This way, all waste heat generated by the components is captured by the dielectric liquid inside the tank, which is capable of absorbing around 1500 times more heat than air, for the same volume. The heat then flows to a heat exchanger where it’s cooled by a secondary air to liquid or liquid to liquid heat exchanger to be dissipated or reused. 

Today, environmental concerns are at the heart of every industry conversation. Conventional cooling methods are no longer sufficient. What can we do? Switching servers off is simply not an option.

AI Is Reshaping Data Center Energy Demands: Key Figures

The global data center market was valued at approximately $348 billion in 2024 and is projected to exceed $1 trillion by 2034, growing at an 11.2% CAGR (GlobeNewswire / McKinsey). Global capacity is expected to double from 103 GW in 2025 to 200 GW by 2030, driven by the explosive growth of AI and cloud computing (JLL, 2026), and with it, an unprecedented infrastructure investment supercycle of up to $3 trillion by 2030.

This rapid expansion comes at a significant environmental cost. Global electricity demand from data centers surged by 17% in 2025, with AI-focused facilities growing even faster at +50% (IEA, 2025). Consumption is projected to nearly double from 485 TWh in 2025 to 950 TWh by 2030, representing approximately 3% of global electricity demand (IEA, Energy and AI Report, 2025). The carbon footprint of AI systems alone could reach between 32.6 and 79.7 million tonnes of CO₂ in 2025, while water consumption could reach up to 764 billion litres (ScienceDirect, 2025).

These figures make one thing clear: conventional cooling methods are no longer sufficient. Reducing the energy and environmental footprint of data centers is no longer optional, it is a strategic and operational imperative.

2CRSi recognized this urgency and works every day to develop new cooling methods to cool IT equipment in all type of settings, such as direct on chip cooling or immersion cooling.

Immersion-cooled GPU and AI servers

Immersion cooling isn't just for general-purpose compute, it has become the cooling solution of choice for the most dense and demanding GPU infrastructures. At 2CRSi, we design and manufacture high-performance GPU servers specifically engineered for immersion cooling for data centers, both single-phase and two-phase.

Our immersion-cooled GPU servers take full advantage of the technology's benefits: optimal heat dissipation, reduced energy consumption, and a PUE close to 1. From AI training to large-scale inference, our immersion-cooled GPU servers are the ideal infrastructure for high-density, energy-efficient data centers.

Our immers​​​​ion cooling servers

Thanks to the dielectric fluid, immersion cooling ensures a much more uniform temperature distribution across components compared to other cooling methods. As a result, components are no longer subjected to the thermal stress caused by sudden temperature fluctuations.

Working in a conventional air-cooled data center can be physically demanding. Ambient noise can reach up to 90 decibels, leading to poor and dangerous working conditions. With an immersion cooling system, fans are out of the equation, therefore noise pollution is avoided.

Using liquid immersion cooling in a data center also results in a reduction of carbon emissions up to 39% and of water consumption up to 91%, according to our own research.

Power Usage Effectiveness (PUE) with Immersion Cooling

One of the most critical metrics for data center operators is PUE, which is the ratio of total facility energy consumption to the energy actually used by IT equipment.

A perfect PUE score is 1.0, meaning 100% of energy goes directly to computing. The global average for air-cooled data centers sits around 1.5 to 1.6, meaning up to 60% of energy is wasted on cooling and infrastructure overhead. Immersion cooling consistently achieves PUE scores as low as 1.02 to 1.05, making it one of the most energy-efficient cooling technologies available today.

Immersion cooling: for which applications?

One of immersion cooling's greatest strengths is the sheer breadth of industries it serves. Any sector that demands high-performance computing and low latency can benefit from immersion-cooled infrastructure.

• AI and LLM training: training large language models and frontier AI systems requires sustained, extreme compute density over long periods.
• Cloud computing and hyperscale data centers: high-density cloud infrastructure demands maximum compute per rack.
• Cloud gaming: real-time rendering and ultra-low latency requirements make thermal management critical.
• Healthcare and medical imaging: AI-powered diagnostics, genomics pipelines, and medical imaging analysis generate intense compute bursts that benefit from stable, efficient cooling.
• IoT and edge computing: compact, energy-efficient immersion cooling systems can be deployed in non-traditional environments, bringing high-density compute closer to the data source.
• Video surveillance and security: large-scale video analytics powered by AI require continuous, reliable processing.

Air Cooling vs Immersion Cooling

Comparison of the water consumption of two data centers

A measure system called WUE (Water Use Efficiency) was created to measure water and energy consumption in data centers. This metric is calculated by dividing data center annual energy source and site water usages (in Liters) by total IT power consumption.

		Data Center A Air cooling	Data Center B Immersion cooling
Electricity	Total IT Power Consumption	4.2 MW	3.36 MW
	Total Facility Power	5.7 MW	3.5 MW
Water	Daily site water usage	507 300 L	43 750 L
	Energy source water per year	94.07 million L	57 million L
	Site Water Usage per year	185.16 million L	15.97 million L
WUE	Site WUE	7.59 L/kWh	2.48 L/kWh

 

Comparison of the carbon footprint of two data centers

Carbon emissions come mainly from the extraction of raw materials and their transformation into electronic components, as well as the electrical production of electricity. As we know, the carbon footprint of each country differs.

In France, 1kWh of electricity is equal to 0.014 kg of CO2. Thanks to nuclear plants, France’s carbon footprint is lesser than the USA, which reaches 0.454 kg of CO2 for 1 kWh of electricity.

Carbon Usage Effectiveness (CUE) is a metric that calculates and determines the amount of carbon gas emitted by a data center daily. It was developed by the Green Grid and adopted around the world. It is calculated by dividing the total CO2 emissions equivalents of the facility’s energy consumption by the total IT energy consumption.

 

		Data Center A Air cooling	Data Center B Immersion cooling
Electricity	Capacity	12 000 servers	12 000 servers
	Average Power Consumption (per server)	350 W	280 W*
	Total IT Power Consumption	4.2 MW	3.36 MW
	Cooling Overhead	30%	2%
	Electrical Overhead	6%	1%
	Effective PUE (Power Usage Effectiveness)	1.36	1.03
	Total Facility Power	5.7 MW	3.5 MW
	Energy Consumption per year	50 million kWh	30.3 million kWh
CO2	USA Carbon emissions per year	22.7 million kgCO2	13.8 million kgCO2
	FRANCE Carbon emissions per year	5.2 million kgCO2	3.2 million kgCO2
CUE	USA Effective CUE	0.62 kg CO2/kWh	0.47 kg CO2/kWh
	FRANCE Effective CUE	0.14 kg CO2/kWh	0.12 kg CO2/kWh

*reduce due to fans removal