Georgia AI Data Center Drained 29M Gallons of Unmetered Water

A Blackstone-owned data center campus in Fayette County, Georgia, secretly consumed 29 million gallons of water over 15 months through two unmetered connections the county didn't know existed. The 6.2 million-square-foot QTS Fayetteville ("Project Excalibur") site, among the largest data center developments in the US, was discovered after nearby residents reported low water pressure.
Key Details
- Volume: 29 million gallons drawn via two unauthorized water connections.
- Duration: 15 months before detection.
- Discovery: Residents complained about low water pressure; county investigation traced it to the data center.
- Penalty: County waived fines, charged only $147,474 in retroactive water charges.
- Scale: 615-acre campus with 13 buildings, total 6.2 million sq ft, with plans for expansion.
- Context: While the data center drew water without permission, the county was simultaneously asking residents to stop watering lawns to conserve water.
This incident highlights growing tensions around AI data center resource consumption. As AI compute demands surge, data centers increasingly compete with local communities for water and power. Unlike residential water use, data centers typically use water for cooling — often thousands of gallons per megawatt-hour of IT load. The lack of oversight and lenient enforcement in this case may set a precedent for other AI infrastructure projects.
📖 Read the full source: HN AI Agents
👀 See Also

AI Interview Platforms Tested: CodeSignal, Humanly, Eightfold in Job Screening
The Verge tested three AI interview platforms including CodeSignal, Humanly, and Eightfold for job screening. The AI avatars conduct one-on-one video interviews, analyze responses, and claim to reduce bias, though bias-free systems remain impossible due to training data limitations.

Microsoft Ends Revenue Sharing with OpenAI, Impact on AI Agents Unclear
Microsoft will stop sharing revenue with OpenAI, its main AI partner, according to a Bloomberg report. The move could affect how developers integrate AI agents via Azure OpenAI services.

Ångstrom Used Claude Code to Train a Model That Beat Meta's UMA-OMC — 100k GPU Jobs on Spot
Ångstrom (YC S24) trained CSP-MACE-Å, an ML model 10,000x faster than DFT with matching accuracy, outperforming Meta's UMA-OMC on crystal structure prediction. They used Claude Code to orchestrate 100,000 GPU jobs on multi-cloud spot via Anycloud CLI.

Illinois Passes SB 315: Third-Party Audits Required for Frontier AI Labs
Illinois passes SB 315 requiring frontier AI labs like OpenAI, Anthropic, and Google DeepMind to have safety practices audited by independent third parties. If signed, it becomes the strongest US state AI safety law.