Warning Signals in the 7481 MW Data Center Construction Boom: The Real Concern Is Supply-Side Grid Connection Capacity

Author: Christopher Johnson (Avanza Energy)

Compiled by TechFlow

TechFlow Editor's Note:As the generative AI arms race rapidly transmits pressure to underlying energy infrastructure, North America's 7481 MW of data centers under construction in the first half of 2026 has set a new all-time high, creating an appearance of prosperity. However, at the very upstream of the industry chain, capacity entering new planning pipelines has contracted sharply for three consecutive quarters. These seemingly contradictory data points accurately reflect a deep fault in the compute infrastructure cycle: the industry is exhausting existing groundbreaking projects at extremely high speed, while available grid interconnection capacity has been pushed into a corner by grid approvals and power supply gaps.

As the core grid hosting the largest data center cluster in the United States, PJM recorded reliability supply deficits of several thousand MW in two consecutive capacity auctions. Administrative price caps failed to resolve the physical power shortage and instead forced the creation of an extraordinary backstop procurement arrangement lasting 15 years at $555 per MW-day, passing the enormous financial cost on to all electricity consumers.

With only six months of pre-lease supply remaining in core U.S. hubs, the core hard constraint limiting technology giants' expansion in the next phase has shifted entirely from server procurement to available grid interconnection capacity.

Data center construction across North America's eight major markets has just set a new all-time high: capacity under construction reached 7481 MW in the first half of 2026, surging 24.8% year over year and surpassing the previous peak set at the end of 2024.¹

Yet just four months ago, Wood Mackenzie reported the opposite trend: new pipeline projects declined for three consecutive quarters, falling 19% in the first quarter of 2026 alone.²

Both sets of data are completely accurate and do not contradict each other. They measure opposite ends of the same construction pipeline, and the gap between them reveals the true industry trajectory for the second half of 2026 better than any attention-grabbing headline.

CBRE counts projects already under construction; Wood Mackenzie counts new projects that have just entered the planning pipeline.

If the existing construction backlog is being consumed faster than new projects are being replenished, a market can easily set both a record high for capacity under construction and a sharp decline in new starts within the same reporting period, and the data show exactly this reality.

Across all major North American markets, remaining capacity available for pre-leasing is less than 1500 MW, which equates to roughly six months of supply at the current rate of demand consumption.¹

Gordon Dolven, CBRE's head of data center research, stated clearly when releasing the first-half 2026 report: 'The next growth bottleneck will no longer be determined by demand, but by where new power supply can actually be delivered and interconnected.'¹

Opposite Ends of the Same Pipeline

More than 80% of all capacity currently under construction has already been pre-leased, up from 74.3% a year earlier; compared with the end of 2025, market vacancy has remained stable at an extremely low 1.4%.¹

Atlanta surpassed Northern Virginia for the first time, becoming North America's most active construction market with nearly 2900 MW under construction.¹

Northern Virginia still leads in total inventory and recorded the highest net absorption among all markets at 467.7 MW, but its vacancy rate has fallen to 0.2%; every MW scheduled for delivery in 2026 has already been fully booked, forcing all new pre-lease negotiations into 2027.¹

Asking rents rose across all deployment sizes, with rents for mid-to-large tenants requiring 3 to 10 MW surging 8.3%.¹ This is clearly a market pricing extreme supply scarcity, not weak demand.

Wood Mackenzie's data corroborate the formation of supply scarcity from another angle. Newly disclosed pipeline capacity fell to 36 GW in the first quarter of 2026, down 19% from the previous quarter, continuing what the firm called 'a slowdown that has persisted for three consecutive quarters.'²

The strongest evidence that this is structural divergence rather than a simple industry cooling lies in the next line of the same report: capacity in late-stage commercial negotiations—the final stage before formal agreements are signed—nearly tripled in a single quarter, surging from 37 GW to 107 GW.²

High-certainty committed capacity backed by signed formal construction contracts or power supply agreements climbed to 195 GW, and disclosed total project capital expenditures surpassed the $1 trillion mark for the first time in history.²

A market simply sliding into decline could not possibly mature its existing pipeline this quickly; the appearance is exactly what one would expect from a market where the existing backlog is being released rapidly while new project starts are entirely blocked.

