Allgood pegs 100MW AI data center at 223M liters of water

There is a number in this week's episode of This Week in Green Software that is worth sitting with. AI ethicist Masheika Allgood, founder of the Sunnyvale-based consultancy AllAI, says a single 100 megawatt AI data center needs roughly 223 million liters of water just to fill and prime its cooling system, then draws about 3 million liters a day to keep running, losing about 1.5% of that daily to drift, evaporation and blowdown.

The reason those numbers exist at all is that nobody was publishing them. Allgood, a former lawyer who worked on AI training software at NVIDIA before starting AllAI, built her own calculator on top of the Uptime Institute's makeup water methodology, an industry-standard sustainability approach that measures operational autonomy if the grid or water supply drops. She is upfront that the model uses reasonable industry assumptions that can be debated, but as she put it on the podcast, "we're not getting the numbers of how, what's the efficiency of the cooling loop."

For context on scale, her own Santa Clara backyard is home to 55 data centers, the second-largest concentration in the US after Loudoun, Virginia, with California capping any single facility at 99 megawatts and another 15 to 20 sites clustered in San Jose. She notes the Bay Area now ranks seventh nationally for fine particle pollution, driven partly by monthly generator maintenance cycles.

The honest caveat is that Allgood's figures are derived from published system autonomy data and vendor assumptions, not from meter readings the hyperscalers themselves have released; nothing in the episode reconciles her numbers against Google's or Meta's own sustainability disclosures. The Georgia case study she references is presented as a community-organizing template rather than an audited before-and-after.

What actually changes if a calculator like this gets adopted is the permitting conversation. Municipal counsel and planning boards get a defensible starting number to negotiate against, and the Wyoming precedent, where regulators tightened wastewater rules after a Meta data center contractor discharged contaminated water, shows that once local governments have a number they trust, they will use it.