AI workloads are reshaping how data centers are planned and operated.

Traditional data centers were built around **steady CPU workloads** and **predictable power growth**, which made it easier to:

• Secure long-term energy supply
• Use mature, air-based cooling technologies
• Plan capacity with reasonable certainty

AI changes this picture in several ways:

• Much higher power density: AI training clusters demand far more power per rack than typical enterprise workloads. Data center electricity consumption has grown about 12% per year over the last five years, and AI training facilities are a major driver of this trend.
• Faster timelines and larger scale: Developers are now targeting gigawatt-scale campuses, with compressed delivery schedules. That puts pressure on power, cooling, land, and permitting decisions.
• Integrated infrastructure needs: Instead of just “another building,” next-generation AI sites are being designed as industrial campuses that integrate power, cooling, transmission, water, digital systems, and long-term operations from day one.

As a result, organizations are reimagining AI data centers as complex energy and infrastructure systems, not just compute facilities. Success depends on building **resilient, flexible, and energy-optimized campuses** that can keep up with AI demand over time.

Access to power is quickly becoming one of the main differentiators in the AI infrastructure race.

Developers are balancing several options, each with trade-offs:

1. Relying on grid power

• Pros: Typically lower long-term energy costs, strong stability and resilience once connected.
• Cons: Capacity constraints and interconnection timelines that can delay projects by years. In fast-moving AI markets, that delay can mean lost opportunity.

2. Behind-the-meter / on-site generation

• Technologies include gas turbines and reciprocating engines.
• Pros: Faster deployment, more operational control, and the ability to bring capacity online before grid upgrades are ready.
• Cons: Higher upfront capital, higher operating costs, fuel dependencies, and more complex permitting (e.g., air quality and emissions).

3. Hybrid and microgrid-based architectures

• Combine grid supply with on-site generation, battery storage, and often renewables such as solar.
• Coordinated via microgrid controls to manage load variability, island during grid issues, and optimize overall performance.

Because **speed-to-market is now a key competitive driver**, many large AI projects are willing to pay a premium for off-grid or hybrid solutions that can be delivered faster.

One example from Texas illustrates the scale of this shift:

• A gigawatt-scale AI training data center supported by 5 GW of gas generation, up to 1.25 GW of solar PV, utility-scale batteries, and a microgrid serving 20 buildings totaling 10 million square feet, each designed for around 250 MW of power demand.

In this environment, **“time to power” is becoming the new measure of AI infrastructure readiness**, often more critical than traditional metrics like cost per kWh alone.

As AI drives higher power densities, both cooling and permitting are being rethought to keep projects viable and on schedule.

Cooling strategies for high-density AI

• Limits of air cooling: Traditional air-based systems struggle with the heat generated by dense AI clusters, both technically and economically.
• Shift to liquid cooling: Liquid cooling is gaining traction because it transfers heat more efficiently and supports the densities AI workloads require. It does, however, introduce new dependencies on liquid infrastructure and vendor ecosystems.
• Heat reuse and integration: Developers are exploring ways to connect waste heat to nearby industrial processes or district heating networks. In regions like the Nordics, district heating from data centers is already common, turning a cost center into a potential value stream.
• Thermal energy storage: By producing chilled water when electricity is cheaper or more available and using it later, AI data centers can:
  • Shift cooling load away from peak periods
  • Reduce grid draw during system stress
  • Lower demand charges
  Combined with batteries and advanced controls, this helps stabilize both the facility and the surrounding grid.

Permitting as a parallel workstream, not an afterthought

• Permitting and regulation now sit alongside technical design as a core constraint.
• Requirements vary by region and project type, affecting:
  • Grid connection approvals and capacity limits
  • Air quality and emissions permits for on-site generation
  • Land use, environmental approvals, and community acceptance
• Teams that address permitting early—alongside power, cooling, and commercial planning—can structure projects that are both technically sound and realistically deliverable.

In practice, this means bringing together **energy providers, technology companies, developers, regulators, and local communities** from the outset. Early alignment across these stakeholders reduces friction between planning, construction, and operations, helping AI capacity come online at the speed and scale the market now expects.