Substack 분석에 따르면 기록적인 7,481MW 규모의 데이터센터 붐이 일고 있으며, 그 이면의 6개월 공급 상황에 대해 경고가 나온다.

Data center construction across North America's eight primary markets just hit a record: 7,481 megawatts (MW) under construction in the first half of 2026, up 24.8% year-over-year and past the previous high set in late 2024.¹ Four months earlier, Wood Mackenzie had reported the opposite trend. New pipeline additions fell for a third straight quarter, down 19% in the first quarter of 2026 alone.²

Both numbers are correct. Neither contradicts the other. They measure opposite ends of the same pipeline, and the gap between them is a more important story for the second half of 2026 than either headline alone.

CBRE counts what has already broken ground. Wood Mackenzie counts what is just entering the pipeline. A market can set a construction record and a slowdown record in the same reporting cycle if the backlog is draining faster than it refills, and that is exactly what the data shows. Less than 1,500 MW remains available for preleasing across every primary North American market, roughly six months of supply at the current pace of demand.¹ "The next phase of growth will be defined less by demand and more by where new supply can realistically come online," said CBRE's Gordon Dolven, Director of Data Center Research, when CBRE released the H1 2026 report.¹

More than 80% of all capacity now under construction is already preleased, up from 74.3% a year ago, and vacancy held steady at 1.4% compared with year-end 2025.¹ Atlanta overtook Northern Virginia for the first time as North America's most active construction market, with nearly 2,900 MW under construction.¹ Northern Virginia still leads on total inventory and posted the highest net absorption of any market at 467.7 MW, but its vacancy has fallen to 0.2%, and every megawatt scheduled to deliver in 2026 is already spoken for, pushing new preleasing activity into 2027.¹ Asking rents rose across every deployment size, led by an 8.3% jump for users seeking 3–10 MW.¹ This is a market pricing scarcity, not softness.

Wood Mackenzie's numbers show scarcity building from the other direction. New disclosed pipeline capacity fell to 36 GW in the first quarter of 2026, down 19% from the prior quarter and "extending a sustained deceleration that has now run for three consecutive quarters," in the firm's own words.² The clearest evidence that this is bifurcation rather than a simple slowdown sits one line further into the same release: capacity in advanced discussions, the stage just before a signed agreement, nearly tripled in a single quarter, from 37 GW to 107 GW.² High-confidence commitments with signed construction or electricity-supply agreements rose to 195 GW, and disclosed project capital spending crossed $1 trillion for the first time.² A market that was simply slowing would not be maturing its existing pipeline this fast. A market draining its backlog while new formation stalls would look exactly like this.

New pipeline entries are shrinking while construction of what's already in the pipeline hits a record: two accurate readings of the same backlog.

Sources: CBRE press release, August 27, 2026; Wood Mackenzie press release, April 2026.

This is also, in part, a correction. Our April 2026 piece on the data center buildout reported the first year-over-year decline in U.S. under-construction capacity since 2020 and said PJM's December 2025 capacity auction failed to clear the market.³ Neither line holds up cleanly against what followed, and CBRE's own words explain the first one: "After power-related project delays briefly softened activity at the end of last year, new construction rose 24.8%..."¹ The mechanism that piece identified, power constraints delaying groundbreaking, was real. What changed is that the delayed backlog finished, and a finished backlog looks exactly like a construction record.

The auction language needs its own fix. PJM's December 2025 auction did clear, at its price cap of $333.44 per megawatt-day for unforced capacity (UCAP). What failed was procuring enough capacity to meet the reliability requirement, which came up 6,623 MW short in what PJM itself calls "the first auction in which the entire RTO, including Fixed Resource Requirement areas, fell short of the reliability requirement."⁴ That is more precise, and, as the next auction showed, more alarming than a one-time miss.

