The riseof artificial intelligence is driving up demand for semiconductors and computing infrastructure. However, their manufacturing and cooling still rely on fluorinated substances, some of which belong to the broad family of PFAS. An investigation by ChemSec, reported by The Guardian, indicates that several major manufacturers plan to expand their capacity, citing, in particular, the needs of chips and data centers. This development reignites a complex debate: how can we reduce chemical persistence and emissions without neglecting the technical functions for which alternatives remain difficult to identify?
What Just Happened
On September 14, 2026, The Guardian reported on the findings of a ChemSec investigation into the world’s top ten PFAS producers. According to the nongovernmental organization, nearly all of them plan to increase their production or capacity. In their corporate communications, several of these investments are linked to demand for semiconductors, batteries, cables, cooling systems, and data centers.[1][2]
Two examples illustrate this trend. In June, Arkema announced a 15% increase in its U.S. production capacity for PVDF, a fluorinated polymer used primarily in semiconductors and the expansion of data centers. In August, Chemours launched two refrigerants for stationary chillers and data centers, highlighting their low global warming potential.[5][6]
However, these announcements do not prove that AI is the sole cause of a global resurgence in PFAS use. The same materials are used in electronics, the automotive industry, energy, healthcare, and numerous industrial processes. The strongest finding is more limited: the expansion of AI infrastructure is now one of the business arguments used to justify new production capacity for fluorinated substances.
Why AI is a factor, even though it isn't the only cause
Large-scale models require accelerators, high-speed networks, memory, and cooling systems. As operators build more specialized data centers, demand is rising throughout the value chain: material extraction and purification, wafer fabrication, photolithography, etching, assembly, wiring, and thermal management.
The chemical footprint of AI therefore exists on two levels. It is indirect in the semiconductor factories that produce the components. It can be more direct in the cooling equipment and circuits, cables, seals, or certain fluids used in data centers. This footprint is less visible than electricity, water, or carbon, as operators rarely publish inventories of substances that are comparable across sites.
The term PFAS also refers to a very broad family of perfluoroalkyl and polyfluoroalkyl substances. Their structure, mobility, toxicity, uses, and legal status may vary. Talking about PFAS as a single product therefore leads to overly general conclusions. Conversely, their shared high persistence justifies a family-based approach and a focus on emissions throughout the entire life cycle.[3][8]
Where are fluorinated substances used?
The technical inventory published by the Semiconductor Industry Association’s PFAS Consortium confirms this diversity of uses. However, it is an industry-based source, produced by companies affected by the proposed restriction. While it provides useful documentation on the functions and substitution constraints, it does not constitute an independent assessment of their impacts.[3]
What the investigation actually allows us to conclude
ChemSec is an advocacy organization dedicated to the replacement of hazardous substances. Its investigation is a valuable resource for identifying industry strategies, but its findings must be cross-checked against corporate announcements, regulatory data, and environmental measures. The Guardian presents this journalistic work, but fails to address the main gap: the lack of a global public database linking production volumes, end uses, and emissions attributable to AI.[1][2]
A low climate impact does not mean there is no persistence
In the refrigeration industry, manufacturers often highlight the global warming potential (GWP) of refrigerants. This metric estimates their contribution to climate change relative to carbon dioxide. Reducing the GWP is helpful, but it does not automatically address issues such as persistence, degradation products, ecotoxicity, or water contamination.
The two refrigerants announced by Chemours illustrate this distinction. The company reports a GWP of approximately 1 for Opteon ZE and approximately 293 for Opteon 515B, well below that of some legacy refrigerants. This claimed climate performance, however, is not sufficient on its own to determine their full environmental impact or their classification as PFAS according to the applicable regulatory definitions.[6]
It is also important to distinguish between direct cooling using a dielectric fluid, water loops, refrigeration units that use a refrigerant, and air cooling. Therefore, not all data centers contain the same substances, have the same exposure pathways, or pose the same risks of leakage.
Why Substitution Remains Challenging
Chip manufacturing involves a combination of temperature, vacuum, plasma, corrosive conditions, extreme purity, and nanoscale dimensions. Substitution is not simply a matter of finding a molecule with a similar function; it must also preserve production yield, component reliability, operator safety, and compatibility with already-qualified equipment.
The validation process can be lengthy, as even a minor modification can introduce defects or reduce a component’s service life. Manufacturers therefore advocate for temporary exemptions when no solution is available on the required scale. This technical constraint is real, but it does not exempt manufacturers from establishing substitution plans, timelines, rejection criteria, and documented evidence of the essential nature of each use.
A responsible approach combines three priorities: eliminating non-essential uses for which an alternative exists; containing and treating emissions from uses that are still essential; and investing in less persistent processes and materials. The issue is not a choice between digital innovation and environmental protection, but rather integrating the chemical cost into architectural and procurement decisions.
