North American data center construction reaches 7,481 MW record, but supply constraints emerge

Author: Christopher Johnson (Avanza Energy)

Compiled by Deep潮 TechFlow

DeepChao Overview: Amid the generative AI arms race exerting intense pressure on underlying energy infrastructure, the 7,481 MW of under-construction data centers in North America through the first half of 2026 has set a new historical high, reflecting a period of robust growth. However, at the very top of the supply chain, the capacity of newly planned projects has declined sharply for three consecutive quarters. These seemingly contradictory figures precisely reveal a deep structural fracture in the compute infrastructure cycle: the industry is rapidly exhausting existing approved projects, while后备并网资源已被电网审批与供电缺口逼入死角.

As the backbone grid supporting the largest data center cluster in the United States, PJM has recorded multi-gigawatt reliability supply deficits in two consecutive capacity auctions. Administrative price caps have not alleviated the physical electricity shortage; instead, they have forced the creation of an exorbitant “backstop procurement” scheme lasting 15 years at a price of up to $555 per megawatt per day, shifting massive financial costs onto all electricity consumers.

With only six months of pre-leasable supply remaining at the nation’s core hub, the primary hard constraint on tech giants’ expansion has fully shifted from server procurement to available space for power grid connection.

Data center construction capacity in North America’s eight major markets has just reached a new all-time high: under construction capacity hit 7,481 MW in the first half of 2026, a 24.8% year-over-year surge, surpassing the previous record set at the end of 2024.¹

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

These two sets of data are entirely accurate and do not contradict each other. They measure the two ends of the same construction pipeline, and the gap between them reveals the true industry trajectory for the second half of 2026 more clearly than any single sensational news headline.

CBRE tracks projects that have already broken ground and are under construction; Wood Mackenzie tracks newly added projects that have just entered the planning pipeline.

If the rate at which backlog projects are being absorbed exceeds the rate at which new projects are added, a market could simultaneously set new records for “highest ever under-construction scale” and “sharpest decline in new project initiation growth” within the same reporting period—and the data precisely reflects this reality.

In all major North American markets, less than 1,500 megawatts of pre-rentable capacity remains available, which at current demand consumption rates equates to roughly six months of supply.¹

Gordon Dolven, Head of Data Center Research at CBRE, clearly stated at the release of the first half of 2026 report: "The next phase of growth will no longer be constrained by demand, but by where new power supply can actually be delivered and connected to the grid."¹

Of all capacity currently under construction, more than 80% has been pre-leased, up from 74.3% a year ago; the market vacancy rate remains at a remarkably low 1.4%, compared to the end of 2025.¹

Atlanta has surpassed Northern Virginia for the first time, becoming North America’s most active construction market with nearly 2,900 megawatts under construction.¹

Northern Virginia remains leading in total stock and recorded the highest net absorption of 467.7 MW across all markets, but its vacancy rate has dropped to 0.2%, with every megawatt of capacity scheduled for delivery in 2026 already pre-leased, forcing new pre-leasing negotiations to be fully deferred until 2027.¹

Rental rates across all deployment scales have risen comprehensively, with rents for mid- to large-scale tenants requiring 3 to 10 megawatts surging by 8.3%.¹ This clearly reflects a market pricing in extreme supply scarcity, not weak demand.

Data from Wood Mackenzie corroborates the formation of supply scarcity from another perspective: newly disclosed pipeline capacity in the first quarter of 2026 fell to 36 gigawatts (GW), a 19% decline from the previous quarter, continuing “a trend of consecutive quarterly slowdowns that has now persisted for three quarters,” according to the firm.

The strongest evidence that this represents structural differentiation rather than mere industry cooling is hidden in the very next line of data in the same report: capacity in advanced business negotiations—the final stage before signing formal agreements—surged nearly threefold in a single quarter, rising from 37 GW to 107 GW.²

The high-confidence committed capacity, backed by signed construction contracts or power supply agreements, has risen to 195 gigawatts, and the total disclosed project capital expenditures have, for the first time in history, surpassed the $1 trillion mark.²

A market simply in decline could never rapidly mature its existing pipeline; the very appearance described here—where existing backlog projects are quickly released while new initiatives face complete obstruction—is precisely what such a market exhibits.

