Why Does the Secondary Side of the Stargate Liquid-Cooled Data Center Adopt the Overhead Piping Scheme?

1. Introduction to the Secondary-Side Piping Schemes for Liquid-Cooled Computer Rooms

The secondary-side piping methods of liquid cooling systems in data centers are mainly divided into two schemes: Overhead Piping and Underfloor Piping.Overhead Piping refers to installing pipes above the server racks, suspending them from the ceiling, similar to water pipes 'running along the ceiling'.

Underfloor Piping involves laying liquid cooling pipes under the floor, which is somewhat similar to the weak and strong electrical wiring in traditional computer rooms. Currently, most liquid cooling projects in China adopt the underfloor piping scheme.

At present, the technical trend for large-scale overseas liquid cooling projects is to adopt the overhead piping scheme, while most domestic liquid cooling projects still use the underfloor piping scheme.

2. OpenAI's Stargate Liquid Cooling Project Adopts the Overhead Piping Scheme

The first phase of the 'Stargate' project started construction in June 2024, including two buildings with a power generation capacity of over 200 MW, and is expected to be put into use in the first half of 2025. The second phase, which broke ground in March this year, will add six more buildings with an additional gigawatt-level capacity and is scheduled to start operation in the middle of 2026.

In total, the eight buildings will operate up to 50,000 Nvidia GB200 NVL72 GPU clusters on a single integrated network architecture. Each building will house a large, fully interconnected GPU cluster, with a total of up to 400,000 of Nvidia's latest Blackwell AI chips. Once completed, it will become the world's largest known AI computing cluster.

For a large-scale liquid-cooled data center project like Stargate—with a GW-level capacity and hundreds of thousands of Blackwell chips—choosing an appropriate layout and design scheme for liquid cooling pipes is particularly critical. The right choice will facilitate subsequent rack expansion, operation, maintenance, and overhaul; a wrong one will lead to tedious troubles every time a problem occurs.

So why did Stargate adopt the overhead piping scheme instead of the underfloor piping scheme commonly used in China? We (Lingke) believe the key reasons lie in safety, maintenance difficulty, and scalability, as detailed below:

1. A project with hundreds of thousands of chips requires a dense network of pipes. If underfloor piping is used, any malfunction would require lifting the floor, resulting in enormous workload. With overhead piping, if there is a water leak or pipe modification is needed, operations can be directly carried out above, significantly improving maintenance efficiency.

2. In large-scale projects, a single water leak may affect thousands of chips or even an entire row of server racks. Overhead piping is equipped with drip trays, which can collect the coolant and drain it to a designated point, reducing leakage risks. Leaks are visible at a glance, enabling timely control and preventing 'hidden leaks' under the floor from causing major accidents.

3. Such projects usually deploy GPUs in phases, and demand may continue to grow. Overhead piping allows for easier later-stage expansion and modification without the need for large-scale floor removal.

The images below show the interior of the 'Stargate' data center project. The objects visible above are the liquid cooling pipes, which transport coolant to various parts of the rack clusters to cool them down.

3. Advantages of the Overhead Piping Scheme for Liquid-Cooled Data Centers

As seen from the Stargate project case, the overhead piping scheme offers greater advantages for ultra-large-scale liquid cooling projects. The detailed advantages of this scheme are as follows:

3.1 Safety Comparison

Overhead Piping Scheme: This is the biggest advantage of overhead piping. Although liquid cooling pipes are under pressure and filled with liquid, and leakage risks have been minimized through strict design, such risks still exist. The overhead piping design minimizes the impact of these risks: leaked liquid will be collected by pre-designed drip trays and drained to a discharge point, and the alarm system will be triggered immediately. This prevents the liquid from coming into contact with the servers, power distribution units (PDUs), and network equipment below, effectively protecting expensive IT equipment and reducing liquid leakage risks.

Underfloor Piping Scheme: When a leak occurs, the liquid accumulates silently under the anti-static floor and is not easily detected in a timely manner. It can soak power cables and data cables, causing insulation damage, short circuits, or even fire hazards. At the same time, accumulated liquid may damage the floor structure and breed mold. Choosing underfloor piping requires supporting it with extremely expensive and sensitive leak detection systems (such as sensing cables) and a complete waterproof and drainage design; otherwise, it will entail huge potential risks.

