By Adam Kotrba, Director of Flat Products for Copper Development Association
Summary: Artificial intelligence workloads are revolutionizing data center cooling strategies. Today’s AI chips consume significantly more power than traditional processors, resulting in much higher heat output within densely packed server racks. While conventional air and HVAC cooling systems remain foundational, they can no longer keep pace with these escalating demands. Consequently, innovative solutions, particularly liquid and direct-to-chip cooling, are rapidly becoming standard features in cutting-edge data center design.
Artificial intelligence workloads are revolutionizing data center cooling strategies. Today’s AI chips consume significantly more power than traditional processors, resulting in much higher heat output within densely packed server racks. While conventional air and HVAC cooling systems remain foundational, they can no longer keep pace with these escalating demands. Consequently, innovative solutions, particularly liquid and direct-to-chip cooling, are rapidly becoming standard features in cutting-edge data center design.
Direct‑to‑Chip Liquid Cooling Basics
Direct-to-chip liquid cooling has emerged as one of the most effective solutions. In this setup, a liquid coolant circulates through custom cold plates attached directly to the hottest components, such as CPUs and GPUs. Rather than depending on air or bulky heatsinks, the liquid loop draws heat away from the chip far more efficiently. This method is purpose-built for the extreme power densities of modern AI processors, where conventional air cooling is reaching its limits. By extracting heat directly from the source, direct-to-chip systems enable data centers to run higher-power chips and pack more equipment into each rack, while maintaining safe operating temperatures. Most current deployments rely on single-phase cooling, where the coolant stays liquid throughout the process, simplifying design and making integration with existing facility cooling systems straightforward.
Cold Plates and Copper
Data centers use large amounts of copper for their construction, especially in power networks, circuit boards, and cooling systems. Copper cold plates are at the heart of modern direct-to-chip liquid cooling systems. In AI servers, these plates are mounted directly onto high-power processors, typically one per chip, and are connected to inlet and outlet lines that circulate coolant throughout the system. The coolant then flows into rack-level manifolds, efficiently distributing liquid across the entire server rack. Managing this loop is the Coolant Distribution Unit (CDU), which pumps and regulates a coolant blend—often about 25% coolant and 75% water. Within the CDU, copper is found in critical components such as heat exchanger coils and headers, as well as in motors and certain fittings.
Copper’s extensive use is driven by its exceptional thermal conductivity, which allows it to rapidly absorb heat from AI chips and transfer it to the circulating coolant. This property is crucial for supporting the extreme power densities of today’s GPU-based systems, where air cooling alone falls short. In most single-phase direct-to-chip systems, both the top and bottom sections of the cold plate are made from copper. Some two-phase designs use plastic for the upper portion while retaining a copper base to maintain optimal thermal performance. Many copper cold plates feature microchannels, tiny internal pathways that dramatically increase surface area and enhance heat transfer between the copper and the coolant. By efficiently removing heat directly at the processor, copper cold plates enable data centers to support higher chip power and greater rack density, making them indispensable to advanced liquid cooling in AI environments. Image: Delta CDU system for Google
Image: Cold Plate System Loop
Image: Cold Plate Internals
Liquid‑Cooled Bus Bars & Power Delivery
After a direct-to-chip liquid cooling loop is installed in a rack, the same coolant system can often be used to cool other high-power components, such as busbars. An emerging application is liquid-cooled bus bars, in which a cooling loop removes heat from the rack’s electrical power distribution system. By adding bus bars to the liquid cooling network, data center designers can efficiently manage heat from both computing equipment and power delivery within a unified thermal system.
Actively cooled bus bars can safely carry much higher electrical current than traditional air-cooled conductors. According to TE Connectivity, liquid-cooled designs can support up to five times more current in some cases and may approach nearly double the capacity of uncooled bars while maintaining safe operating temperatures. This increased current capacity allows engineers to reduce conductor size and overall copper mass while still meeting power delivery requirements. At the same time, improved thermal management helps reduce resistive losses, thereby increasing overall power path efficiency in high-density AI server racks
Liquid-cooled busbar inlet and outlet 
Liquid-cooled busbar cross-section
Copper, Efficiency, and Sustainability
AI workloads require significant resources, but data center material choices can help reduce environmental impact. Copper is widely used in cooling systems, power delivery, and interconnects, but its production carries a measurable carbon footprint. As sustainability reporting advances, operators are increasingly tracking embodied carbon at the material level to better manage infrastructure impact.
Copper remains widely used in modern data centers for its performance and long lifecycle. It is highly recyclable, and using recycled copper lowers embodied carbon. In most cases, operational emissions from AI data centers far exceed the carbon footprint of the materials themselves, and the efficiency gains from copper can offset its production impact and pay for upfront higher material costs.
Copper’s high electrical and thermal conductivity enables efficient cooling and power delivery. High-performance copper conductors and cold plates reduce the need to oversize cables, bus bars, and cooling systems. This lowers the total material required and reduces energy loss during partial-load or standby operation. Efficient heat transfer through copper also allows cooling systems to use less fan and pump power and operate at higher chiller temperatures, increasing energy efficiency.
These benefits extend across the facility. Higher power density and more compute per rack mean fewer racks and less supporting equipment are needed for the same workload. This reduces the demand for structural steel, enclosures, and building materials, while maximizing rack space for IT equipment.
Major hyperscale operators are also pairing efficiency improvements with renewable energy. Many companies are investing in renewable energy to reduce operational emissions. For example, Google has matched 100% of its annual electricity use with renewable energy since 2017. This combination of efficient materials and cleaner energy helps reduce the environmental impact of rapid growth in AI infrastructure.
The evolution of AI and its supporting infrastructure presents both challenges and opportunities for data centers. Paying closer attention to embodied carbon in materials like copper is now a priority in the broader effort to decarbonize digital infrastructure. Industry coalitions such as the iMasons Climate Accord, a group of over 250 data center owners, operators, and equipment suppliers, are leading this effort by addressing the carbon footprint of power, materials, and equipment. By adopting efficient materials and advanced cooling strategies, operators can meet growing demand and support the transition to more sustainable data center operations.