PJM's July 2026 auction for the 2028/29 delivery year procured 138,318 MW UCAP, cleared at a new, lower cap of $325 per megawatt-day, and still fell 6,831 MW short of the reliability requirement.⁵ In PJM's own words, "these most recent auctions were the first in PJM history in which the entire RTO fell short of the reliability requirement."⁵ That makes two in a row: December's 6,623 MW miss, followed by a wider 6,831 MW miss seven months later, even as the auction procured more capacity at a lower price.

Line the two auctions up and the pattern sharpens. Each cleared supply worth $16.4 billion.⁴, ⁵ The second bought more capacity, 138,318 MW against 134,479 MW, at a lower price, $325 against $333.44, and still missed by a wider margin.⁴, ⁵ The cap changed the price. It did not touch the shortfall, and the supply response has been thin: the July auction cleared only 525 MW UCAP of new generation and uprates, against a gap now measured in the thousands.⁵

PJM's July 2026 auction bought more capacity than December's at a lower price cap, and still missed its reliability target by a wider margin.

Source: PJM Interconnection news releases, December 17, 2025 and July 14, 2026.

PJM is not waiting for the next auction to fix this. On July 31, 2026, it filed a Reliability Backstop Procurement with FERC, docket ER26-3380-000: a one-time, up-to-15-year procurement sized to the combined shortfall and aimed, as PJM itself put it, at "the unprecedented addition of large loads, most notably data centers."⁶ The filing's maximum willingness to pay is a MW-weighted average of $555 per megawatt-day.⁶ That figure is not arbitrary. PJM ran the same auction without the collar and published the result: every zone cleared at $554.72 except ComEd, which separated to $776.69.⁷ PJM cautions that the simulation reused existing seller offers and cannot know how those offers would have changed in an uncapped auction.⁷

PJM's July 2026 auction cleared at its $325 price cap. The same auction, simulated without the cap, would have cleared at $554.72 in most zones and $776.69 in ComEd. PJM's backstop filing set its own maximum willingness to pay at $555, a bidding ceiling rather than a result. FERC suspended the procurement September 29; the bid window never opened, and no results are expected on the original December 2 schedule.

Source: PJM Interconnection, July 14 and July 31, 2026 filings; FERC order, September 29, 2026.

The cap did not lower what the shortfall costs. It moved the cost out of the annual auction and into a separate 15-year contract priced almost exactly where the uncapped market would have landed. FERC acted on schedule, but not the way PJM proposed. On September 29, it accepted the backstop procurement in substance while suspending its effective date for five months, to February 28, 2027, subject to refund, pending further proceedings on three pieces it flagged as potentially unjust and unreasonable: cost allocation, the rules for a transmission owner that exits PJM, and collateral requirements for load-serving entities, which for one cooperative ran to roughly $2 billion.⁸ PJM pulled the September 30 bid window. It has said it will refile a revised proposal on October 29 but, as of this writing, has not set a new bid window.⁸ The fight over who pays for the shortfall, the question this filing was supposed to settle, is now itself a five-month proceeding.⁸ Our July 2026 deep dive into the December auction's cost mechanics covers how that first shortfall became a $16.4 billion bill. This is the sequel, where the bill comes due a second time, gets a 15-year payment plan, and now even that plan is on hold.⁹

A third data point adds texture to the shortfall, and it deserves more precision than it has gotten in circulation. Currence, formerly Sightline Climate, reported in February 2026 that at least 16 GW of data center capacity was slated to come online in 2026 across roughly 140 large tracked projects, of which only about 5 GW was under construction. The remaining 11 GW sat in early announcement with no visible groundbreaking.¹⁰ Currence put 30–50% of that pipeline at risk of missing its 2026 timeline, and supplied the number that tells you whether that's alarming: in 2025, 26% of expected capacity slipped and another 10% of projects pushed back.¹⁰ A 30–50% at-risk share, against that base rate, is a materially worse year, short of a collapse.

Of the 16 GW of data center capacity tracked for 2026, about 5 GW