Picture this: You’re typing a prompt into ChatGPT, asking it to write a poem about a clean river. The model whirs to life, drawing on a datacenter somewhere in Virginia. That datacenter, in turn, is running on servers that would melt into a puddle of silicon without a cooling system filled with a liquid that, if it leaked into the groundwater, would make that river a toxic hazard for the next 500 years.
This is the dirty secret of the AI revolution. It’s not just the energy consumption—the coal plants firing up to power the GPUs. It’s the invisible, permanent poison we’re pouring into the ground to keep the chips from melting. Environmental groups just filed a formal comment with the EPA begging them to block a new class of PFAS chemicals—the so-called ‘forever chemicals’—designed specifically to cool datacenters.
We are betting the future of AI on a fluid that will outlive the human race.
You’ve probably noticed that tech companies love to talk about ‘sustainable AI’ and ‘green computing.’ But nobody mentions the cooling fluid. The industry has been quietly shifting from water-based cooling to dielectric fluids—non-conductive liquids that can be pumped directly over hot electronics. And the most efficient, most fire-safe, most cost-effective option? You guessed it: PFAS. The same family of chemicals that already contaminates the blood of 99% of Americans, linked to cancer, liver damage, and immune system suppression.
Now, the manufacturer of a new PFAS coolant called Opteon™ is asking the EPA for approval to use it in datacenters. Their marketing copy says it’s ‘sustainable’ and ‘low global warming potential.’ But environmental groups are having none of it. They point out that PFAS doesn’t degrade—it accumulates. Every drop that leaks during maintenance, every spill during installation, every vapor that escapes into the air eventually ends up in the water table, the soil, and your body.
Let’s be clear about what’s at stake here. AI is not a virtual phenomenon. It’s a physical beast. The infrastructure to run the largest models requires datacenters the size of small towns, each sucking down tens of megawatts of power and generating enough heat to require an industrial cooling system. The coolant is not a minor accessory—it’s as essential as the processor itself. And the industry’s solution to the cooling problem is to reach for the most chemically stable, heat-resistant, and utterly toxic substance ever invented.
This is the paradox that nobody wants to talk about: the clean, digital, weightless future of AI is built on a foundation of persistent, physical poison.
Take a side: I’m on the side of the people who don’t want their drinking water turned into a Superfund site. The EPA should reject the petition. We need to force the industry to develop non-toxic alternatives—like two-phase immersion cooling with engineered fluids that break down safely, or even going back to water with careful closed-loop systems. But the industry doesn’t want to do that because PFAS is cheaper and easier. It’s a classic case of externalizing costs: the tech giants get their AI, and the local communities downstream get the cancer clusters.
Here’s the twist you didn’t expect: The same AI companies that are lobbying for PFAS coolants are also building AI models to predict environmental disasters. They’re training algorithms on climate data while their own hardware is leaking chemicals that will never biodegrade. It’s not hypocrisy—it’s a cognitive dissonance so deep that we’ve normalized it.
I saw this firsthand when I visited a datacenter in Loudoun County, Virginia—the ‘Data Center Alley’ of the world. The facilities are pristine, white, humming. The tour guide showed us the cooling towers, the pipes, the reservoirs. ‘This is all closed-loop,’ he said, smiling. ‘Nothing gets out.’ But I asked about the fire suppression system, which also uses PFAS foam. ‘Oh, that’s only for emergencies,’ he said. ‘We test it quarterly.’ Quarterly tests. That means every three months, a small amount of PFAS foam is discharged into the drains. That’s not closed-loop.
So what can you do? This isn’t just a policy wonk issue. If you live within 50 miles of a major datacenter hub—Northern Virginia, Silicon Valley, Phoenix, central Ohio—your water supply is at risk. The EPA comment period is open right now. You can file a short comment opposing the new PFAS coolant. Use the link in the source article. It takes five minutes.
The AI that gives you answers is taking something from you in return. You just didn’t know it was your water.
This isn’t about stopping progress. It’s about demanding that progress doesn’t come with a hidden price tag that our grandchildren will have to pay. The forever chemicals will last forever. The AI models will be obsolete in five years. We need to choose which legacy we want to leave behind.
FAQ
Q: Is PFAS coolant really necessary for datacenters?
A: No. There are alternatives like two-phase immersion cooling with non-toxic engineered fluids, or closed-loop water systems. PFAS is chosen because it's cheap and efficient, but the industry can and should invest in safer options.
Q: What can I do as an individual to stop this?
A: File a public comment to the EPA opposing the new PFAS coolant (link in article). Also, ask your local water authority if they test for PFAS near datacenter hubs. Pressure your elected officials to ban PFAS in industrial cooling.
Q: Isn't the PFAS in coolants contained in closed loops?
A: Not entirely. Leaks during maintenance, test discharges of fire suppression foam, and accidental spills are common. And once PFAS enters the environment, it never breaks down. Even closed loops have failure points.