The data center landscape in Indonesia is undergoing a massive transformation fueled by the artificial intelligence (AI) explosion. AI workloads, particularly for training Large Language Models (LLMs) and inference, are powered by accelerators like GPUs that generate extremely high heat densities. At these power levels, traditional air cooling becomes impractical, inefficient, and economically unviable. This is forcing a necessary transition towards liquid cooling as the future-proof solution. For data center operators in Indonesia, understanding this liquid cooling architecture landscape is no longer an option but a necessity to remain competitive and sustainable.
Liquid cooling for AI servers offers significant advantages, including improved accelerator reliability and performance, greater energy efficiency, reduced water usage (in certain air-side systems), and lower sound levels. At its core, liquid cooling captures heat directly at the source—the CPU and GPU chips—with far greater efficiency than air ever could.
The two main categories of liquid cooling for AI servers are direct-to-chip and immersion. Regardless of the heat capture method inside the server, a heat rejection ecosystem is ultimately required to dissipate this heat to the outside environment. This is where the crucial role of a Cooling Distribution Unit (CDU) and the broader cooling architecture comes into focus.
A CDU is the heart of a liquid cooling system. Its function is to isolate the IT cooling fluid (within the Technology Cooling System or TCS) from the rest of the facility’s cooling system. A CDU is not just a heat exchanger; it performs five key functions: precisely controlling fluid temperature, controlling flow to ensure heat is carried away, managing pressure within the TCS loop (including negative pressure systems to prevent leaks), performing fluid treatment like filtration and maintaining water chemistry, and of course, facilitating heat exchange and isolation.
When selecting the right liquid cooling architecture, we can map it based on three fundamental elements: the method of heat capture within the server (out of this article’s scope), the CDU type, and the method of rejecting heat to the outdoors.
Understanding CDU Types: Liquid-to-Air vs. Liquid-to-Liquid
The CDU type is primarily determined by two critical attributes: the type of heat exchange and its capacity and form factor.
- Type of Heat Exchange: The two most common types are:
Liquid-to-Air (L-A): Heat from the liquid TCS loop is transferred to a coil or radiator, where fans then reject this heat directly into the data center room air. It’s like a giant radiator for your servers.
Liquid-to-Liquid (L-L): Heat from the liquid TCS loop is transferred to an existing facility water system, such as a chilled water or condenser water loop. This is a deeper integration with the building’s cooling infrastructure.
- Capacity and Form Factor:
Rack-mounted CDU: Installed inside the server rack, typically serving a single rack. Capacities for L-A range from 20-40 kW, while L-L can reach 40-80 kW. This solution is ideal for small-scale deployments.
Floor-mounted CDU: A standalone unit on the floor, serving a row or multiple rows of servers. Floor-mounted L-A CDUs have capacities up to 60 kW, while floor-mounted L-L CDUs can handle loads from 300 kW to over 1 MW, suitable for large-scale AI clusters.
Methods of Heat Rejection to the Outdoors: Leveraging Existing vs. Building Dedicated
Once the heat is captured by the TCS loop, the next step is to reject it outside. There are three common methods:
- Leverage Existing Heat Rejection System: Reject Heat to IT Space Air.
This architecture is the simplest to adopt. L-A type CDUs (either rack or floor-mounted) reject heat directly into the server room air. The existing traditional air cooling systems, like Computer Room Air Handlers (CRAH), then reject this heat outdoors. This architecture is closed-loop and local.
Advantages: High compatibility with existing air-cooled infrastructure, no major modifications needed, can be prefabricated for quick installation, and failures are isolated to a single rack.
Disadvantages: Lower efficiency due to multiple heat exchange stages and fan use, costly for large-scale deployments, and consumes valuable rack or floor space.
When to Implement: Ideal for small-scale deployments (a few servers to several racks), when a connection to chilled water is not feasible, or when speed of deployment is the top priority.