Chart: The flow of new projects entering the pipeline has continued to shrink, while the scale of projects that have already broken ground within the existing pipeline has set a new historical record; both accurately depict the same backlog. Data sources: CBRE press release (August 27, 2026); Wood Mackenzie press release (April 2026).

A Candid Correction

To some extent, this article is also a necessary correction of our earlier analysis. In our April 2026 special report on the data center construction boom, we stated that U.S. capacity under construction had recorded its first year-over-year decline since 2020 and said that PJM's December 2025 capacity auction failed to clear the market.³ In light of subsequent developments, both statements were imprecise.

CBRE's official summary precisely explains the first point: "After delays caused by power supply shortages briefly suppressed construction activity late last year, new starts rebounded strongly by 24.8%..."¹ The core mechanism we identified at the time—that power bottlenecks were preventing projects from breaking ground—was entirely real.

The only change was that previously deferred backlog projects finally began construction and were delivered in a concentrated wave, and that concentrated release of backlog appeared on statistical charts as precisely a record construction boom.

The wording about the capacity auction also needs correction. PJM's December 2025 auction actually did clear, with prices reaching the $333.44/MW-day unforced capacity (UCAP) price cap; the real problem was that it failed to procure enough capacity to meet grid reliability metrics, creating a severe 6623 MW deficiency.

PJM officially defined it as "the first auction in the entire RTO's history (including fixed resource requirement areas) to fail to meet the grid reliability floor standard."⁴ This correction is more precise, and as the subsequent auction revealed, it is more alarming than a one-off failure to clear.

The Bottleneck Shifts to the PJM Grid

PJM's July 2026 capacity auction for the 2028/29 delivery year procured 138318 MW of UCAP and ultimately cleared at a new price cap of $325 per MW-day, but a deficiency of 6831 MW remained against grid reliability requirements.⁵ As PJM's official notice stated: "These auctions were the first in PJM history where the entire region failed to meet reliability standards."⁵

At this point, PJM has suffered two consecutive setbacks: following the 6623 MW deficiency in December, the gap widened to 6831 MW just seven months later, even though the auction procured more nominal capacity at a lower price cap.

Comparing the two auctions side by side makes the underlying pattern clear. The total procurement value cleared in both auctions was $16.4 billion.⁴, ⁵ The second auction bought more capacity at a lower price ($325 versus $333.44)—138318 MW versus 134479 MW—yet the reliability deficiency actually widened.⁴, ⁵

Administrative price caps artificially suppressed clearing prices but did nothing to resolve the physical power shortage, while the market response on the supply side was extremely weak: in July's auction, new generation units and capacity uprates cleared only 525 MW of UCAP, a drop in the bucket against a multi-gigawatt vacuum.⁵

Chart: PJM's July 2026 auction procured more capacity than December's at a lower price cap, but the shortfall against reliability targets widened further. Data sources: PJM Interconnection press releases (December 17, 2025 and July 14, 2026).

PJM did not wait passively for the next regular auction. On July 31, 2026, PJM filed an emergency application with the Federal Energy Regulatory Commission (FERC), titled “Reliability Backstop Procurement” (Docket No. ER26-3380-000).

This is a special long-term procurement with a 15-year term, sized to exactly match the previously accumulated shortfall. In PJM’s own words, it is intended to address “unprecedented large-scale new load growth, especially the rapid expansion of data centers.”⁶ The filing set a maximum willingness to pay of a weighted average of $555 per megawatt-day.⁶

This figure was not conjured out of thin air: after removing the price cap, PJM simulated the same auction and published the results. Without price controls, actual clearing prices in all regions except ComEd reached $554.72, while the ComEd region spiked separately to $776.69 due to grid congestion.⁷

PJM also carefully cautioned that the simulation used existing seller offers and could not predict how seller offers would change strategically in a fully unregulated real auction.⁷

Chart: PJM’s July 2026 auction cleared under a $325 price cap. In an uncapped simulation of the same auction, clearing prices would have reached $554.72 in most regions and $776.69 in the ComEd region. PJM’s backstop procurement filing set its maximum willingness to pay at $555 (a bid cap, not an actual result). FERC suspended the procurement proceeding on September 29; the bidding window never opened, and the scheduled December 2 opening fell through. Sources: PJM Interconnection (July 14 and July 31, 2026 filings); FERC order (September 29, 2026).

