SLB agrees $4.1 billion deal for cooling specialist Kelvion to expand its AI data centre business
As rack power climbs past 100 kW, moving heat out of the building becomes the constraint, and the thermal chain is where the money is going.
Energy services company SLB has agreed to acquire thermal management specialist Kelvion in a deal valued at $4.1 billion, Data Center Knowledge reported on 31 August 2026. The transaction comprises about $3.4 billion in cash plus roughly $700 million of assumed Kelvion debt, and is expected to close in the first half of 2027.
Kelvion makes heat exchangers, dry coolers, hybrid and adiabatic cooling systems and waterless cooling equipment used across several industries. SLB, better known for oilfield services, has been building a data centre infrastructure business around modular construction and engineering. The report says the acquisition brings heat-rejection expertise into that offering, aimed at AI deployments where rack power densities now exceed 100 kW and some architectures approach 1 MW per rack, levels that push cooling requirements beyond cold plates and coolant distribution units alone.
SLB expects its combined data centre business to generate more than $2 billion of revenue in 2026 and has set a target of $4.5 billion to $5 billion by 2028, with adjusted EBITDA of $700 million to $800 million that year. The company anticipates about $120 million of annual EBITDA synergies within three years. The purchase price represents 11 times estimated 2026 adjusted EBITDA before synergies and 8.5 times after them, according to the report.
Alex Cordovil, research director at Dell'Oro Group, told the publication the deal turns SLB into an original equipment manufacturer in one of the fastest-growing parts of the thermal chain, and argued that lasting differentiation in this market comes from system design and controls rather than from individual components.
Why it matters
Cooling has moved from a facilities detail to a strategic asset, and companies from adjacent industries are buying their way in. The shift to liquid cooling inside the rack pushes the bottleneck outward to heat rejection, which is exactly where Kelvion's equipment sits. Consolidation suggests buyers want one supplier for the whole thermal path rather than a stack of vendors. Water use and grid connections will keep shaping where the next wave of AI capacity is built.
Cooling has moved from a facilities detail to a strategic asset, and companies from adjacent industries are buying their way in.
Cold plate on the chip
Coolant absorbs heat directly from the processor package inside the server.
Coolant distribution unit
Pumps and a heat exchanger move heat from the rack loop into the facility water loop.
Facility loop and heat exchanger
Plate heat exchangers isolate the clean indoor loop from the outdoor loop without mixing fluids.
Dry cooler, adiabatic or evaporative unit
Fans, and sometimes water spray, reject the heat to outside air; this is Kelvion's territory.
Outside air
The final sink; climate and water availability decide which rejection method is practical.
What you can learn from this
- Heat has to leave the building, not just the chip. Liquid cooling moves heat from a processor into a coolant loop, but that loop is only a transport mechanism. The heat then passes through a coolant distribution unit into a facility water loop, and finally into equipment such as dry coolers, cooling towers or chillers that reject it to the outside air. Every stage adds thermal resistance and cost, so rising rack density affects the whole chain, not only the cold plate. Designers size each stage for the heat load on the hottest expected day, not the average one.
- A heat exchanger transfers energy between fluids that never mix. Two fluids flow on either side of thin metal plates or tubes, and heat conducts through the metal from the hotter fluid to the cooler one. Effectiveness depends on surface area, flow rate and the temperature difference between the fluids, which is why exchangers become large and expensive when you need to shed a lot of heat across a small temperature gap. Data centres use them to isolate the clean loop inside the building from the outdoor loop.
- Dry, adiabatic and evaporative cooling trade water for electricity. A dry cooler uses only fans and outdoor air, so it consumes no water but struggles when the air is hot. Evaporative systems spray water to exploit the cooling effect of evaporation and reach much lower temperatures, at the cost of significant water consumption. Adiabatic and hybrid designs switch between modes depending on conditions, which is why they have become common where water is politically or physically scarce.
- Rack density drives every downstream decision. A rack drawing 100 kW turns almost all of that into heat that must be removed continuously. Air cannot carry that much heat at practical flow rates, which is what forces liquid cooling, and liquid cooling in turn forces new plumbing, higher floor loading and more heat rejection capacity outside. Site selection now weighs climate and water availability alongside power availability. Density also concentrates risk, because a pump or distribution unit fault can overheat a rack quickly, so redundancy becomes part of the design.
- Controls determine how efficiently the hardware runs. The same cooling plant can consume very different amounts of energy depending on set points, fan curves and how it responds to load and weather. Integration software decides when to switch from dry to wet operation or when to raise supply temperatures. That is why analysts point to system design and controls as the durable advantage: software and integration are harder to copy than a fan or a plate.
We teach this
How to use this in practice
- Trace the heat path for a machine you own. Pick a laptop, desktop or server and write down every stage heat takes from the die to the room: heat spreader, heatsink or cold plate, fan or pump, exhaust, room air, building ventilation. Watch temperatures under sustained load with a tool such as HWiNFO on Windows or
sensorson Linux to find which stage limits performance, and compare idle and loaded temperatures to see how much headroom the cooling leaves. Done looks like: a numbered list of stages with the bottleneck circled. - Calculate the heat load of a hypothetical rack. Take 100 kW, assume it all becomes heat, and compute the water flow needed to carry it away with a 10 degree Celsius temperature rise using the specific heat of water, about 4.18 kJ per kilogram per degree. Then compute the airflow needed for the same job using the specific heat of air, and write down the assumptions you made about inlet temperature, because changing them changes the answer. Done looks like: two numbers, in litres per second and cubic metres per second, and a one-sentence explanation of why air loses.
- Sketch a full liquid-cooled facility thermal chain. Draw cold plate, rack manifold, coolant distribution unit, facility loop, plate heat exchanger and outdoor heat rejection, then mark which stages use water and which use electricity. Add where the control system sits and what it decides, and mark which stages would need redundancy to survive a single pump or fan failure. Done looks like: a diagram with at least six labelled stages and a note on the failure mode at each.
- Read the cooling section of a public data centre sustainability report. Choose one large operator's most recent report, find its stated power usage effectiveness and water usage effectiveness figures, and write a short note explaining what each metric measures and what it leaves out. Done looks like: two definitions in your own words and the two figures you found, with the report's date.
Sources
- SLB's $4.1B Kelvion Deal Expands AI Data Center Push — Data Center Knowledge
Our reporting is an original summary; full coverage is at the links above.
Don't just read about it — build it.
Square 1 teaches the skills behind the headlines, with every line of your work graded by AI. Find your starting point in 3 minutes.
Get your free skill report