Orban LabsOrban Labs
JournalMapAbout
Supporting note · AI x Energy

US Data Center Delays: Half of 2026 Pipeline at Risk, Only 5 GW Under Active Construction

Wood Mackenzie and others now show that of ~12 GW of US data center capacity expected to come online in 2026, only about 5 GW is under active construction. Close to half of the planned 2026 pipeline is delayed or canceled - driven primarily by the unavailability of transformers, switchgear, and batteries.

May 25, 2026 · 3 min read

Summary

Industry data (citing Wood Mackenzie and other sources) now estimates that only about one-third of the ~12 GW of US data center capacity expected to come online in 2026 is under active construction - roughly 5 GW out of 12 GW. Close to half of the originally planned 2026 builds are delayed or canceled. The primary cause is electrical equipment unavailability - large power transformers, switchgear, batteries - not chips, not turbines, not capital. China transformer imports surged from fewer than 1,500 units in 2022 to more than 8,000 units in 2025 through October (Wood Mackenzie data cited by Bloomberg), reflecting US developers reaching outside North American supply for parts.

Headline Numbers

  • 2026 US DC capacity originally planned: ~12 GW
  • Under active construction: ~5 GW (roughly one-third of planned)
  • Delayed or canceled: close to half of planned
  • China transformer imports: 1,500 units (2022) → 8,000+ units (2025 YTD October)
  • Transformer lead times: 24-30 months (pre-2020) → up to 5 years (now)
  • Typical AI data center deployment cycle: under 18 months - much shorter than transformer lead times

What’s Causing the Gap

  1. Large power transformer shortage - primary bottleneck (covered in transformer-lead-times-4-5-years)
  2. Switchgear and breaker delays - similar long-lead supply constraints
  3. Battery / energy storage equipment - same global supply pinch
  4. Interconnection queue length - ERCOT 410 GW backlog, PJM and SPP equally long
  5. Permitting and emissions disputes - particularly for behind-the-meter gas (see microsoft-chevron-pecos-jeti-tax, meta-hyperion-7gw-expansion)

Conclusions

The gap between hyperscaler capex commitment ($725B in 2026) and physical delivery ability (~5 GW of new US capacity actually building) is now publicly audited. The 2x-3x oversubscription of the order book is no longer an analyst projection or an ERCOT queue overhang - it shows up in Wood Mackenzie’s project tracker and in Cleanview’s monitoring. The delivery rate is the binding constraint on AI infrastructure scale through at least 2028.

This is also why behind-the-meter (BTM) is accelerating (see williams-project-neo-682mw, meta-hyperion-7gw-expansion, microsoft-chevron-pecos-jeti-tax). A BTM project orders its own transformers and switchgear as an integrated package with its on-site generation. It does not have to compete for utility-side LPT slots through grid interconnection.

Our Thinking

The “half of planned 2026 capacity delayed or canceled” claim was already circulating in March via Sandstone Group and other analyst notes (covered in Ep4’s transformer-shortage-deepens). What’s new here is that Wood Mackenzie is now backing the directional read with project-level tracking and import data, and Bloomberg is amplifying it. This is the gap going from analyst memo to mainstream finance press in the window 2026-04-20 to 2026-05-25.

What this means for the narrative arc: the central question of the series (“can the system be built fast enough?”) now has a publicly visible empirical answer for the 2026 window - no. The next questions are:

  1. How much of the planned 2027 capacity also slips? Q3 / Q4 print disclosures from hyperscalers and OEMs will reset that read.
  2. What is the BTM share of actual delivered capacity in 2026 vs grid-connected? This is the key variable for the BTM-as-default hypothesis (H4).
  3. Does the China transformer import surge get tariffed or restricted? This would tighten the supply ceiling further.

For Roman’s lens: this is the most important single dataset in the entire energy + AI story right now. The “$725B capex / 5 GW delivered” ratio is the cleanest empirical proxy for the structural gap between commitments and capacity.

