The rapid iteration of artificial intelligence (AI) technology is driving global AI computing infrastructure into a cycle of large-scale construction. During the training of large models and the deployment of high-density computing clusters, core hardware such as servers and chips operates continuously under high loads, generating massive amounts of heat. The shortcomings of traditional air cooling solutions, including limited cooling efficiency and high energy consumption, are becoming increasingly prominent, making it difficult to meet the cooling needs of high-density computing equipment.
In this context, liquid cooling, with its advantages of high efficiency, low energy consumption, and suitability for high-computing-power scenarios, is gradually replacing air cooling as the mainstream upgrade direction for data center cooling. Fluorine-containing PFAS chemicals, as the core medium of liquid cooling systems, have seen a rapid increase in industry application scale, becoming an invisible key material behind AI infrastructure. These synthetic compounds, known as "forever chemicals" — PFAS, demonstrate exceptional cooling efficiency in two-phase immersion cooling systems. Low-boiling-point PFAS fluids absorb heat from servers, vaporize, and then condense for recycling. Their heat transfer capability can be 10 to 100 times that of single-phase solutions. For ultra-high-density cabinets that often bear thermal loads of tens of kilowatts, this performance advantage is highly significant.
The Dual Nature of Performance and Cost
The appeal of PFAS in data center scenarios stems from the extreme strength of its carbon-fluorine bonds. This chemical bond endows PFAS with excellent thermal stability, chemical inertness, and water resistance, allowing it to maintain stable performance under extreme operating conditions while providing absolute insulation, enabling safe direct contact with electronic components. Furthermore, compared to traditional water cooling, PFAS water-free liquid cooling solutions can completely eliminate the reliance on water resources, significantly reducing the water consumption pressure on data centers and aligning with energy and water conservation policy requirements in many regions worldwide.
In high-end AI servers, fluorinated liquid solutions are currently one of the most mature cooling choices for deployment, capable of stably supporting ultra-high-power cabinets of over 50 kilowatts. However, it is precisely the strength of these carbon-fluorine bonds that creates the concerning side of PFAS. These compounds are almost impossible to degrade in the natural environment and can continuously accumulate in the air, water, soil, and living organisms.
PFAS has been detected in the blood of over 99% of the global population, and epidemiological and toxicological studies have linked it to cancer, liver and kidney damage, elevated cholesterol, and developmental toxicity. According to estimates by the chemical watchdog ChemSec, the world's largest PFAS manufacturers profited approximately USD 4 billion in 2022, but the social costs—including healthcare, cleanup, and contaminated water sources—reach as high as USD 17.5 trillion annually. This stark contrast reveals a fundamental contradiction: a massive asymmetry exists between the commercial returns generated by PFAS and the public costs it causes.
Global Giants Concentrate on Capacity Expansion to Seize the New AI Cooling Track
In response to the cooling demand brought by AI infrastructure construction, major global PFAS producers are accelerating capacity layout. Japan's Daikin plans to expand its fluoroelastomer capacity to more than three times its 2022 level. Its new plant in Kashima is expected to be completed in August 2026, with commercial supply starting in 2027. France's Arkema has already put its PVDF (Polyvinylidene Fluoride) capacity expansion project in Kentucky, North America, into operation in June 2026, and plans to increase the PVDF capacity at its Changshu base in China by 20%. This project is expected to be completed and operational by 2028, forming a production network covering North America, Europe, and Asia.
The US-based Chemours is also one of the leading producers in the PFAS field. According to statistics from ChemSec, Chemours produces or uses 57 PFAS substances, with approximately half to two-thirds of its revenue coming from PFAS-related businesses. Its Opteon refrigerant product line achieved a 56% annual sales growth in 2025 and is expected to maintain double-digit growth in 2026, with the surging demand for data center cooling being one of the core drivers. Chemours has launched new refrigerant products for data center immersion cooling and has received priority review status from the US Environmental Protection Agency (EPA).
Belgium's Syensqo (spun off from Solvay) is also expanding its PFAS capacity. Its Galden series of perfluoropolyether (PFPE) fluids are explicitly promoted for high-performance computing and data center immersion cooling, with a boiling point range covering 55°C to 320°C, which can reduce the data center PUE (Power Usage Effectiveness) to below 1.1. In addition, companies such as AGC, Dongyue Group, and Gujarat Fluorochemicals have also been listed by ChemSec as producers currently expanding their capacity.
Industry Divergence and Ambivalence
While PFAS producers are expanding their capacity, a clear divergence is emerging within the industry. 3M completed its comprehensive exit from PFAS manufacturing as scheduled at the end of 2025, halting the production of its flagship Novec and Fluorinert series of fluorinated liquids across the board. Prior to the exit, these two series held a dominant share of approximately 70% in the electronic-grade fluorinated liquid market, leaving a supply gap corresponding to a global market restructuring space of approximately USD 1 to 1.5 billion. BASF also plans to exit PFAS production by 2028, while Archroma is actively developing PFAS-free alternatives. This divergence reflects the deep-seated contradiction in the industry's attitude towards PFAS.