Health and Environmental Risks: What We Know
Persistence is the most concerning common factor: many PFAS degrade very slowly and can accumulate in the environment. Certain well-studied substances have been linked to health effects, particularly on the immune system, development, metabolism, and certain cancers. While these findings cannot be uniformly applied to every family member, they justify prevention, monitoring, and the reduction of exposure.[7][8]
For semiconductors, relevant pathways include air emissions, process water, sludge, waste, and incidents. For data centers, they include upstream manufacturing, fluid leaks, maintenance, fires, and end-of-life equipment. The risk therefore depends as much on the properties of the substance as on the quantity, containment, and exposure.
To date, publicly available data do not allow for the creation of a robust inventory of PFAS consumed or released by AI data centers alone. This limitation must be made clear: while it prevents us from precisely quantifying AI’s share, it does not justify delaying efforts to improve traceability.
In Europe, regulations are still being developed
The European Union already regulates several PFAS, including PFOS and PFOA, through various pieces of legislation. At the same time, five countries submitted a proposal in 2023 for a much broader restriction under REACH. The European Chemicals Agency is continuing its scientific assessment of uses, exemptions, and socioeconomic impacts. As of September 22, 2026, this general proposal does not constitute a uniform ban that has already entered into force.[7]
The debate focuses in particular on the definition of the scope, essential uses, the availability of alternatives, and the duration of the transitions. The semiconductor sector is among those where technical functions and qualification timelines are being carefully examined. The outcome may involve a combination of bans, restrictions, monitoring requirements, and temporary exemptions, rather than a single rule applicable to all uses.
For digital companies, waiting for the final text would be risky. Procurement contracts, website design, and equipment all involve long lead times. Identifying these substances now makes it possible to prepare for compliance, limit pollution liabilities, and avoid having to replace a material at the last minute because it has become difficult to obtain.
What Responsible Governance Should Measure
- An inventory of PFAS and other fluorinated substances by site, equipment, supplier, and life cycle stage.
- The quantities purchased, contained, lost, recovered, destroyed, and transferred to waste or effluent.
- The distinction between direct emissions from the data center and the upstream carbon footprint associated with the manufacturing of chips, cables, and thermal systems.
- The criteria for selecting alternatives, covering climate impact, persistence, toxicity, safety, and technical performance.
- Supplier clauses regarding composition, analytical methods, incidents, end-of-life, and substitution schedules.
- Verifiable public information, including scope, methodology, uncertainties, and annual trends.
This governance encompasses compliance, ethics, and industrial strategy. It protects workers and nearby residents, reduces the financial risks associated with remediation, and helps buyers compare solutions. It also prevents local energy savings—such as more efficient cooling—from masking a shift in environmental impact toward the chemical industry or the end-of-life phase.
What to Watch for Now
Three developments will be decisive. The first concerns the publication of data: volumes by use, measured emissions, and treatment methods. The second focuses on truly qualified alternatives, their industrial availability, and their comprehensive environmental profile. The third will be regulatory, involving the ongoing European assessment and decisions made in other jurisdictions.
We will also need to monitor how AI operators incorporate these substances into their environmental goals. Sustainability reports often detail energy, water, and carbon, but say much less about the chemistry of infrastructure. A credible accounting framework for responsible AI will need to link these dimensions rather than treat them separately.
ChemSec’s investigation therefore does not support the conclusion that AI alone is causing a new global wave of PFAS. It does, however, reveal a real tension: the growth of computing still relies on materials whose environmental persistence can far exceed the lifespan of the equipment itself. Making this dependence visible is the first step toward reducing it.
Learn more
To put this chemical footprint into the context of the environmental, industrial, and infrastructural changes brought about by AI, be sure to check out these analyses from the aivancity blog.
Sources
[1] The Guardian, September 14, 2026. PFAS firms ride AI demand as campaigners warn of a new wave of "forever chemicals." View source
[2] ChemSec, 2026. The world’s top 10 PFAS producers; most are expanding production. View source
[3] Semiconductor Industry Association, PFAS Consortium. Uses of PFAS in semiconductor manufacturing. Accessed September 22, 2026. View source
[4] U.S. Environmental Protection Agency. Semiconductor Industry: Fluorinated Processes and Compounds. Accessed September 22, 2026. View source
[5] Arkema, June 23, 2026. Arkema successfully launched its PVDF capacity expansion in the United States. View source
[6] Chemours, August 10, 2026. Chemours launches Opteon ZE and Opteon 515B refrigerants. View source
[7] European Chemicals Agency (ECHA). PFAS and the REACH restriction proposal. Accessed September 22, 2026. View source
[8] OECD, 2021. Reconciling Terminology of the Universe of Per- and Polyfluoroalkyl Substances. View source