Watch

  • Wood Mackenzie project tracker updates - quarterly
  • Cleanview construction-tracking data
  • Hyperscaler explicit delivery guidance (Q2 / Q3 2026 prints)
  • Any China transformer tariff action
  • DOE Large Power Transformer initiative milestones
  • Marginal BTM vs grid-connected share of 2026 actuals (end-of-year retrospective)
Orban LabsOrban Labs
← AI x Energy
Supporting note · AI x Energy

US Data Center Delays: Half of 2026 Pipeline at Risk, Only 5 GW Under Active Construction

Wood Mackenzie and others now show that of ~12 GW of US data center capacity expected to come online in 2026, only about 5 GW is under active construction. Close to half of the planned 2026 pipeline is delayed or canceled - driven primarily by the unavailability of transformers, switchgear, and batteries.

May 25, 2026 · 3 min read

Summary

Industry data (citing Wood Mackenzie and other sources) now estimates that only about one-third of the ~12 GW of US data center capacity expected to come online in 2026 is under active construction - roughly 5 GW out of 12 GW. Close to half of the originally planned 2026 builds are delayed or canceled. The primary cause is electrical equipment unavailability - large power transformers, switchgear, batteries - not chips, not turbines, not capital. China transformer imports surged from fewer than 1,500 units in 2022 to more than 8,000 units in 2025 through October (Wood Mackenzie data cited by Bloomberg), reflecting US developers reaching outside North American supply for parts.

Headline Numbers

What’s Causing the Gap

  1. Large power transformer shortage - primary bottleneck (covered in transformer-lead-times-4-5-years)
  2. Switchgear and breaker delays - similar long-lead supply constraints
  3. Battery / energy storage equipment - same global supply pinch
  4. Interconnection queue length - ERCOT 410 GW backlog, PJM and SPP equally long
  5. Permitting and emissions disputes - particularly for behind-the-meter gas (see microsoft-chevron-pecos-jeti-tax, meta-hyperion-7gw-expansion)

Conclusions

The gap between hyperscaler capex commitment ($725B in 2026) and physical delivery ability (~5 GW of new US capacity actually building) is now publicly audited. The 2x-3x oversubscription of the order book is no longer an analyst projection or an ERCOT queue overhang - it shows up in Wood Mackenzie’s project tracker and in Cleanview’s monitoring. The delivery rate is the binding constraint on AI infrastructure scale through at least 2028.

This is also why behind-the-meter (BTM) is accelerating (see williams-project-neo-682mw, meta-hyperion-7gw-expansion, microsoft-chevron-pecos-jeti-tax). A BTM project orders its own transformers and switchgear as an integrated package with its on-site generation. It does not have to compete for utility-side LPT slots through grid interconnection.

Our Thinking

The “half of planned 2026 capacity delayed or canceled” claim was already circulating in March via Sandstone Group and other analyst notes (covered in Ep4’s transformer-shortage-deepens). What’s new here is that Wood Mackenzie is now backing the directional read with project-level tracking and import data, and Bloomberg is amplifying it. This is the gap going from analyst memo to mainstream finance press in the window 2026-04-20 to 2026-05-25.

What this means for the narrative arc: the central question of the series (“can the system be built fast enough?”) now has a publicly visible empirical answer for the 2026 window - no. The next questions are:

  1. How much of the planned 2027 capacity also slips? Q3 / Q4 print disclosures from hyperscalers and OEMs will reset that read.
  2. What is the BTM share of actual delivered capacity in 2026 vs grid-connected? This is the key variable for the BTM-as-default hypothesis (H4).
  3. Does the China transformer import surge get tariffed or restricted? This would tighten the supply ceiling further.

For Roman’s lens: this is the most important single dataset in the entire energy + AI story right now. The “$725B capex / 5 GW delivered” ratio is the cleanest empirical proxy for the structural gap between commitments and capacity.

Watch