Schroders Greencoat Pivots to AI Data Centers as Power Becomes the Binding Constraint on Compute
Institutional renewable capital is flowing toward AI infrastructure rather than grid decarbonization, exposing electricity access—not chips or algorithms—as the bottleneck limiting AI deployment.
Schroders Greencoat is targeting data center-linked renewable assets as AI power demand surges, marking a strategic shift in institutional energy investment away from traditional climate mandates toward compute infrastructure.
The firm launched a green digital infrastructure platform in March 2026 through a 50:50 joint venture with Greencoat Renewables, with its first investment supporting a 36MW data center at Drogheda Energy Park in Ireland. The facility will integrate on-site generation, storage, and grid services—architecture designed to bypass rather than depend on congested grid connections.
“There is an exciting opportunity in integrating data centre demand with Renewable Energy generation to support the rapid growth of AI driven digital infrastructure,” said Paul O’Donnell, Partner at Schroders Greencoat.
Power, Not Silicon, Defines the Bottleneck
Data center electricity consumption reached 415 TWh globally in 2024—1.5% of world electricity—and has grown at 12% annually since 2017, according to the Brookings Institution. AI-focused facilities drove consumption up 50% in 2025 alone, per the International Energy Agency. Lawrence Berkeley National Laboratory projects U.S. data center demand will rise from 176 TWh in 2023 to between 325 and 580 TWh by 2028—potentially 12% of total national consumption.
Modern AI chips draw 700W to 1,200W per processor compared to 150W-200W for standard CPUs, requiring 30-80 kW per rack versus 8-15 kW for conventional Data Centers. Nvidia’s Rubin GPU system, launching later in 2026, will demand around 300 kW per rack, with industry projections approaching one megawatt per rack within 18 months.
According to Center for Strategic and International Studies analysis, access to electricity supply is the binding constraint on expanded computational capacity and therefore on continued U.S. leadership in AI.
High-voltage transformer lead times have extended from 24-30 months pre-2020 to five years today, creating a critical bottleneck for data center projects, according to analysis from Tech Investments. Grid connection timelines now stretch 4-10 years in congested markets, while AI data center planning cycles operate on 2-3 year horizons—a structural mismatch driving demand for off-grid or grid-adjacent solutions.
Sightline Climate tracked 12 GW of announced U.S. data center capacity for 2026, but only 5 GW is under construction. The remaining 11 GW sits in the ‘announced’ stage, awaiting power infrastructure that may not materialize for years.
Capital Reallocation Creates Two-Tier Energy Markets
Renewable investment capital is now flowing toward compute-dense regions rather than following decarbonization mandates. The IEA projects renewables will grow 22% annually through 2030, meeting nearly 50% of data center electricity demand growth—but this capacity is being contracted directly by hyperscalers and AI developers rather than utilities serving general grid loads.
Power purchase agreement prices rose 35% in 2024, driven largely by AI developer procurement. This bidding competition creates geographic clustering: operators with existing power capacity, renewable PPAs, and grid connectivity gain structural advantages in hosting AI compute, while regions dependent on greenfield infrastructure face multi-year delays that effectively exclude them from frontier model training.
About 30% of power flowing into data centers is not used for computation—it’s consumed by cooling systems and lost in long-distance transmission. This inefficiency, combined with exponential growth in per-chip power draw, is driving architectural redesign toward distributed, on-site generation models that eliminate transmission losses and reduce cooling loads through proximity to renewable sources.
Schroders Greencoat’s pivot exemplifies this shift. “More recently, we’ve also made investments into platforms to help develop sites for data centers. We’ve got as energy specialists a real role to play in that space,” Duncan Hale, Portfolio Manager at Schroders Greencoat, told Bloomberg.
Geopolitical Asymmetry in AI Capacity
The power constraint is creating geopolitically asymmetric advantages. Nations with surplus renewable capacity—Norway’s hydropower, Middle Eastern solar, or Texas wind—can offer immediate grid access that regions with constrained or aging infrastructure cannot match. This dynamic is fragmenting AI compute capacity along energy rather than technological lines.
Electricity access now defines competitive position in AI development more than chip manufacturing or algorithm research, according to the Center for Strategic and International Studies. Elon Musk stated in recent comments that “very soon, maybe even later this year, we’ll be producing more chips than we can turn on.”
The shift toward localized, AI-adjacent renewable projects represents a structural break from traditional energy transition investment. Where renewable capital once chased policy subsidies and carbon reduction mandates, it now follows compute clustering patterns—optimizing for latency, grid independence, and power density rather than emissions reduction. This reallocation exposes a new reality: in the AI economy, electricity access is not a commodity input but a source of competitive moat.
What to Watch
Monitor announcements of co-located renewable-plus-data-center projects, particularly in regions with surplus baseload capacity. Track high-voltage transformer production capacity expansions—if lead times remain at five years, the power bottleneck will persist through 2030 regardless of chip or model advances. Watch for regulatory reform around grid connection timelines; jurisdictions that streamline permitting will attract disproportionate AI infrastructure investment. Finally, observe whether hyperscalers begin acquiring utility-scale generation assets directly rather than relying on PPAs—a signal that vertical integration of power supply is becoming strategically necessary for AI deployment at scale.