Nxtra’s hyperscale facilities require energy management systems capable of distinguishing thermal loads by cooling mode for billing and Scope 2 reporting
The System Pressure
The conventional cooling assumption — that air-handling units, evaporative towers, and a manageable PUE target can absorb any compute density the market produces — is failing in high-heat, water-stressed geographies. Operators across the Middle East, South Asia, and Southeast Asia report that traditional air-cooling architectures are no longer adequate for the rack densities that AI inference and training workloads demand.
The pressure is not purely thermal. These regions simultaneously face extreme ambient dry-bulb temperatures, regulatory and community pressure on freshwater withdrawal, and government ambitions to become AI compute hubs. That combination forces a decision most energy and infrastructure teams in temperate Western markets have not yet had to make: commit to a cooling architecture before the rack density trajectory is confirmed. Deferring that decision — building for today’s density and retrofitting later — produces a more expensive outcome than designing for flexibility from the start, a point several operators in the source have made explicitly.
The Drivers, Dependencies, and Constraints
Three distinct strategies are emerging across the region, each with a different energy implication.
In Saudi Arabia, DataVolt’s planned AI factory in NEOM is designed to reject the entirety of its heat load to seawater. Seventy percent routes directly through plate heat exchangers; the remaining thirty percent moves through water-cooled chillers before the condenser loop also terminates at the sea. The energy dependency is narrow in one direction — this architecture avoids the compressor loads that evaporative cooling would require at scale in desert conditions — but it introduces a long-cycle capital constraint: biofouling control, marine-grade materials, and coastal thermal loop maintenance are not one-time CapEx items. They compound across asset life and are difficult to cost-model without operational precedent.
In India, Nxtra by Airtel is pursuing portfolio flexibility rather than single-system optimization. Nxtra’s hyperscale facilities require energy management systems capable of distinguishing thermal loads by cooling mode for billing and Scope 2 reporting. Water-stress analysis is integrated into site selection, and the stated target is zero liquid discharge. The operational constraint that follows is significant: a facility running four cooling modes requires energy management systems capable of distinguishing thermal loads by mode, with metering granular enough to inform billing and Scope 2 reporting accurately.
In Southeast Asia, ST Telemedia Global Data Centres is designing new builds around air-cooled chiller plants and closed-loop systems to protect freshwater budgets, while ensuring facility water reaches the perimeter of every data hall for future liquid-cooling conversion. The dependency that creates is architectural: perimeter water infrastructure is a sunk cost that yields value only if tenant density evolves toward liquid cooling within the facility’s economic lifecycle.
Across all three geographies, Gulf Data Hub’s framing of the decision map is analytically useful for energy teams. The four operative tradeoffs are water consumption versus energy efficiency, upfront CapEx versus future density readiness, system complexity versus operational resilience, and sustainability pace versus speed to market. These map directly to budget line items that global energy heads own or influence.
Open Dependencies
Several assumptions embedded in these strategies remain unresolved from the source evidence.
The NEOM seawater rejection system is planned, not yet operational. Long-term thermal performance in a marine environment involves biofouling and corrosion variables that are difficult to model without data from comparable live facilities. Whether the energy efficiency gains from eliminating compressor loads hold across the full operational lifecycle is not established by current evidence.
The zero liquid discharge target in India is a design intent, not a certified operational outcome. Water-stress site analysis improves selection but does not eliminate the risk that local grid and water conditions shift after capital is committed. Nxtra’s multi-mode flexibility also leaves an open question the source does not answer: at what rack density does the facility pivot from air-primary to liquid-primary operation, and which party — operator or tenant — absorbs the transition cost?
STT GDC’s perimeter water model depends on future tenant demand arriving at the right density and the right point in the facility lifecycle to activate embedded optionality. That is a commercially reasonable assumption given AI demand trends, but it is not confirmed in the source as a contracted or scheduled outcome.
The Operating Exposure for Global Heads of Data Center Energy
PUE is the most immediate financial lever. Energy procurement models structured against a 1.4 or 1.5 PUE baseline become mispriced if cooling architecture shifts mid-contract toward denser, lower-PUE configurations. PPAs and utility tariffs built on historical consumption curves need scenario testing against the architectures now entering design in these markets.
The second exposure is water as an energy proxy. Closed-loop and seawater rejection systems reduce freshwater consumption but shift the energy intensity of heat rejection. Air-cooled chiller plants operating under high ambient temperatures carry compressor loads that evaporative systems avoid — at a water cost. The energy cost of water-free cooling needs to be modeled against water procurement cost and scarcity risk, a calculation that differs materially by geography and falls within the energy head’s budget scope.
The third is transformation timeline risk. If a facility designed with liquid-cooling optionality activates that mode faster than planned because AI rack density accelerates beyond forecast, the power infrastructure serving that facility must scale ahead of the procurement cycle. Transformer lead times in most markets currently run eighteen to thirty-six months. A cooling transition that pulls forward peak electrical load without corresponding infrastructure readiness creates a stranded capacity problem in reverse — not empty data halls waiting for power, but powered data halls waiting for the right thermal infrastructure.
Signals the System Is Shifting
Three indicators are worth tracking. First, whether seawater and closed-loop projects in NEOM and Southeast Asia move from design to operational confirmation — performance data from live facilities would sharpen energy cost modeling for comparable projects globally. Second, whether tenant contracts in India’s hyperscale market begin to specify cooling mode preferences explicitly, which would force operators to commit to a primary architecture rather than maintain full optionality. Third, whether AI workload density in these regions exceeds planning assumptions on an eighteen-to-twenty-four-month horizon — an outcome that would stress cooling and power supply chains simultaneously and force energy procurement decisions before interconnection and transformer timelines can accommodate them.
Sources
- Indiatimes — Hot regions drive liquid cooling adoption for ai data centers, ETDatacenters (Link)
