AI Data Centers Are Driving New PFAS Demand—Despite the Push to Phase Them Out

AI infrastructure is creating two connected sources of demand for PFAS: the semiconductor supply chain that produces AI chips, and the thermal-management systems needed to run increasingly dense servers. That growth is occurring as regulators, environmental groups, and some companies push to reduce or eliminate these persistent chemicals.

ChemSec’s September 2026 review identifies AI and data-center infrastructure, semiconductor manufacturing, and lithium-ion batteries as the three principal demand drivers behind planned PFAS-capacity expansion. It reports that, with few exceptions, major producers are expanding capacity, and names Chemours, Syensqo, Arkema, Solstice, Daikin, and AGC among companies linking investments to AI infrastructure or chip fabrication. 31

How AI increases PFAS demand

1. More AI hardware means more semiconductor processing

PFAS are used in semiconductor manufacturing applications including surfactants, etching and cleaning processes, and chemically resistant tubes and fittings, according to ChemSec. As demand rises for AI accelerators, memory, and the broader equipment needed to manufacture them, that can raise demand for these specialty materials upstream at fabs. 31

The connection is broader than the server itself. NRDC notes that PFAS can be associated with cooling, fire suppression, and the manufacture of semiconductors and other electronic components used in data centers. 12

2. Dense servers make liquid cooling more attractive

Air cooling becomes harder as computing density and heat loads rise. Immersion cooling places electronics in an electrically nonconductive dielectric liquid rather than relying solely on air. In a single-phase design, the fluid circulates without boiling; hydrocarbons such as poly-alpha-olefins are commonly used. In two-phase cooling, a fluorinated refrigerant can boil at the hot component, carry heat away as vapor, then condense and recirculate. 38

That second approach is a potential PFAS-demand channel. ChemSec specifically identifies immersion-cooling fluids and thermal management for AI and data-center infrastructure as a driver of capacity expansion. 31

Why PFAS are technically useful—and environmentally contested

PFAS are valued in demanding industrial settings for properties such as heat tolerance, water and oil repellency, chemical resistance, and electrical insulation. Those characteristics help explain their use where chipmaking chemicals, high temperatures, and sensitive electronic components must be handled reliably.

But the same durability creates the central environmental concern: PFAS do not readily break down in the environment. Food & Water Watch describes data-center cooling systems as a possible additional use pathway for persistent PFAS, even where those systems may reduce rather than eliminate water use. 2

This does not mean every liquid-cooling fluid is PFAS-based, or that every data center uses immersion cooling. It does mean AI’s cooling challenge can become a material-selection issue with long-lived environmental consequences.

Which PFAS producers are expanding—and who is leaving?

Expansion: Chemours and other major suppliers

ChemSec’s research says the majority of major PFAS producers are expanding capacity in response to AI/data-center infrastructure, semiconductor manufacturing, and batteries. Its assessment identifies Chemours, Syensqo, Arkema, Solstice, Daikin, and AGC as companies explicitly framing relevant investments around AI or chip fabrication. 31

Chemours provides the clearest public example of an AI-related commercial strategy:

• The company said a recent expansion of Teflon PFA production capacity was completed in 2024 and characterized the material as critical to semiconductor manufacturing. 21
• Chemours has partnered with Navin Fluorine International to manufacture its Opteon two-phase immersion-cooling fluid as part of an expanded liquid-cooling venture aimed at the heat, energy, and water demands of advanced data centers and AI hardware. 25
• It also announced an agreement with DataVolt to develop and demonstrate liquid-cooling approaches, including two-phase direct-to-chip and two-phase immersion systems. 24

Chemours has projected commercialization of Opteon two-phase immersion cooling by 2026, subject to regulatory approvals. Its investor presentation also promotes potential cooling-energy reductions, but these are company performance claims rather than independently verified outcomes for every deployment. 21

Withdrawal: 3M’s exit changes the supply picture

3M announced in 2022 that it would phase out PFAS manufacturing by the end of 2025. Its Novec and Fluorinert product lines had applications in cooling, semiconductor manufacturing, and fire prevention, according to Data Center Dynamics. 28

The result is not a simple, industry-wide exit. It is a market reshuffling: 3M’s departure coincides with other suppliers seeking to serve semiconductor and advanced-cooling demand. 2831

How big is the AI-PFAS market?

A credible public estimate for the portion of PFAS demand attributable specifically to AI infrastructure is not available in the supplied evidence. ChemSec’s analysis focuses on expansion plans and demand drivers, not a quantified AI-only production volume. 31

The broader immersion-cooling-fluid market offers context, but it should not be treated as a PFAS market. MarketsandMarkets estimates the data-center immersion-cooling-fluids market will grow from $180 million in 2025 to $830 million by 2032, a projected 23.9% compound annual growth rate. That category includes direct-to-chip and immersion fluids and is not limited to fluorinated or PFAS-based products. 41

Regulation and PFAS-free alternatives

Policy and procurement pressures are increasingly encouraging alternatives. In the United States, EPA began prioritizing reviews of new chemicals intended for data-center projects or related manufacturing under the Toxic Substances Control Act review process. 16 At the same time, reporting requirements and possible future restrictions are prompting operators to consider exposure to PFAS in their supply chains. 35

Alternatives already exist for some cooling uses. Single-phase immersion often uses hydrocarbon dielectric liquids, while the broader market is developing non-fluorinated synthetics, esters, silicones, and bio-based fluids. 3842 Direct-to-chip cooling and improved air-cooling designs can also reduce the need for two-phase immersion in certain deployments.

However, the evidence does not establish that non-PFAS options are drop-in replacements for every semiconductor process or every high-density two-phase cooling application. Material qualification, reliability, safety, and system design all matter.

The AI infrastructure trade-off

For data-center builders, the issue is not simply whether to use more liquid cooling. It is whether efficiency and water-management goals should be met with persistent fluorinated materials where alternatives may be available, still maturing, or unsuitable for a particular use.

ChemSec argues that AI demand risks locking in a new generation of PFAS capacity just as broader phaseout efforts gather momentum. 31 NRDC’s lifecycle framing adds that the PFAS question extends beyond a server room to chip fabrication, cooling, fire suppression, and end-of-life handling. 12

The most important distinction is therefore between AI-driven demand for advanced thermal management and a presumption that PFAS are the only way to deliver it. The former is accelerating; the latter remains a technical, regulatory, and environmental choice.