What is documented: data centers in California are required to file a Water Supply Assessment under CEQA before approval. But according to the reporting, projects can sidestep full environmental review by falling under specific size or water-use thresholds, or by qualifying for alternative municipal approval pathways. That's a real gap, and it's worth noting separately from whatever the headline number turns out to be.

The community response online was predictably scattered. Several commenters defaulted to the almond deflection — a tired move that treats agricultural water use as a reason to ignore industrial use, as if both can't be problems simultaneously. Others proposed a $50/liter tax on unrecycled water, which is a number someone typed without referencing existing water pricing structures or enforcement mechanisms.

For context: large language model training runs and inference workloads are genuinely water-intensive. Google disclosed in its 2023 environmental report that its global water consumption increased 20 percent year-over-year, attributing part of that to AI infrastructure. Microsoft reported similar trends. Whether California-specific data center water use is a crisis, a manageable externality, or something in between depends on figures that, by the framing of this story, aren't publicly available in a consolidated form.

That's the actual story: not that data centers are definitely destroying the water table, but that the disclosure framework has holes large enough that we genuinely don't know. That's a regulatory design question, and it's a slower, less satisfying story than the headline implies.