Two-Phase Immersion Cooling Systems for AI & HPC
Two-phase immersion cooling is the most advanced liquid cooling technology available for high-compute data center infrastructure today. Like single-phase immersion cooling, it works by fully submerging servers in a dielectric liquid, but it goes one step further, leveraging the physics of phase change to achieve unmatched cooling efficiency, extreme compute density, and a PUE as low as 1.01. For AI training clusters, LLM infrastructure, and HPC environments pushing the boundaries of compute density, two-phase immersion cooling is the definitive answer.
Two-Phase Immersion Cooling: How Does It Work?
The initial setup of two-phase immersion cooling is similar to single-phase: servers are fully submerged in a dielectric liquid and enclosed inside a sealed tank. However, the key difference lies in the type of fluid and the cooling mechanism itself.
Two-phase systems use a fluorocarbon-based dielectric liquid with a boiling point of around 50°C. As the servers generate heat, this fluid absorbs it and begins to boil, transforming from liquid to vapor and passively transferring heat away from the components in the process. This phase change is what makes two-phase cooling so exceptionally efficient: the latent heat absorbed during vaporization removes far more thermal energy than a liquid alone could.
The vapor then rises naturally to a condensing coil at the top of the tank, where it is cooled back into liquid state and falls back down to re-enter the cooling cycle. The entire process is fully passive: no pumps, no fans, no complex secondary circuits. It results in an ultra-efficient, low-maintenance cooling loop. This cycle eliminates all mechanical cooling hardware, achieving the lowest PUE scores in the data center industry: 1.01 to 1.02.
Is Two-Phase Immersion Cooled System Better? 5 advantages
Dual-phase immersion cooling represents the pinnacle of data center cooling technology. Here is why it outperforms both air cooling and single-phase liquid immersion cooling across every key metric.
Low PUE
Two-phase immersion cooling tanks require no fans and no air conditioning, eliminating auxiliary power consumption almost entirely.
The result is an average PUE of 1.01 to 1.02, the lowest achievable in the data center industry, compared to 1.03 for single-phase immersion cooling and approximately 1.6 for traditional air cooling.
Easier Maintenance
Technicians can remove server boards directly from the tank without needing to drain the fluid first, unlike single-phase systems where the entire liquid volume must be drained before any intervention.
This reduces maintenance downtime and simplifies servicing operations considerably.
Extended Component Lifespan
The dielectric fluid is maintained at a constant boiling state, meaning component temperatures remain stable at all times.
Combined with the complete elimination of fans and their associated vibrations, this thermal stability extends the lifespan of server components, reducing hardware replacement costs and improving long-term infrastructure ROI.
High Computing Density
The phase change mechanism of two-phase immersion cooling dissipates heat so effectively that server components can be packed at densities far beyond what air or single-phase liquid cooling can support.
Cooling Efficiency
The latent heat absorbed during the liquid-to-vapor phase change allows two-phase immersion cooling to dissipate heat far faster and more efficiently than both air and water cooling.
Cost of Dual-Phase Immersion Cooling vs Single-Phase
While two-phase immersion cooling delivers superior PUE and cooling efficiency, it is important to acknowledge that it comes with a higher upfront investment than single-phase systems. The fluorocarbon-based dielectric fluid used in two-phase cooling is significantly more expensive than the mineral or synthetic oils and the sealed tank design adds to the initial CAPEX.
However, when viewed through the lens of TCO, the picture changes considerably. The near-perfect PUE of 1.01 to 1.02 translates into long-term energy savings, particularly at scale. For large AI training clusters or hyperscale facilities running thousands of GPU servers 24/7, the energy cost differential between a PUE of 1.02 and 1.6 (air cooling) can represent millions of euros per year. Combined with extended component lifespan, reduced maintenance costs, and no need for CRACs, CRAHs or chillers, two-phase immersion cooling typically delivers a faster ROI than its upfront cost suggests.
Single-Phase vs Two-Phase: Comparison Table
Single-Phase | Two-Phase | |
Fluid type | Mineral oil / synthetic oil | Fluorocarbon-based |
Phase change | No | Yes |
PUE | ~1.03 | ~1.01 to 1.02 |
Maintenance | Full drain required | Board removal only |
Compute density | High | Extreme |
Fluid cost | Lower | Higher |
Complexity | Lower | Higher |
For Which Industries is Two-Phase Immersion Cooling Useful?
Two-phase immersion cooling is purpose-built for the most thermally demanding environments in existence. Any industry running sustained, high-density GPU or HPC workloads stands to benefit significantly from its extreme cooling efficiency and near-perfect PUE.
AI, LLM Training & ML
The exponential growth of AI model complexity is pushing GPU thermal output to levels that single-phase cooling struggles to handle sustainably.
Training frontier LLMs such as GPT-scale or Llama-scale architectures requires weeks of continuous full-GPU operation, generating thermal loads that demand the absolute best in cooling technology.
Two-phase immersion cooling's passive phase-change mechanism handles these extreme, sustained loads without any mechanical intervention, while its PUE of 1.01 to 1.02 ensures that the massive energy consumption of AI training infrastructure is kept as efficient as physically possible.
Financial Services & HFT
In high-frequency trading, microseconds matter. The thermal stability delivered by two-phase immersion cooling goes beyond what single-phase systems can achieve.
The constant boiling state of the fluorocarbon fluid maintains component temperatures within an exceptionally narrow range, virtually eliminating any risk of thermal throttling even under the most intense, sustained compute bursts.
This level of thermal precision is critical for trading algorithms where a single millisecond of performance degradation can translate into significant financial losses.
Aerospace & Defense
Aerospace and defense environments impose constraints that go far beyond those of a standard data center, extreme temperatures, vibration, shock, and the absolute intolerance of downtime make conventional cooling technologies unsuitable.
Dual-phase immersion cooling's fully passive mechanism has no moving parts whatsoever. This makes it resilient in harsh operating conditions, while its hermetically sealed tank design provides an additional layer of protection against dust, humidity, and environmental contamination.
Edge & Cloud Computing
While single-phase immersion cooling already outperforms air cooling in edge and cloud environments, dual-phase takes it to another level.
Edge data centers and hyperscale cloud facilities face relentless pressure to increase compute density while reducing their energy and physical footprint.
Two-phase immersion cooling's passive phase-change cycle achieves rack densities that simply cannot be matched by any other cooling technology, making it the only solution capable of meeting the extreme density requirements.
Focus on the Atlas 1.8GG: The First Server Designed for Dual-Phase Immersion
Designed for dual-phase, not adapted to it
Most immersion cooling servers on the market today are either air or single-phase servers that have been retrofitted or adapted for dual-phase use: a compromise that introduces engineering trade-offs and limits performance potential. 2CRSi took a different approach.
The Atlas 1.8GG is the first server purpose-built from the ground up for dual-phase immersion cooling. Designed entirely by our R&D teams with two-phase constraints at the core of every engineering decision, it eliminates all the adaptation problematics that plague retrofitted solutions, delivering a cleaner, more reliable, and more thermally optimised result.
Thanks to this native dual-phase architecture, the Atlas 1.8GG supports up to 8 GPUs in a hyper-dense configuration, a feat only made possible by the extreme heat dissipation capacity of dual-phase immersion cooling.
