TOPIC
Direct-to-chip, immersion, single-phase and two-phase liquid cooling technologies.
Korea has no Foxconn. But a group of mostly mid-sized companies — Wellmade, KTNF, Solum, Global TopNet, Hanwool, eSlim Korea, Lian System and Elice — is now trying to build the Korean AI server, from the motherboard and chassis to the power supply and the cold plate.
As Tech Week Singapore 2026 approaches, Supermicro and Giga Computing are presenting two distinct paths toward the AI factory. Supermicro is moving into the NVIDIA Vera Rubin era with its DLC-2 liquid-cooling architecture, while Giga Computing is scaling Blackwell Ultra through GIGAPOD and its GAIFA validation model. The comparison shows how the competition is shifting from individual AI servers toward complete rack, cooling and data-centre infrastructure.
Single-phase direct-to-chip cooling remains the dominant architecture for high-performance AI systems, but rising GPU power is pushing it toward a new thermal boundary. With industry analysis placing Vera Rubin in the 1.8–2.3kW range and IDTechEx forecasting broader two-phase D2C adoption from 2026–2027, AI infrastructure may be approaching its next major cooling transition.
NVIDIA’s Vera Rubin generation reveals an important shift in AI infrastructure design. Cooling manifolds, quick disconnects, power whips and even the rack busbar are becoming integrated parts of the computing architecture. Field images of Vera Rubin reference hardware show how liquid and electrical distribution are being redesigned together for higher-density AI factories.
Korea’s AI data center cooling market is moving from imported equipment toward domestic manufacturing, local integration and multi-megawatt liquid-cooling infrastructure. Samsung’s planned FläktGroup production line in Gwangju could accelerate that transition from 2028.
Liquid cooling in AI data centers is moving beyond GPUs and CPUs. As NVIDIA’s Vera Rubin generation pushes AI factories toward fully liquid-cooled, fanless rack architectures, enterprise SSDs are becoming part of the same thermal transition. Samsung’s new PCIe 6.0 PM1763 and SK hynix-affiliated Solidigm’s cold-plate-cooled D7-PS1010 show that storage is becoming the next frontier of Direct Liquid Cooling.
LG Electronics is rapidly expanding its chip-to-chiller cooling portfolio for AI data centers, while two-phase Direct Liquid Cooling is emerging as a potential next step for ultra-high-density AI racks. At OCP Korea Tech Day on August 21, 2026, a technical session also highlighted the transition from today’s single-phase DLC infrastructure toward future two-phase architectures. Israel-origin ZutaCore provides one of the clearest commercial references for how that transition may develop.
NVIDIA's next AI racks are moving toward megawatt-class power, but will single-phase liquid cooling remain enough? DATAAD believes the most practical long-term path may be two-phase direct-to-chip cooling that evolves from today's existing DLC infrastructure rather than replacing it entirely.
Air cooling remains simple and economical for conventional data centers, but AI racks are moving beyond the densities it was designed to handle. DATAAD compares air and direct liquid cooling across rack density, energy, cost, infrastructure, maintenance and future scalability.
As AI racks move beyond 100 kW, the coolant distribution unit is becoming one of the most important pieces of data-center infrastructure. DATAAD compares leading CDU suppliers including Vertiv, Schneider Electric/Motivair, CoolIT, Delta, Rittal, LG, Cooler Master and nVent.
Single-phase direct liquid cooling is becoming the standard architecture for today's high-density AI racks. But rising chip and rack power could push the industry toward two-phase DLC from around 2028 — creating a new market for low-GWP fluids, cold plates, QDs, hoses, fittings, CDUs and facility HVAC.
AI liquid cooling is moving beyond simple connector compatibility. NVIDIA’s deeper involvement in OCP, UQD/UQDB V2, smaller server-side MQD interfaces and the rise of two-phase DLC are reshaping what the industry expects from cooling couplings.
As AI infrastructure scales, the constraint is no longer compute alone. Power density, heat removal, liquid-cooling architecture and reliable fluid interfaces are becoming critical to how modern data centers are designed and operated.