Urgent.News

What's breaking now, across thousands of outlets.

Tech

American data centers in just seven states are reportedly using more water than all the cities in California

Data centers in seven states are being linked to the consumption of 10.4 million acre-feet of water annually.

American data centers in just seven states are reportedly using more water than all the cities in California

A new report by think tank Ceres reveals that data centers in just seven states consume more water than all the cities in California. The seven states are Virginia, Texas, California, Illinois, Georgia, Ohio, and Arizona. According to Ceres, these data centers withdraw 3.4 trillion gallons of freshwater annually, or 10.4 million acre-feet. This figure may be overstated as it includes water returned to rivers immediately by hydropower plants.

Eighty-eight percent of electricity in these states comes from water-using power plants, with 66% of these plants located in water-stressed areas. Ceres's findings paint a partial picture as it measures water withdrawals, or the total volume of water pulled from sources like rivers, lakes, or aquifers, including water returned shortly afterward.

The California Legislative Analyst's Office report states a different figure of 8 million acre-feet, representing water actually delivered and applied to various uses in California.

The majority of power generated for data centers is hydroelectric, which may seem ironic given the concern for water usage. However, this approach provides a strong indicator of risk exposure to communities already facing drought. The high number of withdrawals may be higher than expected; Arizona State University researchers estimate overall US data center consumption, including power generation, at 650,000 acre-feet annually. If this consumption rises significantly, it could be 16 times the reported figure.

Despite these concerns, projections suggest that water for power generation could account for 72 percent of total US data center water consumption by 2030, even as cooling efficiency improves. Current projects have faced opposition due to water and grid concerns, leading to $130 billion in projects disrupted within the first quarter of 2026.

Ceres emphasizes the need for a more inclusive approach, as no single stakeholder can address these challenges alone. The future of AI companies investing in data centers and their cooling and power requirements remains uncertain.

Written by urgent.news from TechRadar's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at techradar.com →

More in Tech

How to Send Email from Cloudflare Workers

There are two real ways to do this: Cloudflare's own native Email Service binding, or calling an external email API like Notify over fetch() .

  • Native Email Service requires Paid plan, domain registration, DNS records, and Wrangler config
  • External API method (e.g., Notify) is portable, avoids Node.js issues, needs API key as secret

The GNU strip backdoor is the case AI code review can't see

This paper (arxiv 2607.24888) is worth a stop if you evaluate AI code review. It shows Ken Thompson's trusting-trust attack is not compiler-specific: the authors pull it off with GNU strip, a build…

  • GNU strip backdoor can implant payload in ELF binaries
  • AI code review tools cannot detect binary-level malicious code
  • Strip-trust problem requires separate verification from source code review

I Want More Coding Agents to Work Like This

💻 One thing I dislike about coding-agent setups is how quickly they become part of one specific machine. Provider config goes in one place, session state somewhere else, local models live in another…

  • OpenClaude-Portable packages coding agent, runtime, and data in a single folder.
  • Supports nine model providers, including Anthropic Claude and OpenAI.
  • Offers cloud and local model options, with local models reducing latency significantly.

More from Tuesday 8 September →