Six billion liters of water per year not drawn from local supplies: that’s the estimated saving across the two campuses EdgeConneX is building in Cyberjaya and Johor, Malaysia. The number alone marks a departure in Southeast Asia, where data centers increasingly compete with agriculture and residential needs for water. But it’s the visible result of a more radical engineering choice: scrapping evaporative cooling in favor of a closed-loop system designed to handle power densities that most legacy facilities can’t touch.

The figures are significant. EdgeConneX is deploying over 500 MW of aggregate capacity, with racks built to absorb from 20–200 kW for near-term hybrid workloads up to more than 600 kW when next-generation GPU clusters ramp up. Each data hall can pull from modular 12 MW power blocks, scaling with demand. And it’s all NVIDIA DGX certified: compliance with the partner’s reference standards means power delivery, cooling, and network topology are validated for DGX systems, cutting technical risk and deployment complexity for enterprise and hyperscaler customers.

Underpinning the flexibility is an architectural philosophy the company calls “data-center-as-a-backplane”: the building becomes a programmable platform, where cooling configurations and electrical distribution can be reconfigured without heavy structural work. An aisle can run 30 kW CPU racks next to a 200 kW GPU cluster, and everything can evolve over a 20–30 year facility lifespan. For organizations evaluating on-premise LLM deployments, that has immediate implications. It means avoiding disruptive retrofits every time chip architectures change, which can happen every 18–24 months, keeping TCO in check and preventing vendor lock-in to a single hardware generation – a factor that grows critical for enterprises that want to retain data sovereignty without being tied to obsolete infrastructure.

The other pillar is thermal management. Instead of evaporative towers that use water to dissipate heat, the Cyberjaya and Johor campuses rely on non-evaporative heat rejection technology and high-efficiency chillers, capable of handling liquid-cooled racks without consuming a drop. At full load, the daily water saving is estimated at 9 million liters in Cyberjaya and 7.6 million in Johor – a crucial difference in a region where multiple operators struggle to secure water abstraction permits. EdgeConneX targets water neutrality by 2030 and UL 2799 zero-waste-to-landfill certification at the new sites within 18–24 months of start-up, replicating what it has already achieved in Jakarta.

Energy independence completes the picture. Both campuses will be anchored by a 275 kV substation, the first of its kind in the greater Kuala Lumpur region, enabling operation independent of the national grid if needed. It’s a direct response to the grid constraints that are slowing data center growth in other Southeast Asian economies, and a bargaining chip for organisations that need continuous uptime for mission-critical inference workloads.

For the local ecosystem, EdgeConneX pledges predominantly local hiring and technical training programs starting in the second half of 2026, building on similar initiatives in Dublin, Atlanta, and Silicon Valley. What exactly fills these facilities remains to be seen: phase one is expected in the first quarter of 2027, and the actual mix of training, inference, and hybrid workloads will depend on how fast customers’ AI ambitions materialize. But the direction is unmistakable: building a data center for AI today can no longer afford to consider only today’s power requirements, or yesterday’s water assumptions.