Wood Mackenzie's analysis this week led with the same number: most of the gigawatts requested for US data centers will never be built. Three more analysts, using three different methods, reached close to the same range independently, which is the rare case where a headline percentage is actually solid. What none of them led with is that the utilities on the other side of these queues have already split into two tiers, one that priced the phantom-load risk into its interconnection contracts back in 2025 and one that is still finding out the hard way, in July culls and August audits, which of its own commitments were ever real.

Wood Mackenziedata center interconnectionPJMERCOTExelongrid capacityphantom load

PJM Overrules Its Own Two-Thirds Vote

PJM's board filed a mandatory curtailment plan for data centers on July 27, 2026, five weeks after two-thirds of stakeholders voted for a voluntary-only backstop instead. The new rule takes effect June 1, 2027.

5 min read

Energy

CenterPoint's 14 Gigawatts Come With a Forfeiture Clause

CenterPoint raised its 10-year capital plan to $66.7 billion on 14 GW of Texas data-center demand and promised $5 billion in customer savings. ERCOT's collateral rules, not the demand forecast, are what make that number credible.

5 min read

Decisions

The Third Pivot Was the Warning Magic Leap Missed

Magic Leap filed a WARN notice for 193 layoffs effective 1 October 2026, its fourth strategic pivot after $4 billion raised. The lesson is the pattern, not the pivot.

4 min read

← All articles