3.2 Maintenance Difficulty

Overhead Piping Scheme: Daily inspections (checking valve status, presence of condensation, and integrity of joints), regular maintenance (replacing filters and pumps), and fault diagnosis are very convenient. This greatly reduces the labor intensity and skill requirements for maintenance personnel, shortens maintenance time, improves maintenance response speed, reduces human operation errors, and thus lowers long-term maintenance costs.

Underfloor Piping Scheme: Maintenance personnel must lift the floor and work in a narrow, dim space that is usually filled with other cables. This makes any inspection, maintenance, or modification extremely difficult, time-consuming, and prone to damaging other facilities. Minor issues may be overlooked and eventually develop into major failures. Each maintenance session means higher labor costs and longer downtime.

3.3 Scalability and Flexibility

Overhead Piping Scheme: It offers extremely high flexibility. When adding or reconfiguring racks, pipes can be extended or adjusted on the existing sturdy cable trays, as conveniently as expanding network cables. This causes minimal interference to the existing operating environment and supports the rapid launch of services and elastic growth, making it a future-oriented design.

Underfloor Piping Scheme: It is relatively rigid. The space under the floor is limited and already occupied by various cables. Adding or modifying pipes usually requires large-scale floor removal and rewiring, which involves complex engineering, high costs, and significant interference to existing services. Choosing underfloor piping may mean facing huge challenges and costs for future expansion, or even becoming a bottleneck for technological development.

From the perspectives of safety, maintenance difficulty, scalability, and other aspects, the overhead piping scheme has multi-dimensional advantages. Therefore, overseas liquid cooling projects (especially large-scale ones with GW-level capacity and 100,000+ chips) have gradually begun to adopt this more optimal overhead piping scheme.Of course, in addition to factors such as technology and deployment scale, the differences in liquid cooling pipe routing between overseas and domestic data centers are also partly determined by construction habits, computer room design concepts, and operation and maintenance cultures.

Kres(Suzhou) Power Systems Technology Co., Ltd is an emerging company focused on fluid power solutions, and serve for the markets of Energy, Construction, Agriculture, Manufacturing, Transportation, Medical, Data Centers, etc., and was founded in 2022 with leading-edge engineering capabilities and innovative products to keep up with the changing customer needs. Including OCP standard UQD, UQDB, BMQC, manifold, EPDM hose and hose assemblies, FF9 full flow dual interlock quick disconnect coupling (a better replacement for FD83 series products), API heat transfer products, Gates EPDM hose and Wika products.

KRES FF9 connectors are specifically engineered for fluid transmission, electronic cooling, and data center liquid cooling applications, where full flow, fluid compatibility, and safety are of paramount importance. The FF9 series ball valve quick-disconnect couplings adopt a full-flow design and are now available in 2-inch, 1.5-inch, and the popular 1-inch sizes. The FF9 double interlock liquid cooling quick-change connector features a swivel structure at its tail, facilitating easy connection and disconnection. Additionally, its two connector halves are completely identical, simplifying maintenance, installation, and service processes.

· 100% full-flow interlock ball valve: The FF9 double interlock ball valve has a diameter of up to 21mm, ensuring large flow while reducing pressure drop, achieving large flow and low pressure drop, and has obtained SGS ROHS reports and Huace performance test reports.

· KRES interlock ball valve adopts a double interlock design with two locking mechanisms to prevent accidental disconnection and ensure data security.

· The FF9 liquid cooling quick-change ball valve has identical connector halves, making maintenance, installation, and service simple.

· As a liquid cooling quick-change ball valve for data centers, it has reliable and stable performance with minimal leakage, meeting the maintenance needs of safe machine relocation or shutdown.

· Standard body material of the liquid cooling quick-change connector: 304 stainless steel, with wide fluid compatibility, and other materials can be provided upon request.

· Standard sealing material of the liquid cooling quick-change connector: EPDM ethylene propylene diene monomer synthetic rubber, and other materials can be provided upon request.

· Operating temperature range of the liquid cooling-specific ball valve: -40℃ ~ 120℃