Scalding hot AI accelerators have put datacenters in hot water. Two-phase cooling could chill them out

The investable angle is not the coolant chemistry itself; it’s who gets to control the service layer when AI racks become thermally constrained. If two-phase systems reduce chiller dependency and raise permissible inlet temperatures, the marginal value shifts toward OEMs and integrators that can certify, warranty, and maintain the stack — that favors DELL first, with HPE and LNVGY trailing as more optionality than certainty. The bottleneck is qualification, not lab performance, so near-term revenue impact is likely small even if the technical case is real.

For NVDA, the effect is second-order but meaningful over 6-18 months: lower cooling overhead improves data-center TCO, which supports faster GPU deployment and better utilization of high-TDP racks. That said, the adoption ceiling is set by CDU capacity and field serviceability; if rack power ramps faster than the thermal ecosystem can scale, this becomes a deployment throttle rather than an accelerant. In other words, better cooling can expand addressable demand, but only if OEMs standardize it and operators can service it without voiding warranties.

MMM is the cleanest loser on a structural basis because the PFAS overhang continues to close off legacy coolant pathways and reduces optionality in adjacent specialty-chemicals categories. But I would not overstate the equity impact: this is more about lost participation in an enabling technology than a material earnings hit. The bigger competitive spillover is to incumbents in datacenter thermal infrastructure — if two-phase architectures win, pricing power migrates away from proprietary fluids and toward CDUs, manifolds, and OEM-qualified cold-plate ecosystems.

The contrarian point is that the market may be too quick to extrapolate a broad liquid-cooling boom from one vendor claim. Until we see repeatable OEM blessing and standardized service contracts, this remains a niche procurement decision, not a platform shift. The thesis is falsified if hyperscalers continue to choose single-phase systems at scale or if higher-density racks force a move to full immersion instead of this intermediate two-phase path.