Back to briefings
Download PDF
US Data Center Briefing · December 26, 2025

December 26, 2025

Hyperscale power procurement driving new thermal + renewables + storage proposals Rising local opposition and bill-impact scrutiny for data-centre-linked generation Grid stability solutions for 500 MW–2 GW load loss events gaining attention Policy and grant-funding volatility complicating clean power and grid upgrade timelines Distributed energy zones (South Korea) explicitly targeting AI/data-centre loads

Market overview (Global | 26 Dec 2025)

AI-driven compute demand continues to pull forward investment in power, grid interconnection, and network infrastructure, with several datapoints underscoring the tight coupling between hyperscale buildouts and local energy/policy dynamics. Today’s flow is dominated by (i) power-supply and grid-stability considerations for gigawatt-scale loads, (ii) emerging national and local policy friction around environmental impacts and permitting, and (iii) enabling infrastructure (fiber and fluids) positioning for the data-centre value chain.

Risks and watchpoints (near-term)

Key deals & strategic moves

Energy/industrial adjacency

  • BP / Stonepeak — Castrol JV: BP agreed to sell 65% of Castrol to Stonepeak for $6bn, valuing Castrol at $10.1bn (Stonepeak 65%, BP 35%). Castrol’s strategy includes diversifying into data centre fluids alongside mobility/industrial lubricants (BP sells 65% of Castrol to Stonepeak for $6B).

Digital infrastructure (connectivity enabling data centres)

  • Indonesia — FiberCo JV: Indosat Ooredoo Hutchison signed an investment agreement with Arsari Group and Northstar Group to spin off fiber assets into FiberCo, valued at ~IDR 14.6tn (~$869.87m). Indosat retains ~45%. Asset footprint: 86,000 km integrated fiber (backbone, domestic subsea, access) on an open-access wholesale model; coverage cited at ~45% of Java and ~55% of non-Java regions (Indosat, Arsari and Northstar form FiberCo to expand fiber nationwide).

Power & grid / interconnection highlights

US (Arkansas): thermal + renewables + storage package tied to hyperscale load

  • Entergy proposal: Jefferson Power Station (745 MW); paired project: Cypress Solar (600 MW solar + 350 MW battery), described as serving Google’s West Memphis data center; customer affordability impact flagged (bill increases up to $20/month) (Arkansas’s Most Notable Farm and Environment Stories 2025).

South Korea: distributed energy “special zones” with explicit data-centre linkage

  • The Ministry of Climate, Energy and Environment designated three new distributed energy special zones: Pohang, Ulsan, Seosan (total now seven). Planned projects include:

Bulk power system resilience (research signal)

  • Substation-level stabilization concept: high-voltage (345 kV) circuit-breaker-operated braking resistors inserted for 0.25–0.85 seconds to stabilize frequency/voltage following sudden loss of 500 MW–2 GW data-centre loads; positioned as scalable for gigawatt-scale clusters (High-voltage braking resistors to stabilize gigawatt data centers).

Policy & regulation / public-sector demand signals

United States

  • DOE Genesis Mission: DOE launched the Genesis Mission to build an integrated AI platform across 17 national laboratories, targeting an initial operating capability within 270 days and milestones at 60/90/120/240/270 days. Collaboration agreements cited with 24 organizations, including AWS, Google, Microsoft, NVIDIA, OpenAI, Anthropic, IBM, Intel, AMD and xAI—potentially supportive of incremental public-sector compute demand and associated infrastructure requirements (DOE launches Genesis Mission to integrate AI across labs).
  • Environmental policy uncertainty: Reporting cites EPA rule rollbacks, termination of a local EPA employees’ union in Chicago, and cancellations of ~$150m rooftop solar and ~$600m methane/grid grants, while Illinois officials grapple with AI data-centre energy/water demand and a stalled environmental-justice ordinance (Trump, AI’s thirst and Mayor Johnson’s stalled environmental ordinance).

Philippines

Technology & efficiency notes (operators / supply chain implications)

  • PUE forecasting (research): A Bidirectional GRU (BiGRU) model was developed to predict data-centre PUE using an EnergyPlus-simulated Singapore dataset with RFECV feature selection; performance compared versus a GRU baseline (MSE/MAE/R²) and presented at IJCNN 2025 (BiGRU neural model predicts data center PUE).
  • Multi-datacenter networking for AI training (research): A framework for pipeline-parallel LLM training across multi-datacenter optical networks reports 31.25% lower iteration time and 13.20% fewer blocking requests vs baselines—relevant to campus-to-campus fiber capacity planning and latency-sensitive design (Framework optimizes LLM training over multi-datacenter optical networks).

What to watch

Subscribe to Data Centers Briefings

Get the briefing in your inbox. Free. One email a day.

Region