- Leverage Existing Heat Rejection System: Reject Heat to Facility Water Systems.
Here, L-L type CDUs transfer heat from the TCS loop to an existing facility water system (chilled or condenser water loop). The heat is then rejected outdoors by the facility’s chillers or reused for other purposes like district heating.
Advantages: Reduces investment by leveraging existing infrastructure, higher efficiency and quieter operation than the L-A approach, and can recapture space previously used by CRAHs in retrofit projects.
Disadvantages: Requires more on-site installation work to connect the CDU to the facility water piping.
When to Implement: Suitable for medium to large deployments in data centers with existing chiller plants, where water connections are available, and energy efficiency is prioritized over absolute installation speed.
- Build a Dedicated Heat Rejection System.
For large-scale AI deployments, building a dedicated heat rejection system for liquid cooling is the most efficient option. This architecture uses L-L CDUs connected to an independent heat rejection system, such as a dry cooler with compressor assist. This system is optimized for high return water temperatures (up to 40°C), which significantly increases free cooling hours.
Advantages: Highest energy efficiency, high return water temperatures open opportunities for heat reuse, and implementation does not disrupt existing cooling system operations.
Disadvantages: Requires significant capital investment, design of a new piping system, and a longer deployment timeline.
When to Implement: When the deployment of liquid-cooled servers is significant, energy efficiency is an absolute priority, and there are plans to reuse waste heat.
A Guide to Choosing the Right Architecture
Choosing the best architecture is about balancing key factors: existing infrastructure compatibility, deployment scale, speed of deployment, and energy efficiency.
For Small Scale & Proof-of-Concept (1-10 racks): The “Reject Heat to IT Space Air” architecture with rack-mounted L-A or L-L CDUs is the best choice. This solution is quick to implement, low-risk, and non-disruptive to the existing facility.
For Medium to Large Scale (>10 racks) with Existing Chiller Infrastructure: The “Reject Heat to Facility Water Systems” architecture with floor-mounted L-L CDUs is a highly efficient and economical choice. It offers a lower cost per kW and superior performance.
For Dedicated, Large-Scale AI Clusters: The “Dedicated Heat Rejection System” architecture with floor-mounted L-L CDUs is the pinnacle of efficiency and sustainability. Although it requires a larger upfront investment and longer timeline, the operational savings and heat reuse opportunities make it a strategic long-term choice.
Implementation Challenges in Indonesia and The Role of a Local Partner
Indonesia’s tropical climate adds a layer of complexity in designing efficient heat rejection systems. Choosing a technology partner that understands local challenges, possesses deep technical expertise, and can provide reliable components and systems is key to success. This is where partnering with a trusted distributor and integrator of data center cooling systems, like Climanusa, becomes critical.
Climanusa, with its portfolio of precision cooling solutions, positions itself as a strategic partner for data centers in Indonesia that are preparing for the wave of AI computing. From providing reliable CDUs and piping systems to supporting solutions like integrated Precision Air Conditioners and Chillers, Climanusa ensures that every layer of the liquid cooling ecosystem—from the chip to the outdoors—is designed and implemented flawlessly. Their expertise enables data center operators not only to choose the right architecture in theory but also to realize it in practice, ensuring operational availability, efficiency, and sustainability amidst the unique challenges of the Indonesian climate.
—
Why Choose Climanusa?
In facing the era of AI-intensive computing, choosing the right partner for your data center cooling solution is a strategic decision. Climanusa is not just a distributor for data center cooling; we are an integration partner that understands the complexities of liquid cooling from end to end. With a comprehensive product portfolio and deep technical expertise, Climanusa is ready to support you in designing and implementing the most efficient and reliable liquid cooling architecture, tailored to your specific needs and the climatic challenges in Indonesia. Entrust your data center transformation to our experts, and turn the heat of AI workloads into a cool advantage for your business sustainability.
For more information, please click here.
–A.M.G–
