The short version: Immersion cooling beats every rival on paper. It achieves a lower PUE, higher rack density and has fewer moving parts. Yet direct-to-chip cold plate is becoming the go to architecture for hyperscalers. The reasons have little to do with thermodynamics and everything to do with warranties, serviceability, and a regulatory cloud forming over the fluids that two-phase systems depend on.
What immersion cooling does better
Immersion cooling submerges live servers in a dielectric (electrically non-conductive) fluid. Single-phase systems pump liquid through a heat exchanger. In a two-phase system the fluid is allowed to boil off and condense on a coil overhead, exploiting the latent heat of vaporization for better thermal control. In both methods the fluid is in direct contact with the silicon. This makes it highly efficient with heat capture approaching 100%.
Vertiv reports immersion deployments delivering up to 100 kW of cooling per rack, with potential for cooling racks approaching 900 kW. These are densities that are impossible for air and even cold plate struggle to reach. Single-phase immersion deployments commonly report PUE in the 1.03–1.10 band, with two-phase pushing toward 1.02. Compare this to 1.15–1.30 for direct-to-chip. For a sense of how aggressive that is, Lawrence Berkeley National Laboratory’s Center of Expertise for Data Center Efficiency frames liquid cooling broadly as the path past the efficiency ceiling. On a spec sheet, immersion cooling wins every time.
Why direct-to-chip quietly took the market instead
If immersion is so great, why is the technology with the worse PUE winning? In Direct-to-chip coolant is piped across metal plates bolted to the CPU and GPU, never touching the board. Overall liquid cooling has climbed toward 37% of deployments in 2026 from about 3% in 2021 and direct-to-chip accounts for roughly 70% of these deployments.
The reason is compatibility. Direct-to-chip fits into a more or less conventional server and a more or less conventional rack. The major GPU platforms ship with direct-to-chip reference designs. IEEE Spectrum’s survey of data-center liquid cooling makes the same point operators keep arriving at. The AI buildout is moving too fast to wait on an architecture that requires them to redesign the whole data center hall. Cold plate is the path of least resistance, and at hyperscale, least resistance usually wins.
Warranty and serviceability
Most enterprise server warranties and OEM support contracts are written around air-cooled or direct-to-chip operation. Dropping a server in a tank of dielectric fluid is, for many vendors, outside the supported operational specs. This can complicate or void warranty coverage. Operators need to shift to immersion-specific servers and support arrangements.
Then there’s serviceability. A failed hard drive in an air-cooled rack is a two minute hotswap. The same job in an immersion system means hoisting the server out of fluid, letting it drain, working on it wet, and re-immersing it. A much slower and messier workflow that needs trained hands and a place to do it. A Two-phase system adds another layer of complexity. There are vapor management requirements on top. These complexities rarely appear in the marketing literature.
The regulatory cloud over two-phase fluids
This is more serious as it threatens the entire branch of the technology. Two-phase immersion depends on engineered fluorochemical fluids. These fluids fall under the umbrella of PFAS, the “forever chemicals” that are becoming heavily regulated.
In December 2022, 3M, the dominant maker of the fluorinated fluids many two-phase systems were built around, announced it would exit all PFAS manufacturing by the end of 2025. The demand side is moving too. In 2023, five European countries submitted a universal PFAS restriction proposal to the European Chemicals Agency, one of the broadest chemical restrictions ever contemplated. For an operator considering a multi year, two-phase immersion cooling rollout this is a genuine concern.
This is however largely a two-phase problem. Single-phase systems typically run on hydrocarbon, synthetic, or bio-based dielectric fluids that sit outside the PFAS debate. However, as previously discussed, it is two-phase cooling that provides the best efficiency numbers.
What it means for operators planning AI capacity
So how does the future look for immersion? We foresee both technologies continuing to be deployed. Cold plate handles the bulk of enterprise, cloud, and AI deployments because it’s OEM-supported and serviceable, whereas immersion earns its place in specialized, ultra-dense corners where the density and PUE advantages outweigh the operational cost.
For an operator then, the question becomes not which architecture is best on paper. Read the fluid datasheet and the support terms before the thermal spec, model the wet-maintenance workflow honestly, and, if you’re looking at two-phase, get a straight answer on fluid sourcing past 2025. Immersion cooling’s numbers are real. So are the reasons it isn’t the default. Those reasons, not the thermodynamics, are what will decide where it lands.