The role of liquid cooling in AI and digital infrastructure
Examining the impact of AI and digital services on power consumption and the advantages of liquid cooling.
The digital age is evolving at a breakneck pace, driven largely by advancements in artificial intelligence (AI), cloud computing, big data, and Internet of Things (IoT) technologies. However, this explosion in digital demand comes with a steep costs, rapidly escalating power consumption by data centers that support these services.
This has raised critical concerns about energy sustainability, infrastructure resilience, and environmental impact. In this context, liquid cooling has emerged as a key enabler of more energy-efficient data center operations, especially as traditional air-cooling methods struggle to address the increasing thermal load.
AI workloads, particularly those involving deep learning and large language models (LLMs), are computationally intensive. Training models like OpenAI’s GPT series or Google’s Gemini require tens of thousands of GPUs running for weeks or months. For example, GPT-3 training alone is estimated to have consumed several gigawatt-hours (GWh) of electricity.
Beyond training, AI models require power for inference across millions of devices daily, providing responses, predictions, or classifications. Even devices at the edge, such as autonomous vehicles or industrial robots, rely on data center backends to process complex tasks in real-time.
The global shift to cloud computing, accelerated by remote work and SaaS models, demands large-scale server farms to host applications, storage, and compute services. In multi-tenant environments cloud providers pack servers densely to maximize ROI, increasing heat density and, in order to meet high availability expectations, 24/7 uptime requires redundancy and backup power infrastructure, compounding energy demands.
Streaming platforms, online gaming, and social media services generate massive data traffic, while billions of IoT devices constantly upload and process telemetry, video, and sensor data.
Data centers now process zettabytes of data annually. According to the International Energy Agency (IEA), data centers accounted for about 1-1.5% of global electricity consumption as of 2022. With the acceleration of AI and digital services, this could rise to 3–4% or more by 2030, potentially surpassing some national industrial sectors in total energy usage.
The liquid cooling opportunity
In terms of increasing data center high energy use, compute-intensive AI chips, redundant systems to help guarantee uptime, the necessary mechanical and electrical infrastructure and the cooling systems are the major ‘culprits’.
Work is being done to improve energy efficiency in all these areas. In terms of the cooling opportunity, there is also the requirement to develop and operate new cooling technologies which can cope with the higher compute densities required by AI workloads – with more and more heat being generated.
Traditional cooling methods rely on air circulated by fans, Computer Room Air Conditioning (CRAC) units, and raised floor systems. This model is facing severe limitations in terms of being able to address the demands of AI and the wider digital world. Air has poor heat capacity, making it inefficient at transferring large quantities of heat. Airflow requires powerful fans and compressors that consume electricity continuously. To maintain effective cooling, racks must be spaced apart, reducing server density. Uneven cooling can lead to hardware degradation or catastrophic failures.
As server density grows, reaching as much as 100 kW per rack in some AI data centers (and with talk of the 1mW rack), air cooling is no longer adequate.
Step Forward Liquid Cooling
Liquid cooling uses fluids with high thermal conductivity to absorb and carry away heat from servers. Various methods include:
- Direct-to-Chip (D2C): Coolant flows through cold plates attached to processors.
- Immersion Cooling: Servers are submerged in non-conductive fluid.
- Rear Door Heat Exchangers (RDHx): Heat is removed from air exiting the rack using liquid-cooled panels.
The advantages of liquid cooling for AI and digital services include superior thermal efficiency, reduced energy consumption, smaller footprint and increased scalability, extended hardware lifespan and the potential to enable sustainable AI infrastructure.
In terms of the superior thermal efficiency, water and dielectric liquids absorb and transport heat far more effectively than air - up to 3,000 times higher thermal conductivity. This allows data centers to manage higher densities and thermal loads without overheating. Liquid cooling supports rack densities exceeding 50–100 kW per rack - ideal for GPU farms and AI clusters and also provides consistent thermal conditions, eliminating hot spots and enabling better thermal predictability.
Cooling accounts for a major portion of data center power use. Liquid cooling systems reduce this burden dramatically. In practical terms, this means a reduction in PUE from typical air-cooled values of 1.6–2.0 down to 1.05–1.2 with liquid cooling. Additionally, direct contact with heat sources means fans can be downsized or eliminated, saving additional energy.
Higher cooling efficiency means servers can be packed more densely, providing more compute per square foot and also making AI-capable micro data centers viable in remote or space-constrained environments (vertical and/or edge deployments).
Liquid cooling reduces thermal cycling and exposure to dust and air particles, minimising physical stress on components, extending equipment life and leading to lower failure rates, reducing downtime and maintenance costs.
Data center operators are under pressure to reduce carbon footprints and liquid cooling results in lower overall energy use when compared to air cooling, potentially reducing reliance on fossil-fuel power. Significantly, waste heat from liquid-cooled systems can be redirected to warm buildings, greenhouses, or industrial facilities.
A note of caution, liquid cooling Capex is higher than air cooling, although lifetime operating costs should be much lower. Additionally, liquid cooling requires new infrastructure, more specialised maintenance, new safety protocols – a willingness to embrace significant innovation. It is also worth remembering that not all servers are designed for liquid cooling and that technology standardisation and interoperability are still evolving.
AI and digital services are ushering in a new era of technological capability and a parallel surge in power demand. The increasing complexity, density, and thermal output of modern data center hardware makes traditional air cooling insufficient. Liquid cooling offers a high-performance, energy-efficient, and scalable solution that aligns with both business and environmental imperatives.
From hyperscale cloud providers to edge computing environments, liquid cooling is rapidly transforming from a niche innovation into a cornerstone of future-ready infrastructure. As data centers prepare for the next wave of AI-driven demand, liquid cooling is poised to become not just an option, but a necessity.