Chart: Projects newly added to the pipeline continue to shrink, while the scale of projects already under construction within the existing pipeline has reached a historic high—both accurately depict the same backlog of projects. Source: CBRE Press Release (August 27, 2026); Wood Mackenzie Press Release (April 2026).

To some extent, this article also serves as a necessary correction to our previous analysis. In our April 2026 special report on the data center construction boom, we noted that the U.S. under-construction capacity experienced its first year-over-year decline since 2020 and stated that the PJM capacity auction for December 2025 failed to clear the market.³ In light of subsequent developments, both of these statements were not entirely precise.

CBRE's official summary precisely explains the first point: "After construction momentum was briefly suppressed at the end of last year due to power supply shortages causing project delays, newly started construction surged by 24.8%..."¹ The core mechanism we identified—that power bottlenecks hindered projects from breaking ground—was indeed real.

What has changed is that previously delayed projects have finally begun construction and delivery in bulk; the concentrated release of these backlog projects appears on statistical charts as a record-breaking construction surge.

The wording regarding the capacity auction also needs to be corrected. The PJM auction in December 2025 did indeed clear, with the clearing price reaching the upper limit of $333.44 per megawatt-day for Unforced Capacity (UCAP); the real issue was the failure to procure sufficient capacity to meet grid reliability requirements, resulting in a severe shortfall of 6,623 megawatts.

PJM officially defined it as "the first auction in the history of the regional transmission organization—including areas with fixed resource requirements—to fail to meet the grid reliability baseline standard."⁴ This revision is more precise, and, as the subsequent auction revealed, it is more concerning than a simple one-time failure to clear.

PJM procured 138,318 MW of UCAP in the capacity auction held in July 2026 for the 2028/29 delivery year, clearing at a new price cap of $325 per MW-day, yet still leaving a gap of 6,831 MW below grid reliability requirements.⁵ As PJM officially stated: “These auctions mark the first time in PJM’s history that the entire region failed to meet reliability standards.”⁵

Thus, PJM has now suffered two consecutive setbacks: following a 6,623 MW shortfall in December, the gap widened further to 6,831 MW just seven months later, despite procuring more nominal capacity at a lower price cap.

When comparing these two auctions side by side, the underlying pattern becomes clear. Both auctions resulted in a total procurement amount of $16.4 billion.⁴,⁵ The second auction purchased more capacity (138,318 MW vs. 134,479 MW) at a lower price ($325 vs. $333.44), yet the reliability gap widened further.⁴,⁵

Administrative price caps artificially suppress clearing prices but fail to address physical power shortages, and market responses on the supply side remain extremely weak: in July’s auction, newly added generating capacity and expansion upgrades cleared only 525 megawatts of UCAP—insignificant compared to the thousands of megawatts of shortfall.

Chart: PJM’s July 2026 auction procured more capacity than the December auction under a lower price cap, yet the gap in meeting reliability targets widened further. Source: 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 plan with a term of 15 years, designed to fully address the previously accumulated shortfall. In PJM’s own words, its goal is to meet “unprecedented large-scale increases in electricity demand, particularly from the rapid expansion of data centers.”⁶ The application sets a maximum willingness to pay of a weighted average of $555 per megawatt per day.⁶

This figure is not fabricated: PJM simulated the same auction after removing price caps and published the results, showing that, without price controls, all regions except ComEd—due to grid congestion, which spiked separately to $776.69—had actual clearing prices of $554.72.⁷

PJM also cautions that this simulation uses existing seller quotes and cannot predict how seller bids might strategically change in a fully unregulated real auction.⁷

Chart: The PJM July 2026 auction cleared at a price cap of $325. Without a cap, the clearing price in most regions would have reached $554.72, with the ComEd region reaching $776.69. PJM’s backup procurement application set its maximum willingness to pay at $555 (this is the bid cap, not the actual outcome). FERC suspended the procurement process on September 29; the bidding window never opened, and the scheduled bid opening on December 2 was canceled. Sources: PJM Interconnection (filings dated July 14 and July 31, 2026); FERC Order (September 29, 2026).