On the one hand, the expansion of AI infrastructure has created deterministic demand growth, providing coolant suppliers with strong incentives to expand capacity and seize market share. On the other hand, regulatory pressure and public opinion are rapidly tightening, making the commercial prospects of PFAS full of uncertainty.
For data centers that promise a service life of 10 to 15 years, choosing a coolant that might be banned means facing the risk of being forced to replace the system during the equipment's lifecycle. The core argument Chemours uses to defend its PFAS cooling products is that two-phase immersion systems are closed loops with very low fugitive emissions during operation.
However, environmental organizations point out that no closed system is absolutely sealed. Leaks from valves, seals, pumps, and expansion tanks could allow the coolant to enter the soil or stormwater systems directly, and the decommissioning and replacement of equipment would similarly increase the risks of emissions and disposal. The case of the AI data center project in Fouju, France, which emits approximately 15 tons of PFAS into the atmosphere annually, has further exacerbated public doubts about the closed-loop claims.
Tightening Regulations and the Rise of Alternative Routes
As the application of PFAS in AI data centers becomes increasingly widespread, its hidden environmental risks have attracted high attention from global environmental organizations and regulatory agencies. Industry controversies continue to ferment, and compliance pressure is gradually transmitting to the upstream and downstream of the industrial chain.
In July 2026, 17 environmental organizations jointly wrote to the US Environmental Protection Agency (EPA), requesting the rejection of Chemours' application for accelerated review of its new PFAS chemical, Opteon 2P50. These organizations accused Chemours of underestimating the toxicity and climate risks of the chemical in its application, relying on laboratory rat strains with weak toxicological responses to PFAS to argue for lower health risks, while a large amount of toxicological data is missing.
PFAS, widely used in two-phase fluids, faces stricter reporting requirements from the US EPA and bans in multiple states. For data centers that require long-term stable operation, this regulatory uncertainty is tipping the balance of economics. In January this year, the French public filed a lawsuit against the two chemical giants Arkema and Daikin over PFAS pollution issues, further highlighting public concerns about the development of the PFAS industry.
Meanwhile, alternative technology routes are accelerating towards maturity. Germany's WACKER Chemie has launched Helisol EC, a silicon-based dielectric coolant, claiming it is less harmful than PFAS fluids. Synthetic hydrocarbon coolants do not contain PFAS, and their environmental attributes align with green computing power policies. Their procurement and operation and maintenance costs are far lower than those of fluorinated liquids, and they are currently being applied in mainstream commercial intelligent computing scenarios of 20 to 40 kilowatts. However, their boiling points generally exceed 180°C. Within the normal operating temperature range of servers, they can only rely on liquid convection sensible heat to dissipate heat, unable to achieve boiling phase change heat transfer. There is a natural constraint on the upper limit of cooling, making it difficult to stably support ultra-high-density immersion cabinets of over 50 kilowatts at this stage.
Chinese companies are playing an increasingly important role in this wave of substitution. Relying on a complete fluorine chemical industry chain, Juhua Group has planned a coolant capacity of 5,000 tons per year; Capchem's fluorinated liquid products have been mass-produced and used in major global wafer fabs, while also laying out a Non-PFAS coolant product line; Dongyue Group's hydrofluoroether and perfluoropolyether products have passed the verification of NVIDIA's A100 and H100 platforms. The product strategies of these companies present a pragmatic balance: seizing the market gap in fluorinated liquids left by 3M's exit, while also reserving space for alternative solutions in case PFAS regulations are further tightened.
Conclusion
From the perspective of industrial development, the popularization of PFAS is a phased technological choice under the rapid development of AI computing power. Its highly efficient and stable cooling characteristics have effectively solved the cooling pain points of high-density computing clusters, filled the shortcomings of traditional cooling solutions, and provided critical support for the large-scale deployment of AI infrastructure. However, it cannot be ignored that the long-term environmental costs brought by "forever chemicals" far exceed the short-term economic benefits of the industry and cannot support the long-term sustainable development of the AI industry.
In the future, the core of development in the AI computing cooling track will no longer be the mere pursuit of improved cooling efficiency, but rather finding a balance among technological efficiency, industrial costs, and environmental safety. As regulatory policies continue to tighten and environmental protection technologies iterate continuously, the research, development, and deployment of new cooling media with low pollution and biodegradability will accelerate, driving AI computing infrastructure to truly achieve efficient, green, and sustainable development.