Warning Signs in the 7,481 MW Data Center Construction Frenzy: The Real Focus Should Be on Room for Supply-Side Rollout

Author: Christopher Johnson (Avanza Energy)

Compiled by: TechFlow

TechFlow Briefing: As the generative AI arms race intensifies its transmission to underlying energy infrastructure, North America’s 7,481 MW of under-construction data centers in H1 2026 has shattered historical records, painting a picture of booming prosperity. Yet at the very top of the industrial chain, project capacity newly entering planning pipelines has plunged for three consecutive quarters. These seemingly contradictory data points precisely map a deep fault line in the computing infrastructure cycle: the industry is consuming existing groundbreaking projects at an unprecedented pace, while backup grid-connection resources have been cornered by grid approval bottlenecks and power supply deficits.

As the core grid hosting the nation's largest data center cluster, PJM recorded reliability supply deficits in the tens of megawatts across two consecutive capacity auctions. Administrative price caps not only failed to resolve physical power shortages but were forced to generate a massive, 15-year “reliability backstop procurement” plan at a staggering unit cost of $555 per MW-day, transferring enormous financial burdens onto all ratepayers.

With only six months’ worth of pre-leaseable supply remaining at core national hubs, the next hard constraint on tech giants' expansion has fundamentally shifted from server procurement to physical grid-interconnection capacity.

Data center construction scale across North America’s eight major markets has just hit a new all-time high: under-construction capacity reached 7,481 MW (megawatts) in H1 2026, surging 24.8% year-over-year and surpassing the previous record set at the end of 2024.¹

Yet merely four months prior, Wood Mackenzie reported the exact opposite trend: newly planned pipeline projects declined for three consecutive quarters, plummeting 19% in Q1 2026 alone.²

Both datasets are entirely accurate and non-contradictory. They measure opposite ends of the same development pipeline, and the gap between them reveals the true industry trajectory for H2 2026 far better than any single sensational headline could.

CBRE tracks projects that have already broken ground; Wood Mackenzie counts newly added projects just entering the planning pipeline.

If the digestion rate of existing backlogged projects outpaces the replenishment of new ones, a market can absolutely break both “record-high under-construction volume” and “plummeting new-project growth” metrics within the same reporting period—the data presents exactly this reality.

Across all major North American markets, available pre-lease capacity stands below 1,500 MW. At current demand absorption rates, this equates to roughly six months of supply inventory.¹

Gordon Dolven, Head of Data Center Research at CBRE, explicitly stated upon releasing the H1 2026 report: “The next phase of growth bottlenecks will no longer be driven by demand, but by where new power supply can actually land and interconnect.”¹

The Two Ends of the Same Pipeline

Currently, over 80% of all under-construction capacity has been pre-leased, up from 74.3% a year ago; compared to the end of 2025, market vacancy rates remain stable at a historically low 1.4%.¹

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

Northern Virginia retains the lead in total inventory and posted the highest net absorption across all markets at 467.7 MW, but its vacancy rate has fallen to 0.2%. Every MW slated for delivery in 2026 is fully pre-booked, forcing all new leasing negotiations to shift entirely to 2027.¹

Listed rents have risen across all deployment sizes, with mid-to-large tenants requiring 3–10 MW seeing premiums surge by 8.3%.¹ This is undeniably a market pricing in extreme supply scarcity, not weak demand.

Wood Mackenzie’s data corroborates this supply squeeze from another angle. Newly disclosed pipeline capacity dropped to 36 GW in Q1 2026, down 19% quarter-over-quarter, which the firm noted “continues a three-quarter streak of sequential moderation”.²

The strongest evidence proving this is structural divergence rather than mere industry cooling lies in the very next line of the report: capacity in advanced commercial negotiation stages (the final step before signing formal agreements) surged nearly two-fold in a single quarter, jumping from 37 GW to 107 GW.²

High-certainty committed capacity backed by signed formal construction or power purchase agreements climbed to 195 GW, with disclosed project capex totals crossing the $1 trillion mark for the first time in history.²

A purely declining market could never mature its existing pipeline so rapidly. Exactly such a market—one where backlog releases accelerate while brand-new approvals face headwinds—produces precisely this appearance.

