My D. Truong, Chief Product and Technology Officer, ZutaCore
For most people, data centres used to sit outside the public conversation. They powered cloud software and business operations, but the buildings themselves remained largely unseen.
AI has completely changed that. As organisations push models into operations, the infrastructure behind the technology is becoming impossible to ignore. New sites are seeking larger grid connections, while existing facilities are being asked to carry denser racks and hotter chips.
This shifts the conversation for operators and their surrounding utilities. AI infrastructure has clear societal value, but its resource footprint is drawn directly from local ecosystems. This creates a dual burden: pulling power from grids already under strain, and extracting water from regions dealing with drought, scarcity, or rising household utility costs. If data centre growth is to retain public consent, the industry must prove that digital capacity can expand without placing avoidable pressure on the communities around it.
Power and cooling: two sides of the same coin
The speed of that expansion explains why the debate has become more urgent. The International Energy Agency recently reported that electricity demand from data centres rose by 17% in 2025, with AI-focused facilities scaling fastest. For utility planners, this changes how the sector is viewed. Electricity demand, however, tells only half the story.
Power and cooling are two sides of the same coin. Every watt a server draws must leave the facility as heat. When cooling cannot keep pace, chips throttle, meaning the site still consumes its grid allocation but delivers less computational output. Ground-breaking AI growth cannot be delivered by securing more power alone; it demands an equally fundamental rethink of cooling strategy.
The limitations of traditional air cooling become clear in high-density AI environments. While it remains effective for moderate workloads, it can struggle when high-powered accelerators operate close to peak capacity for extended periods. Forcing more air through a room spikes fan energy consumption and leaves performance hostage to minor airflow disruptions.
Good neighbours: earning the right to grow
AI workloads do not run in abstract environments. They live in physical facilities that share grids, water systems and skylines with the people around them. If those facilities are to keep expanding, they must be positive contributors to the communities they are built in, not silent draws on shared infrastructure. That demands answers rooted in physical design, not broad sustainability language.
Nowhere is that truer than with water, one of the most polarising topics in data centre development. Evaporative cooling is increasingly difficult to defend in water-stressed regions, but the conversation cannot focus only on where water is scarce. Water use must be considered in the context of each site and its surrounding area, because designing only for today’s conditions can create risks later. Even where water is plentiful, risks can still arise. Climates shift, populations grow, and allocations tighten over a facility’s lifetime; so water demand that looks manageable today may become harder to justify later.
Rather than optimising conventional cooling architectures around local conditions, the more valuable question is what becomes possible when a facility is designed to be water-free from end to end.
Discarding convention: the end-to-end water-free alternative
The answer starts at the chip, where the heat is created. In waterless, two-phase direct-to-chip cooling, a dielectric fluid in a sealed loop vaporizes as the processor heats up, carrying heat away efficiently before condensing back to the cold plate. No evaporative towers, no facility water loops and no water consumed anywhere in the heat-rejection chain.
The approach scales with next-generation AI processors – now climbing past 4,000 watts per chip – while avoiding the fluid-quality, maintenance and leak risks of water-based systems. And because the loop is sealed, the same design performs identically in the desert or beside a river, decoupling AI growth from local hydrology altogether.
That decoupling changes what operators can offer the places they build in. Siting no longer hinges on water rights, and heat captured directly at the chip supports higher rack densities within the same grid allocation – more computational value for every megawatt a community grants.
The heat itself, so often treated as a liability, can become a community benefit. Every megawatt of IT load leaves the facility as thermal energy. When that heat is captured directly at the chip, it emerges concentrated and usable, rather than being diluted into warm air and released into the atmosphere.
That opens up genuinely creative possibilities. Piped into local heat networks, recovered heat can warm nearby homes, schools and swimming pools through the winter. In coastal or water-stressed regions, it can drive thermal desalination, allowing a data centre that consumes no water to help produce it for the community around it. Paired with heat-to-power technologies such as organic Rankine cycle systems, waste heat can even be converted back into electricity, returning capacity to the very grid the facility draws from. The IT load stops being purely a burden on local infrastructure and starts giving something back.
As development accelerates, utilities and communities will demand reassurance that local resource systems are not being stretched for avoidable reasons. Operators that remove heat at the source and eliminate water dependency will have the strongest case for expansion – proof that AI infrastructure can scale as a positive contributor to the places around it, not a hidden cost carried by the people who call them home.
