“Especially in the last three to five years, this industry has seen strong growth in terms of capacity additions,” Sharma said, specifically in high density data processing and storage. “The growth brings with it greater demand for power consumption.”

Data center density is by far the highest in the United States, she noted, followed by Germany, the United Kingdom and China.

Sharma noted that according to the International Energy Agency, the energy consumption globally by data centers – excluding for crypto – has increased from 200 terrawatt-hours in 2015 to 240-340 TWh in 2022, for an increase of 20%-70%.  A terawatt-hour, or TWh, is a unit of energy representing one trillion watts of power used for one hour. The use of power for crypto mining energy increased from 4 TWh in 2015 to 100-150 TWh in 2023.

In addition to this increase in data center power demand, the water crisis in many countries is another factor. She noted that the United Nations estimated that by 2025, 50% of the world’s population is projected to live in water-stressed areas, making data center water usage a key environmental area to prioritize change.

The global growth of date for a wide variety of application is what’s generating data center growth. “Globally, the amount of data created is skyrocketing,” Sharma said.

For context, she explained that in 2010 when the Internet was maturing and globalization of data processing started, data amounted to about 2 zettabytes. By 2023, that total reached 120 zettabytes, she said, and it is projected to grow to as much as 2,100 zettabytes by 2035. A single zettabye contains one billion terabytes and is a term used to measure large aggregate amounts of data, including global data totals.

Some of the factors driving growth in data centers include an explosion in cloud-based computing, growing demand for edge and micro data centers, increasing global use of smart phones and tablets and what’s termed a “gold rush” for artificial intelligence. The latter includes large language model and generative AI, machine learnings and a big technology investment surge.

“It is expected data creation is going to see an exponential growth in the next 10 to 15 years,” Sharma said. With the data center growth, she said, the information technology infrastructure industry is under immense pressure to reduce its power consumption, from the standpoint of regulations as well as saving on costs.

Within the immersion cooling category, different types of cooling systems have emerged. She said the two variants designed to handle the highest thermal load density per rack (up to 250 kW) are single phase immersion cooling using fluorocarbon and two-phase immersion cooling. For thermal loads up to a little more than 150 kW is single phase immersion cooling using oil.

The value chain for data centers includes a variety of segments – data centers, engineering deployers, cooling specialists, server manufacturers, chip manufacturers and a growing number of immersion cooling fluids suppliers. She said that so far more than 30 such fluids have emerged, with examples including products made using gas-to-liquid base oil, polyalphaolefins, low viscosity mineral oils and esters, among others. Among the many fluid suppliers, examples include Shell, ExxonMobil, Castrol, TotalEnergies, SK Lubricants, Eneos, Honeywell, Svensqo, Mivolt, Engineered Fluids, Chevron Phillips Chemical, Lubrizol, Chemours and Cargill. Sharma noted that the number of immersion cooling fluid supplies is increasing fast in the market, as they establish partnerships and offer different products.

In the current data center market, immersion cooling is driven by racks with power densities higher than 30 kW. That comprised about 15% of global average rack power densities in 2023, meaning about 0.5% immersion cooling penetration or about 4,000 standard racks. “Immersion cooling is at its initial stage of its life cycle,” she noted, and added it’s important to note that not all IT equipment can be submerged.

“Immersion cooling currently has a few scaled deployments and also some test beds established to test this technology,” she said. “There are hyperscale data centers where this technology is being tried at the rack end, and some edge set ups are trying this technology and co-location facilities. All of that included, estimated of more than 4,000 standard racks have been submerged to data with this technology that’s been deployed.”

 Kline estimated current volumetric size for “in-use” immersion cooling fluids at about 4 million or 5 million liters in 2023. Those are in high performance computing applications and are also in some crypto mining setups in data centers.

“Immersion cooling in data centers is certainly poised to grow,” out to 2033, she said. “We estimate demand for immersion cooling fluids is expected to grow by more than 10 times in the future from the current levels.” A tipping point is expected between 2028 and 2033, when chip and server manufacturers are expected to recommend immersion coolant specifications. A proven return on investment is also expected to help move hyperscalers to move to immersion cooling.

She noted that the market is currently in its formative stages. “All cooling requirements are not the same in data centers,” she said. “In the long term future, with the shift expected to become more intense as this technology starts to mature, it will be more widely adopted by larger players in the data center industry. “Stronger growth is expected in the longer term.”

Since 2007, data center average annual power usage effectiveness has declined and essentially flatlined since around 2013, due to changes in technologies, hardware and software at data centers. She said factors pushing the market towards finding more effective power consumption include a continuing rise in rack power density that causes servers to run hotter, chip power density increasing to meet power-hungry AI applications and a rise in real estate prices that drives the need for more compressed computing. “There is a strong need in the data center industry for reducing the power that is going into keeping the equipment powered and overall thermal management,” she added.

She explained that high thermal loads in typical data centers today stand at 15-30 kilowatts per rack, which is addressed by legacy cooling systems. These are typically air cooling systems that are considered best-in-class. However, in the next few years, to drive better CPU and graphic p processing unit performance, a higher thermal load would need efficient cooling, she said. Air cooling is not considered suitable for AI clusters about 20 KW per rack, while a bespoke cooling distribution design complicates deployment.

Current air cooling consists of massive cooling towers installed on the roof of a data center. Immersion cooling – referred to as a “disruptive” cooling technology – uses high power density servers that are immersed in cooling tanks filled with dialectic fluids. “An increasing number of data centers are looking at liquid cooling and immersion cooling as a viable solution,” she said.

These emerging technologies are being evaluated for the needs of tomorrow. “The decisions data centers take are going to be based on certain factors, such as computing requirements, the cost of switching from a legacy cooling system to a new cooling system and new technology,” she said. Another factor could be the push data centers may see in different part of the world for meeting different requirements in different country markets, she added.

The webinar was based on insights from Kline’s “Immersion Cooling Fluids in Digital Infrastructure” report.