Why a coolant with no fluorine is harder than it sounds

What the question is

Two-phase cooling for AI servers has run for years on fluorinated liquids. The fluid boils on a hot chip, the vapour rises, and a cooler turns it back into liquid. It works well because these liquids boil at a low temperature and do not burn. That last property matters when the fluid sits next to live electronics.

The trouble is that fluorinated fluids sit inside a much larger family of chemicals called PFAS. The US Environmental Protection Agency describes PFAS as long-lasting chemicals that break down very slowly. One maker of these fluids announced that it would leave PFAS manufacturing. The home page notes that Europe is moving to restrict the whole class.

So the obvious wish is a coolant with no fluorine at all. This post explains why that wish is harder to grant than it sounds, using the screen that FluxZero runs.

What we counted

We built a catalogue of every two-part blend we could make from ordinary catalogue molecules, one low-boiling fuel paired with one pump liquid. The screen holds 14,559 blends, and its stage-by-stage counts are published; the per-blend table is not. We scored each one on heat removal, electrical behaviour, safety, climate impact and boiling range.

The climate test alone removes more than half of them. A total of 7,020 blends clear the ceiling we set for global warming potential. That leaves a large space, and it is fair to ask whether a safe, low-carbon choice with no fluorine sits somewhere inside it.

• The screen is the full list, not a hand-picked sample.
• It is computed from literature values, so each score is an estimate.
• Its stage-by-stage counts are a published file, linked at the end of this post; the per-blend table is not published.

The count that matters

The next question is how many of those low-carbon blends are rated non-flammable. The screen finds 174 of them. Then we recounted their fuels from the lab’s blend table. Every one of them has a fluorinated fuel. Half use one fuel from the family sold as hydrofluoro-olefins. Half use a fluid sold under the name Novec, which is itself a PFAS. Theorems & Bounds states this recount beside the counts above.

Read that carefully. It does not say a safe fluid with no fluorine cannot exist. It says that, inside the space we enumerated, none appeared.

Why the easy answer fails

Fuels with no fluorine are easy to find. Pentane is one. Methanol is another. They boil at convenient temperatures. The two lead blends on our home page use fuels like these. A count of the six components this site names, the five in those two blends (pentane, dimethyl carbonate, methanol, triethylene glycol and boron nitride) and gamma-butyrolactone, another pump the site names, finds 0 fluorine atoms.

The difficulty is that these fuels burn. Pentane is flammable, so a pentane blend needs a closed tank under nitrogen. Methanol is flammable too. Our methanol lead has not had its flash point measured, and the vapour it gives off while boiling would be mostly methanol. So it is not shown to be safe from fire.

Here is the trade, stated simply:

1. A fuel with no fluorine tends to burn.
2. A fuel rated non-flammable tends to carry fluorine.
3. Adding a second liquid can shift the balance, but the shift has to be measured, not assumed.

Why the pump liquid does not rescue it

In our design the second liquid is meant to act as a pump. Its surface tension is higher than the fuel’s, and in theory the difference pulls boiling liquid back to the hottest spot, which would help heat transfer. That effect has not yet been simulated with surface-tension-driven flow or measured for these blends: our model assumes it, it does not show it. Even if it works, it does not make a flammable fuel safe, and nobody has yet shown that it changes the fire behaviour of the mixture.

What it means for the PFAS question

There are two ways to read the finding, and both are fair.

A cautious reading

If you need a coolant rated non-flammable with a low climate impact, then today’s candidates in our space all rely on fluorine. Within our screen, a buyer who wants to avoid PFAS completely would have to accept a fluid the screen does not rate non-flammable, plan for it to burn until a flash point is measured, and design the tank, the nitrogen blanket and the fire plan around that.

A hopeful reading

The catalogue is limited to molecules we could look up. It leaves out new molecules, and it leaves out mixtures of three or more parts. The finding points to where a search should go next: a fuel rated non-flammable that has no fluorine.

A caution about the word

Whether a given fluorinated molecule counts as a PFAS depends on the definition a regulator uses. We do not settle that here. We only report which molecules our screen kept.