The administrative price cap has done nothing to reduce the real financial cost of the shortfall. It has merely shifted this enormous cost from annual spot auctions into a 15-year special long-term contract, with the contract price almost exactly matching the unregulated market equilibrium price.

Although the FERC issued its ruling as scheduled, it did not fully adopt PJM’s demands.

On September 29, FERC essentially recognized the necessity of the backup 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 “unjust and unreasonable”: cost allocation mechanisms, rules governing transmission asset owners’ withdrawal from PJM, and collateral requirements imposed on load-serving entities—including a collateral burden of nearly $2 billion faced by one electric cooperative.

PJM subsequently withdrew the bidding window originally scheduled to open on September 30; although it indicated it would resubmit a revised proposal on October 29, no new bidding schedule has been confirmed as of the writing of this article.⁸ What was originally intended as a backstop bill to address “who pays for electricity shortages” has now evolved into a five-month legal and administrative standoff.⁸

Our in-depth report in July 2026 on the December auction cost mechanism meticulously broke down how the first shortfall evolved into a massive $16.4 billion bill; what is unfolding now is the sequel to that drama: the enormous bill has returned, repackaged as a 15-year installment plan, and even this installment scheme has now come to a halt.⁹

The third set of industry data provides a vivid micro-level illustration of the specific forms of power shortages, yet this data has clearly been oversimplified in market circulation.

Industry think tank Currence (formerly Sightline Climate) reported in February 2026 that, of approximately 140 large planned projects it tracks, at least 16 gigawatts of data center capacity were scheduled for grid connection in 2026; however, only about 5 gigawatts are currently under construction, while the remaining 11 gigawatts remain in early planning stages with no signs of physical construction yet.¹⁰

Currence assesses that 30% to 50% of the pipeline's capacity faces a risk of not being delivered on schedule within 2026. However, the agency also provides key historical benchmarks to determine whether this data is truly critical: in 2025, 26% of expected delivery capacity was delayed, and an additional 10% of projects were entirely postponed.¹⁰

Compared to this historical norm benchmark, a 30% to 50% exposure does signify a significantly worse year, but it should not be equated with a market cliff dive.

Chart: Of the 16 gigawatts of planned data center capacity tracked for 2026, only about 5 gigawatts are currently under construction. Currence estimates that 30% to 50% of the entire pipeline faces risks of not being delivered on the 2026 schedule—higher than the 26% delay rate in 2025, but far from the sensational claim by some media outlets that “half of the projects will be completely canceled in 2026.” Source: Currence, “Data Center Industry Outlook” (February 24, 2026).

The renowned research firm SemiAnalysis, in refuting the widely circulated industry claim this year that “half of all projects will be canceled by 2026,” directly challenged the 5-gigawatt figure used as the denominator, rather than engaging with its qualitative interpretation: “Just counting the two largest hyperscale cloud providers, their actual under-construction capacity already far exceeds 5 gigawatts.”¹¹

If the statistical denominator significantly underestimates the actual scale of construction, then the purported proportion facing risk will naturally shrink substantially; and those truly exposed to default risk are almost entirely early-stage, paper-only projects that have not yet broken ground—precisely matching the layer of the pipeline that is contracting most rapidly in Wood Mackenzie’s data.²

When examining these two sets of studies together, an objective and impartial conclusion emerges: a significant proportion of projects in the 2026 pipeline will indeed fail to deliver on their original schedules, but delays are not equivalent to project cancellations—yet media headlines deliberately misrepresent "delays" as "cancellations."

Three sets of disconnected data, when viewed in isolation, seem to tell three unrelated stories: construction projects are booming, newly approved projects have come to a standstill, and the PJM regional grid is struggling to sustain operations.

However, when pieced together, they point to a stark and uniform reality: the entire industry is rapidly consuming its existing pipeline of projects at an unprecedented pace, while new projects available to replace them are dwindling; meanwhile, grid operators have failed twice in a row to meet their reliability commitments, forcing them to pay massive risk premiums lasting up to 15 years, attempting to mask—rather than genuinely resolve—the structural supply deficit through administrative price controls.

The six-month supply available for pre-rental is not an immediate catastrophe, but rather a timer that has already started. What will truly determine the market’s path to resolution is not the current hype in headline news, but the next quarter’s hard data.