Forever Chemicals Ride With the AI Buildout
Everyone in AI spent Monday arguing about whether to slow the frontier. I got stuck on the physical side of that argument, the bit about what chips and data centres are actually made with and cooled in. ChemSec now pulls together producer announcements and investor material and finds expansion of PFAS capacity across the major makers with very few exceptions, driven in its telling by AI and data centre infrastructure including immersion cooling fluids and thermal management alongside semiconductor manufacturing uses like surfactants, etching, cleaning and tubes and fittings, plus lithium ion battery materials.1 That is not a footnote to the AI boom, that is the plumbing for it, and it makes the slowdown chat look strangely airy, all models and morals with no chemistry attached.
The politics bites hardest in fabs. Ars Technica reports that Intel and other semiconductor firms have teamed up with industrial materials businesses to fight American clampdowns on PFAS, substances described as slow to break down in the environment, a lobbying push that broadens opposition to new rules and bans.2 I read that as necessity talk rather than denial, and the paperwork points the same way, with the trade body SEMI maintaining a dedicated explainer on types of PFAS regulation for semiconductor manufacturing.3 You do not build that kind of compliance guide if your line is that there is no problem. You build it if your line is that the rules have to fit a process you say you cannot run without this chemistry. That is clever, because it lifts a pollution fight into industrial policy and national security, and anyone who wants tighter controls is suddenly cast as the bloke trying to kneecap AI itself.
The trouble is I have watched this lot use indispensability and safety as commercial cover before, and the current mood music gives me pause in both directions. Of the frontier slowdown calls, Ars notes “Safety is the watchword, but there could be ulterior benefits for the industry.”4 One developer thread punter put the cynical version in budget terms, suggesting part of the push to slow development is about adding regulatory pressure that gives the labs cover to train fewer models and slow their cash burn.5 I am not saying the PFAS lobby and the pause letter crowd are the same people. I am saying both plays work the same way, by turning a cost problem into a virtue problem. When chip supply is declared strategic, any environmental rule looks like an attack on progress. When training runs get ruinously dear, a principled pause starts to look like a budget cut with better PR.
My take is that the hinge of this fight really is irreplaceability, and that is where scrutiny belongs. If these chemicals are genuinely baked into etch, cleans, seals and coolants with no qualified substitute at volume, then regulators face a real choice about derogations, timelines and closed loop handling. If workable alternatives already exist in some steps, then a blanket claim of no substitute starts to look like bargaining for the widest exemption. What I have in front of me does not settle which steps are which, and I will not pretend otherwise. What it does establish is the shape of the lobbying: acknowledge persistence, emphasise function, and force governments to choose between clean chemistry and chip capacity. That is a harder contest for green groups than arguing with denialists, and a more convenient one for chipmakers than defending pollution. Watch what gets exempted, for how long, and on what proof of no alternative. That paperwork will tell you whether irreplaceable means physically true or just commercially annoying.
Sources
- 1 (PDF) Expansion plans of the world’s top PFAS producers - ChemSec — chemsec.org
- 2 Chipmakers fight spread of US crackdowns on “forever chemicals” - Ars Technica — arstechnica.com
- 3 PFAS Explainer: Regulations for Semiconductor Manufacturing | SEMI — semi.org
- 4 AI leaders want to hit the brakes after years of reckless speed — arstechnica.com
- 5 @TSiege on Anthropic tells investors it will be profitable for second straight quarter — ycombinator.com
How this was made
- 01-research ~z-ai/glm-flash-latest $0.021
- 04-nominate ~deepseek/deepseek-v4-flash-latest $0.001
- 05-select ~google/gemini-flash-latest $0.006
- 06-write meta/muse-spark-1.3-contributor $0.004
- 08-visualise anthropic/claude-sonnet-5 $0.042
total $0.073