Scaling the Grid for the Next Era of AI and Data Centres

The growth of AI, cloud services, edge computing and real-time analytics is driving a dramatic surge in data-centre power demands. Many facilities are scaling from tens of megawatts (MW) to hundreds of MW — even approaching the gigawatt (GW) level. This transformation brings with it immense infrastructure, supply-chain, regulatory and grid-integration challenges.

The Scale of the Challenge

Industry research forecasts astonishing growth: according to Deloitte, the power demand from AI-driven data centres in the U.S. could grow more than 30-fold, reaching 123 GW by 2035 (from 4 GW in 2024). Deloitte Another report from Goldman Sachs indicates that data-centre power consumption may increase by 165% by 2030. Goldman Sachs For perspective: traditional large data centres have drawn tens of MW; now operators are designing campuses capable of delivering 500 MW to 1 GW+ of power and cooling capacity. McKinsey & Company

Infrastructure Implications

Scaling power supply to this magnitude involves multiple infrastructure facets:

• Generation & connection: New substations, grid-tie transmission lines, step-up transformers, high-voltage feeders. Energy scarcity, permitting delays and grid bottlenecks are major risk factors.

• Cooling & thermal management: With rising power comes rising heat loads. The cooling infrastructure must scale accordingly — cooling tens or hundreds of MW of IT load continuously.

• Power delivery architecture: Traditional 480 VAC may not be optimal. Newer designs like 400/800 VDC or high-voltage AC are gaining attention to reduce conversion losses, simplify distribution and support high-density racks. For example, ABB and NVIDIA recently announced development of high-efficiency scalable power delivery for next-gen AI data centres. ABB Group

• Site selection and build-time: Obtaining the land, grid access and permits for GW-scale facilities is complex and time-consuming. McKinsey notes that these campuses must evolve from tens of MW to hundreds of MW within five years. McKinsey & Company

Key Challenges & Bottlenecks

• Grid constraints and local power availability: Many geographic regions cannot support additional large loads. This can lead to delays or incremental build-outs.

• Heat and cooling synergy: A 1 GW data-centre campus may require hundreds of MW of cooling capacity, plus dealing with heat rejection, water use (if used), or alternative cooling mediums.

• Operational reliability: Ensuring redundancy, power quality, backup generators, uninterruptible power systems (UPS), and coordinating with thermal management becomes increasingly complex at this scale.

• Energy efficiency and sustainability: As power demand soars, scrutiny increases from regulators, enterprises and sustainability officers. Efficiency (PUE), carbon footprint, water usage, renewable sourcing all matter.

• Time to build vs demand ramp: If build-out takes too long, demand (for compute capacity) may outpace supply of power/infrastructure, causing capacity shortfalls or inflated costs.

Strategies for Mitigation

Operators and infrastructure providers are adopting strategies such as:

• Modular power-delivery and cooling building blocks: Prefabricated, pre-tested modules that can be “plugged in” to accelerate deployment.

• Closer coupling of cooling and power design: Designing power infrastructure in tandem with cooling systems (thermal, airflow, liquid-cooling) rather than as separate silos.

• Leveraging renewables, onsite generation and energy storage: To reduce dependency on grid supply and provide flexibility (though this is addressed further in another article).

• Future-proofing for expansion: Designing campuses with headroom for growth, reserved grid connections, oversized substations, and cooling loops designed with modular expansion in mind.

• Demand-side management and shift loads: Using AI and workload scheduling (covered in the next article) to optimise when and how compute loads draw power and how cooling can be coordinated.

Why This Matters for Cooling Power

Large-scale power delivery and cooling are inseparable at GW-scale facilities. As power input increases, heat rejection requirements grow, capacity risks rise, and the coupling of power & cooling becomes tighter. A mis-matched cooling system can bottleneck compute, reduce uptime or inflate costs. Thermal strategy must evolve in lock-step with power strategy.

Closing Thoughts and Outlook

As data-centre power demands creep into the gigawatt realm, only those operators with holistic infrastructure strategies (power + cooling + site + sustainability) will be able to scale cost-effectively. Vendors, utilities and regulators must also step up. The next decade will test the agility of both developers and the grid.

Serge Boudreaux – AI Hardware Technologies, Montreal, Quebec
Peter Jonathan Wilcheck – Co-Editor, Miami, Florida

References

1. “AI infrastructure gaps | Deloitte Insights”, Deloitte. https://www.deloitte.com/us/en/insights/industry/power-and-utilities/data-center-infrastructure-artificial-intelligence.html Deloitte

2. “AI to drive 165% increase in data center power demand by 2030”, Goldman Sachs. https://www.goldmansachs.com/insights/articles/ai-to-drive-165-increase-in-data-center-power-demand-by-2030.html Goldman Sachs

3. “Building data centres bigger, faster | McKinsey”, McKinsey & Company. https://www.mckinsey.com/industries/private-capital/our-insights/scaling-bigger-faster-cheaper-data-centers-with-smarter-designs McKinsey & Company

4. “ABB to develop next-generation AI data centres with NVIDIA”, ABB News. https://new.abb.com/news/detail/129805/abb-to-develop-next-generation-ai-data-centers-with-nvidia ABB Group

5. “The New Texas Energy Barons”, D Magazine. https://www.dmagazine.com/publications/d-ceo/2025/november/the-new-texas-energy-barons/ D Magazine