Greg Brockman describes closed-loop cooling systems in AI data centers that keep ongoing water consumption low after initial fill and comparable to a medium office building or several households

Greg Brockman explains how the public story about AI data center water use is partly wrong. Because the cooling-systems use a closed-loop design that circulates the same stored water instead of constantly pulling fresh water. i.e. it works less like a running tap and more like a sealed pool, where water absorbs heat from servers, moves through cooling equipment, then returns to the same circuit. The argument here is not that AI infrastructure has no resource cost, but that public debate often mixes up different cooling designs and treats every data center as if it burns through water the same way. The important distinction is water withdrawal versus water consumption, because a site can hold a large amount of water inside its pipes while using far less new water day to day. OpenAI's official blog on Stargate project also says the same thing: "Water is one area where details matter. Like many data centers, the Abilene site uses closed-loop cooling rather than traditional evaporative cooling towers. Once the system is filled, water continuously moves through sealed pipes and is recirculated rather than consumed. For Abilene, the one-time initial fill for each building is equal to roughly two Olympic-sized swimming pools. After that, annual water use for the entire cooling system at full buildout is expected to be comparable to a medium-sized office building, or about four average households." --- From 'The Knowledge Project Podcast' YT channel (link in comment)