Immersion cooling used to be a conversation about tanks. It has become a conversation about liquid, because the fluid decides the warranty, the fire strategy, the service intervals, and what happens to the whole system in ten years. That has drawn in suppliers who were never in the data center business.
The oil and lubricant companies
Castrol, Shell, and South Korea's S-Oil have all moved into immersion fluids, and their advantage is manufacturing and distribution at a scale no specialist can match, plus decades of experience with how fluids behave over years rather than months. All three are visible in 2026: Castrol signed an agreement to package immersion cooling for Japanese data centers with Woodman and Eole, Shell's gas-derived fluid is in a twelve-month trial inside a working Keppel facility in Singapore, and S-Oil is running pilots in Korea with the equipment maker Global Standard Technology.
The system vendors who supply their own fluid
Submer and GRC sell fluid alongside the tanks, and the argument for buying both from one supplier is that a single company then owns the outcome. When a fluid degrades or a seal fails, there is no question of whose problem it is. The trade is that the fluid choice is largely made for you, and switching later means renegotiating the whole system rather than one line item.
The two-phase fluid problem
Two-phase immersion, where the fluid boils to remove heat, has been constrained less by engineering than by chemistry. The fluids that work best have historically been fluorinated compounds, and those are the subject of tightening environmental regulation and, in one prominent case, a manufacturer exiting production. Anyone evaluating two-phase immersion should treat fluid availability over the system's life as a primary risk rather than a procurement detail. Single-phase systems, which do not boil the fluid, use hydrocarbon-based oils and do not carry the same exposure.
What actually decides the choice
Fluids are compared on thermal performance, and thermal performance is rarely what causes trouble. The problems are material compatibility, where a fluid attacks a seal, a label, or the plastics inside a server; the warranty position, since a server vendor has to agree that submerging their hardware is acceptable; and the local fire code, which governs what may be stored in what quantity in an occupied building. The Japanese partnership above made a point of the fluid being handleable under national fire rules, which tells you how often that is the obstacle.
- Written warranty acceptance from the server vendor, for that fluid
- Material compatibility across seals, gaskets, labels, and plastics
- Fire classification and the quantity limits it imposes on the room
- Service life, testing requirements, and the cost of replacing a full tank
- Availability over the system's life, especially for two-phase fluids
How to read supplier efficiency claims
Fluid suppliers publish large numbers, and they are usually comparisons against air cooling rather than measurements of a specific site. Shell's published benchmarks claim up to 48 percent better energy efficiency and up to 40 percent lower capital and operating cost, and the Japanese partnership claims a power usage effectiveness of about 1.10. These may well be achievable, but each carries an implicit set of conditions. The reason the Keppel trial is worth following is that it puts one of these claims inside an operating facility in a hot, humid climate, which is the hardest test of them.