Meta’s 2026 New Albany expansion is its first U.S. campus designed from the ground up for direct‑to‑chip (DLC) liquid cooling as the default thermal architecture. The new AI hall uses warm‑water cooling loops (30–45°C supply) feeding high‑flow cold plates mounted directly on Blackwell‑generation GPUs, with facility‑scale distribution manifolds and redundant coolant distribution units (CDUs). The expansion block is sized for 120–150 kW per rack, with electrical rooms and busways re‑engineered to support high‑density AI clusters. Meta uses a water‑based, treated closed‑loop coolant, avoiding dielectric fluids and minimizing water consumption through recirculation and heat‑recovery integration. The project represents Meta’s shift to liquid‑first mechanical design, with air cooling relegated to secondary components only.
Supermicro’s 2026 San Jose integration campus is a manufacturing‑plus‑deployment hub producing factory‑integrated liquid‑cooled racks for hyperscale AI customers. These racks are built around NVIDIA Blackwell and Rubin GPUs, each with 1,000–2,300W TDPs, requiring direct‑to‑chip cold plates and high‑flow manifolds. The racks ship as 120 kW+ liquid‑cooled blocks, pre‑plumbed with quick‑disconnect fittings and tested using water‑glycol coolant. While the campus itself is not a hyperscale data center, the 2026 deployments tied to it are with thousands of racks delivered to U.S. and APAC AI clusters. This is one of the first global supply‑chain nodes where liquid cooling is integrated at the factory, not retrofitted onsite, reducing commissioning time and ensuring consistent thermal performance across deployments.
OVHcloud’s new 2026 Limburg data hall is its first German facility designed entirely around two‑phase immersion cooling, using OVH’s proprietary dielectric fluid and tank architecture. Each server is submerged in a sealed bath of boiling dielectric coolant, which vaporizes at low temperature and condenses on integrated heat‑exchanger coils, enabling extremely stable thermal performance. The hall is sized for 50–120 kW per rack equivalent, with immersion tanks replacing traditional rack rows. OVH integrates district heat‑reuse, exporting high‑grade heat from the condenser loop to local municipal systems. This project represents one of Europe’s most advanced immersion‑native AI and HPC deployments, eliminating server fans, reducing water consumption to near zero, and enabling high‑density compute in a compact footprint.
Keppel’s 2026 expansion at SGP5 introduces a direct‑to‑chip liquid‑cooled AI block engineered for 150–200 kW rack densities, making it one of the densest commercial deployments in Southeast Asia. The system uses closed‑loop water‑glycol coolant feeding cold plates, with high‑efficiency CDUs and heat‑rejection systems optimized for Singapore’s tropical climate. Keppel integrates liquid‑ready manifold corridors, redundant pump sets, and high‑temperature chilled‑water interfaces to reduce mechanical load. The block is designed for AI training clusters requiring extreme power density, and the mechanical plant includes elevated supply‑temperature cooling to improve energy efficiency in humid conditions. This is one of the region’s first liquid‑native hyperscale expansions.
Digital Realty’s KIX13 expansion in Osaka adds a new 2026 AI hall engineered for direct‑to‑chip liquid cooling across its entire compute footprint. The hall supports 120–180 kW racks, with high‑flow water‑glycol loops feeding cold plates and optional RDHx for mixed workloads. Digital Realty redesigned the base building MEP to include liquid‑ready risers, manifold rooms, and high‑capacity CDUs, making this its first Japan facility where liquid cooling is integrated from day one rather than retrofitted. The coolant loop is fully closed‑circuit, with plate‑and‑frame heat exchangers interfacing with the building’s chilled‑water system. Osaka’s power‑dense AI demand makes this hall a flagship example of Japan’s shift toward liquid‑first hyperscale design.
| Project | Location | Liquid‑Cooling System | Coolant | Rack Density | Facility Scale |
| Meta – New Albany (Prometheus) | New Albany, Ohio, USA | Direct‑to‑chip warm‑water DLC | Water‑based closed loop | 120–150 kW | Gigawatt‑scale AI supercluster |
| Supermicro + NVIDIA – San Jose Integration Campus | San Jose, California, USA | Factory‑integrated DLC racks | Water‑glycol | 120 kW+ | Multi‑building integration & manufacturing campus |
| OVHcloud – Limburg Immersion Hall | Limburg, Germany | Two‑phase immersion + hybrid water loop | Dielectric immersion fluid | 50–120 kW | New immersion‑native hall |
| Keppel – Singapore SGP5 Liquid‑Cooled Block | Jurong, Singapore | Direct‑to‑chip DLC | Water‑glycol | 150–200 kW | High‑density AI block expansion |
| Digital Realty – Osaka KIX13 AI Hall | Osaka, Japan | Direct‑to‑chip DLC | Water‑glycol | 120–180 kW | 21 MW AI‑ready hall |