The administrative price cap did nothing to reduce the real financial cost of the shortfall. It merely carved this massive cost out of the annual spot auction and stuffed it into a special 15-year long-term contract, with pricing landing almost exactly at the true market-clearing price that would have emerged in a fully unregulated market.

Although FERC issued its ruling as scheduled, it did not adopt PJM’s requests wholesale.

On September 29, FERC essentially recognized the necessity of the backstop procurement but delayed its effective date by five months to February 28, 2027 (subject to refund conditions), pending further hearings on three core issues alleged to be potentially “unjust and unreasonable”: the cost-allocation mechanism, the rules for transmission asset owners exiting PJM, and collateral requirements imposed on load-serving entities (one electric cooperative faced a collateral burden of up to nearly $2 billion).⁸

PJM then withdrew the bidding window originally scheduled to open on September 30; although it said it would resubmit a modified proposal on October 29, no new bidding schedule had been set as of the time of writing.⁸ The backstop measure, originally intended to resolve “who pays for the power shortage”, has itself become a five-month legal and administrative tug-of-war.⁸

Our July 2026 deep-dive research report on the December auction cost mechanism detailed how the first shortfall evolved into a massive bill of as much as $16.4 billion; what is happening now is the sequel to that drama: a massive bill has arrived again, packaged as a 15-year installment plan, and now even that repayment plan has stalled.⁹

The Overhyped Project Cancellation Narrative

A third set of industry data provides a vivid micro-level footnote to the specific shape of the power shortage, but this data was clearly oversimplified as it circulated in the market.

Industry think tank Currence (formerly Sightline Climate) reported in February 2026 that, among approximately 140 large planned projects it tracks, at least 16 GW of data center capacity was scheduled for grid connection and delivery in 2026, but only about 5 GW was actually under construction; the remaining 11 GW remained in early paper-planning stages with no signs of groundbreaking.¹⁰

Currence estimates that 30% to 50% of capacity in the pipeline is at risk of missing its 2026 delivery schedule. The firm also provided a key historical benchmark for determining whether this figure is truly critical: in 2025, 26% of expected delivered capacity was delayed, and another 10% of projects were postponed indefinitely.¹⁰

Compared with this historical baseline, the 30% to 50% risk exposure does point to a significantly worse year, but it should not be equated with a cliff-edge market collapse.

Chart: Of the 16 GW of planned data center capacity tracked for 2026, only about 5 GW is actually under construction. Currence estimates that 30% to 50% of the entire pipeline faces the risk of missing 2026 delivery schedules, higher than the 26% delay rate in 2025 but far from the media's alarming claim that 'half of 2026 projects will be completely canceled.' Data source: Currence, Data Center Industry Outlook (February 24, 2026).

In rebutting the 'half of 2026 projects canceled' narrative widely spread by industry media this year, prominent research firm SemiAnalysis directly questioned the 5 GW under-construction statistical denominator rather than debating its qualitative interpretation: 'Counting only the top two hyperscale cloud providers, their actual under-construction capacity already far exceeds 5 GW.'¹¹

If the statistical denominator severely understates actual construction volume, the proportion described as at risk would naturally shrink significantly; and the projects truly exposed to default risk are almost all early-stage paper plans that have not yet broken ground, which corresponds precisely to the fastest-shrinking tier of the pipeline in Wood Mackenzie's data.²

Viewing the two sets of research together, an objective and fair conclusion emerges: a substantial proportion of projects in the 2026 planned pipeline will indeed fail to meet their original schedules, but delays are not cancellations, while media headlines deliberately misinterpret 'delays' as 'cancellations.'

Deeper implications behind the data

Taken in isolation, three disconnected data sets seem to tell three unrelated stories: construction is booming, newly approved projects are stalling, and the PJM regional grid is unsustainable.

Yet once combined, they point to the same sobering reality: the industry is consuming existing under-construction inventory at a record pace, while fewer new projects are starting to replenish it; meanwhile, grid operators have failed to meet their own power reliability commitments twice in a row and are being forced to pay massive risk premiums lasting up to 15 years, attempting to mask, rather than truly resolve, a structural supply deficit through administrative price caps.

Six months of inventory available for pre-lease is not an immediate existential disaster, but an alarm clock that has already started counting down. What will truly determine the market's path forward is not the overheated boom in current headlines, but the hard data from the next quarter.

Source: 深潮 TechFlow