AI Data Center Site Selection
Choosing a site for a gigawatt-scale AI data center is no longer a real-estate decision. It is an infrastructure planning problem that decides, for twenty years or more, what the campus pays for electricity, how fast it can connect, and how resilient it will be.
IWE AI Infra uses the IWE Multicriteria Site Selection (IMSS) methodology — a structured, systems-engineering approach developed from our work siting wind and solar projects and, more recently, planning a green AI data center hub with the State of Gujarat. The primary criterion is the delivered cost of electricity over the life of the asset: the cost and availability of wind, solar and battery storage, transmission access, and the carbon credits needed to offset residual grid power.
What we evaluate
- Power and energy — proximity to high-capacity transmission (typically 400 kV and above), available interconnection capacity, and the delivered cost of renewable energy.
- Digital connectivity — long-haul fiber, multiple carriers, latency, distance to cable landing stations, and redundant network paths.
- Water and cooling — reliable water supply, or suitability for water-efficient, closed-loop cooling.
- Land and geophysics — large contiguous parcels; areas with slopes above 5° are excluded; seismic and geotechnical risk.
- Risk and security — distance from borders and conflict zones, and other long-term security risks.
- Environment and community — environmentally sensitive areas are excluded and social impacts are assessed early.
- Climate resilience and readiness — flood, sea-level rise and extreme-weather exposure; roads, ports and construction supply chains.
Availability zones, not just sites
Hyperscale AI workloads need geographic redundancy. Rather than concentrating capacity in one place, we design triangular availability zones: clusters of campuses close enough to work together, far enough apart to avoid correlated risks. The result is redundant power and network paths, load balancing across campuses, and resilience to local disruptions.
Training campuses and edge sites
Not every AI workload needs a gigawatt campus. Training clusters belong at a few large sites on 400 kV transmission. Inference — the majority of AI compute by volume — can follow the power: edge AI data centers of 10–50 MW placed next to renewable plants on existing 33/132 kV feeders can be built in 12–24 months instead of waiting 4–6 years for a transmission connection, while absorbing renewable energy that would otherwise be curtailed.
Outcome
A site decision that is defensible and bankable — ranked on lifecycle cost and risk across power, connectivity, land and environment, rather than on any single factor or short-term opportunity.
Related reading