Figure: Projects newly entering planning pipelines continue to contract, while volumes of groundwork-breaking projects within existing pipelines break records. Together, they accurately depict the same batch of backlogged assets. Source: CBRE press release (August 27, 2026); Wood Mackenzie press release (April 2026).

A Candid Correction

To some extent, this piece also serves as a necessary correction to our prior analysis. In our April 2026 deep dive on the data center construction boom, we noted a year-over-year decline in U.S. under-construction capacity for the first time since 2020, and characterized PJM’s December 2025 capacity auction as failing to clear the market.³ In light of subsequent developments, neither statement holds up rigorously.

CBRE’s official summary perfectly addresses the first point: “After construction pacing was briefly dampened by project delays stemming from power supply shortages at the end of last year, new groundbreaking activity rebounded strongly by 24.8%...”¹ The core mechanism we identified then—power bottlenecks stalling groundbreaking—is entirely real.

The only thing that changed is that previously backlogged and delayed projects have finally begun breaking ground and delivering in concentrated waves. On statistical charts, this delayed release simply looks like a record-breaking construction peak.

The wording regarding the capacity auction also requires adjustment. PJM’s December 2025 auction did effectively clear, with transaction prices touching the Unforced Capacity (UCAP) cap of $333.44/MW-day; the actual issue was the failure to procure enough capacity to meet grid reliability targets, resulting in a severe shortfall of 6,623 MW.

PJM officially defined it as “the first time in the organization’s history (including Base Residual Auctions) that it failed to procure sufficient capacity to meet Reliability Requirement standards.”⁴ This correction is more precise, and as revealed by the subsequent auction, it signals a far graver warning than a one-off clearing failure.

Bottlenecks Shift to the PJM Grid

In the July 2026 Forward Capacity Auction covering the 2028/29 delivery year, PJM procured 138,318 MW of UCAP, ultimately clearing at a new price cap of $325/MW-day. Yet a shortfall of 6,831 MW remained against grid reliability requirements.⁵ As PJM’s official notification stated: “These auctions represent the first instance in PJM’s history where the entire region failed to meet Reliability Standards.”⁵

Thus, PJM has suffered consecutive backfires twice: following a 6,623 MW shortfall in December, the deficit widened further to 6,831 MW just seven months later, despite the auction securing more nominal capacity at a lower price ceiling.

Lining up these two auctions side-by-side exposes the underlying dynamic clearly. Both cleared for a total procurement value 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/MW-day), yet the reliability gap paradoxically widened.⁴, ⁵

Administrative price caps artificially suppressed clearing prices but did nothing to resolve physical power shortages, while market supply response was exceedingly weak: in July’s auction, new generation builds and expansions cleared only 525 MW of UCAP, a drop in the bucket against multi-megawatt vacancies.⁵

Figure: PJM’s July 2026 auction procured more capacity than December’s at a lower price cap, but the reliability shortfall widened further. Source: PJM Interconnection press releases (December 17, 2025, and July 14, 2026).

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

This is a special long-term procurement spanning 15 years, sized to completely match the previously accumulated shortfalls. In PJM’s own words, its objective is to address “unprecedented, large-scale increments in electric load, particularly the rapid expansion of data centers”.⁶ The petition sets maximum willingness-to-pay at a weighted average of $555/MW-day.⁶

This figure is certainly not fabricated: PJM simulated removing the price cap and published results showing that without regulatory controls, actual clearing prices reached $554.72/MW-day across all regions except ComEd, where grid congestion pushed prices to $776.69/MW-day.⁷

PJM also prudently cautioned that the simulation relied on current seller bids and could not forecast how sellers would adjust their strategies in a truly unregulated auction.⁷

Figure: PJM’s July 2026 auction cleared at a $325 price cap. Under uncapped simulations, nearly all regions would have cleared at $554.72/MW-day, with ComEd reaching $776.69/MW-day. PJM’s backstop procurement petition sets its maximum willingness-to-pay at $555 (this is a bid cap, not an actual outcome). FERC paused the procurement procedure on September 29; the bidding window never opened, and the originally scheduled October 2 opening was cancelled. Source: PJM Interconnection (July 14 and July 31 documents); FERC Order (September 29, 2026).

