Evaporative cooling does exactly what the name says. Water enters a tower, thermal energy leaves with the vapor, and the aquifer never gets it back. My D. Truong, chief product and technology officer at ZutaCore, built his case on that physics in DataCenterDynamics on July 27, 2026. The page carries a sponsored label. The numbers hold anyway. He cites the International Energy Agency forecasting global data center water withdrawals above 1,200 billion liters a year by 2030, the Environmental and Energy Study Institute pegging a single large facility at up to five million gallons a day against a town of 10,000 to 50,000 people, and LBNL's 2025 Update putting data centers near 12 percent of US electricity by 2030 against 4.7 percent in 2024. Each traces back to the primary source intact. LBNL's reference case is 11.8 percent, inside a 9.5 to 15.3 percent band. The 4.7 percent baseline is 192 TWh. Truong took the reference case and left the ceiling alone.
The argument is a ladder. Open loop evaporative draws facility water and boils it off, the rung Truong ties to permit disputes running from Arizona to the Great Lakes. Closed loop with dry coolers seals a water-glycol mix in a circuit and rejects heat to air. He credits that with largely solving facility consumption, then goes after it anyway. Water still runs over multi-billion-dollar compute, dragging leak risk, corrosion, and standing fluid-chemistry maintenance into the white space. The third rung is his own. Waterless two-phase direct-to-chip, a dielectric boiling in the cold plate, no water present anywhere in the heat rejection chain. Truong says accelerator roadmaps are climbing past 4,000 watts per chip. Racks that once drew tens of kilowatts are now specified in the hundreds. That is the load case he says favors phase change over pumped volume. The middle rung is what the industry has spent two years celebrating, from Microsoft's Fairwater closed-loop design forward. Truong is telling operators they stopped one rung short.
Truong faults dry coolers for losing efficiency in hot regions, where summer temperatures force oversized equipment or evaporative trim on the worst days. That critique lands on his own architecture. The waterless reference design ZutaCore and Munters announced together pairs HyperCool on the technical loop with Munters SyCool. The joint release describes SyCool as a refrigerant thermosiphon that pulls rack heat up risers to modular condensers on the roof and rejects it to ambient. A rooftop condenser has no wet bulb to work against either. At 110F in Phoenix, both architectures are pinned to the same dry-bulb ceiling and take the same capacity derate. The op-ed never quantifies the condenser area, the fan power, or the PUE penalty that buys the water savings. Ask a vendor for the summer design-day condenser count on a 50 MW hall and watch how fast the answer becomes a follow-up call.
Chemistry is the second thing the piece leaves alone. Truong names the working fluid only as a dielectric heat transfer fluid and moves on. Two-phase direct-to-chip ran on 3M Novec until 3M announced its PFAS exit in December 2022 and set March 31, 2025 as the last day to place an order. ZutaCore moved to Chemours Opteon SF33 as a bridge and committed to a PFAS-free formulation in 2026. That history is why the thermally strongest cooling approach on the market nearly stalled out when the working fluid underneath it disappeared. An operator signing a 20-year campus on a sealed dielectric loop is underwriting a molecule. The op-ed does not name it, price it, or say what a recharge costs when the formulation turns over again.
Deleting the onsite loop removes the cooling tower from the site permit. It touches nothing at the generating station feeding the load. Roughly 72 percent of data center water consumption already happens offsite, upstream at power generation. A 300 MW waterless campus still carries that draw on somebody else's withdrawal permit. Truong's own framing makes this the weak point. He argues AI growth gets negotiated community by community, and community groups have gotten good at reading the utility filing next to the campus filing.
Truong closes with a sales line: "The technology exists. The industry simply has to choose it." Choosing it means a rooftop condenser field sized for the hottest hours of the year, a dielectric inventory with an unsettled supply chain, and a CDU population that has to hold containment for two decades. In Phoenix, Abilene, or anywhere a withdrawal permit is the thing holding up the shovel, those are tradeoffs worth taking. In Trondheim or Quebec, where the ambient air is cold and nobody is fighting over the well, the operator pays the two-phase premium and books zero permit value against it. Price the waterless premium against the permit it clears. That is the only column where it pays.