Nearly half of US data centers planned for 2026 are facing delays or cancellation

Sightline Climate estimates 30–50% of US AI data-center projects planned for this year will be delayed or canceled, reflecting major supply and power constraints. Of ~32 GW cited for pre-2027 deployment (16 GW slated before end-2026 + 16 GW announced), only ~5 GW is under construction and typical builds take 12–18 months; projects >25 GW are scheduled for 2027 but <10 GW are being built. Key bottlenecks are shortages of electrical hardware (transformers, batteries) and insufficient grid power, forcing continued reliance on Chinese components despite US tariffs and reshoring efforts.

Analysis

The current bottleneck is shifting economic rents away from hyperscalers and into a narrow set of electrical-equipment and interconnection providers; that shift is likely to persist for multiple quarters because lead-times on heavy electrical kit and permitting cycles are structurally longer than semiconductor build-times. Expect supplier order books and utility interconnection queues to become the primary pacing indicators for AI capex rather than headline demand for chips — that changes where pricing power and margin expansion show up (equipment OEMs and regulated utilities, not GPU makers).

Regionalization will create two distinct markets: fast-to-interconnect corridors where developers pay up for guaranteed capacity, and slow-growth markets where projects stall and real estate valuation for data-center-adjacent land and brownfield campuses re-rates downward. This bifurcation benefits players who monetize interconnection speed (utilities, grid-services vendors, turnkey electrical contractors) and penalizes landlords whose valuation depends on a steady lease pipeline.

Timing and catalysts are concrete and trackable over the next 3–24 months: transformer lead-times, queue movement in major ISO/RTO interconnection lists, utility rate-case decisions on cost recovery, and quarterly capex guidance from the hyperscalers. A derivatives-aware approach — buying optionality on vendors and putting asymmetric hedges on operators — captures upside if orders convert while limiting exposure if policy or rapid reshoring eases constraints faster than expected.

Key Decisions for Investors

• Long Eaton (ETN) — buy shares or 12–18 month calls sized 2–4% portfolio. Rationale: direct exposure to higher-margin electrical distribution equipment and shorter replacement cycles as interconnection orders convert. Risk/reward: target 20–30% upside on visible backlog translating to revenue within 12–24 months; downside ~10–15% on execution or cyclical pullback.
• Long Tesla Energy (TSLA.Energy) or Bloom Energy (BE) — selective 6–24 month exposure to grid-scale storage/backup power suppliers. Rationale: independent onsite generation and storage become premium as grids constrain new builds. Risk/reward: asymmetric upside from project wins and long-term contracts; downside is project execution and commodity input inflation.
• Pair trade: short Digital Realty (DLR) vs long Eaton (ETN) — 3–9 month horizon. Rationale: DLR has valuation tied to forward build/bookings that are most exposed to project delays, while ETN captures vendor rents if projects resume. Risk management: size short to 1–2% with bought puts on DLR and protective calls on ETN; expected mid-single to low-double-digit P/L swing if delay headlines continue.
• Insurance trade on semiconductor demand: buy 3–6 month NVDA puts (small allocation, <1% portfolio) as tail-hedge. Rationale: sustained build delays materially compress near-term GPU drawdown and could shock sentiment/pricing. Risk/reward: low premium for asymmetrical protection against a demand re-rate; expiration should align with next two earnings cycles.