Administrative price caps do nothing to reduce the genuine financial toll of the shortfall. They merely strip this colossal cost out of annual spot auctions and shunt it into a bespoke 15-year long-term contract, priced almost precisely at the fully unregulated market equilibrium.

Although FERC issued its ruling on schedule, it did not adopt PJM’s proposal wholesale.

On September 29, FERC substantially acknowledged the necessity of the backstop procurement but delayed its effective date by five months to February 28, 2027 (with refund provisions attached), pending further hearings on three core disputes alleged to be “unjust and unreasonable”: the cost allocation mechanism, rules governing transmission asset owners exiting PJM, and collateral requirements imposed on Load Serving Entities (one power cooperative faces a collateral burden nearing $2 billion).⁸

PJM subsequently withdrew its bidding window originally scheduled to open on September 30; although it indicated it would resubmit a revised proposal on October 29, a new bidding schedule remains unresolved as of this writing.⁸ Originally designed to answer “who pays for the power shortfall,” the backstop measure has instead devolved into a five-month legal and administrative stalemate.⁸

Our July 2026 deep-dive into the December auction cost mechanics meticulously unpacked how the initial shortfall ballooned into a $16.4 billion bill; everything unfolding now is the sequel to that drama: another colossal invoice arrives, packaged as a 15-year installment plan, but even this repayment schedule has ground to a halt.⁹

The Overstated Narrative of Project Cancellations

The third dataset provides vivid micro-level context on the shape of power shortages, yet it has evidently been oversimplified as it circulated through the market.

In February 2026, industry think tank Currence (formerly Sightline Climate) reported that among the roughly 140 large-scale projects it tracks, at least 16 GW of data center capacity was originally scheduled to grid-connect and deliver in 2026. However, only about 5 GW is actually under construction, with the remaining 11 GW still stuck in early paper-planning stages, showing zero signs of groundbreaking.¹⁰

Currence assesses that 30% to 50% of pipeline capacity faces the risk of missing 2026 deadlines. Yet the firm simultaneously provides a crucial historical benchmark to contextualize whether this is truly fatal: in 2025, 26% of expected delivery capacity experienced delays, with another 10% of projects completely pushed back.¹⁰

Relative to this historical baseline, a 30–50% risk exposure undeniably signals a significantly worse year, but it falls far short of the cliff-like market collapse suggested elsewhere.

Figure: Of the 16 GW of data center planning capacity tracked for 2026, only ~5 GW is actually under construction. Currence estimates 30–50% of total pipeline capacity risks missing 2026 timelines. While higher than 2025’s 26% delay rate, it is hardly the shocking media narrative claiming “half of 2026 projects were completely cancelled.” Source: Currence, “Data Center Industry Outlook” (February 24, 2026).

Fighting the viral industry media claim of “half of 2026 projects canceled,” prominent research outfit SemiAnalysis directly questioned the 5 GW denominator used for tracking construction, rather than debating qualitative interpretations: “Just counting the top two hyperscale cloud providers, their actual under-construction capacity alone vastly exceeds 5 GW.”¹¹

If the statistical denominator severely underestimates real-world construction volume, then the purported risk ratio naturally contracts sharply as well. Those truly exposed to default risk are almost exclusively early-stage paper plans that haven’t broken ground, which aligns precisely with the fastest-contracting layer in Wood Mackenzie’s pipeline data.²

Juxtaposing these two studies yields an impartial conclusion: a substantial portion of 2026 pipeline projects will indeed fail to materialize on original schedules, but schedule delays are not outright cancellations, yet headlines deliberately twist “delay” into “cancelled”.

The Deeper Meaning Behind the Data

Three disconnected datasets, viewed in isolation, seem to tell three unrelated stories: construction booms, new approvals stagnating, and PJM region grids buckling under strain.

Yet once pieced together, they point to one stark, unified truth: the industry is burning through existing construction backlogs at record speeds, while fewer new projects are coming online to replenish the pipeline. Simultaneously, grid operators have failed to honor their reliability commitments twice in a row, forcing massive 15-year risk premium payouts in desperate attempts to mask—not genuinely solve—structural supply deficits via administrative price controls.

Six months of pre-leaseable inventory is not an immediate apocalypse, but a countdown timer already ticking. What will ultimately determine the market’s breakout path is not the feverish prosperity splashed across today’s headlines, but the hard data of the coming quarter.