Big data takeover
Charlie Robinson welcomes Hakeem Anwar back to Macroaggressions to discuss his six-month investigation into AI data centres. Anwar has built an open-source map and report at aidatacentermap.org tracking roughly 4,500 data centres across the United States, including 2,000 operational facilities and 623 hyperscale (AI) data centres. Robinson frames the issue as one that crosses party lines: most people do not feel they need faster AI responses, yet big tech has pledged $2.4 trillion—equivalent to the US highway budget for 44 years or multiple national high-speed rail networks—to expand data centre infrastructure.

Water and Power Consumption
Anwar explains that hyperscale data centres consume millions of gallons of water daily. A single GPU costs $30,000–$50,000 and performs a trillion calculations per second, generating heat within one order of magnitude of the surface of the sun. Water is run through cooling tubes, evaporates in cooling towers, and is lost to the atmosphere. Anwar stresses that aquifers take hundreds of years to recharge, and when millions of gallons are extracted daily, recharge cannot keep pace. He has not found a single data centre with an active water recycling programme; claims of recycling are, he says, "lip service." The water also picks up heavy metals, requiring cleaning before any potential return to the water supply.
The power demands are equally stark. Anwar cites a Lawrence Berkeley National Labs paper estimating 4.45 litres of water per kilowatt used. Because data centres operate at gigawatt scale, total water use is enormous—plausibly exceeding all bottled drinking water consumed by people. The PJM interconnection (a federally regulated regional grid in the Northeast) fell 6.6 gigawatts short of its reliability margin, roughly equivalent to the power needed for 6 million homes. Electricity prices in PJM-served areas, including Virginia, have risen roughly 300%. In Nevada, 50,000 Lake Tahoe residents face energy diversion to data centres.
Regulatory Capture and Corruption
Anwar describes a pattern of regulatory evasion and preferential treatment:
- Code names: Developers use names like "Blue Goldfish" rather than corporate identities when seeking zoning approvals, skirting public scrutiny.
- Rezoning: Agricultural land carries stricter permitting; once rezoned, developers face fewer constraints.
- Air permits: At xAI’s Memphis data centre, 35 gas turbines were brought on site and operated before permits were obtained. Permits were requested individually to avoid fleet-wide environmental review. The Southern Environmental Law Center funded a thermal flyover, confirmed 35 turbines running, and is now suing xAI for Clean Air Act violations.
- Wholesale water pricing: Data centres receive preferential water rates while local residents bear the cost.
- Subsidies: 32 US states offer data centre subsidy programmes, but only 20 disclose the amounts. Texas lost over $1 billion in 2025 alone. Former Texas Governor Rick Perry is involved in Project Matador, one of the nation’s largest data centres, named after President Trump.
Anwar attributes the unanimous pro-data-centre votes by city councils to "pure corruption and conflict of interest."
Health and Environmental Impacts
Robinson raises electromagnetic fields (EMF) as a potential health angle; Anwar has not yet studied EMF but is looking into it. He highlights noise pollution from a Marathon Computing crypto-mining operation (300 MW, far smaller than AI hyperscalers). Nearby residents described living next to a 24/7 vacuum cleaner, experiencing stress, high blood pressure, hearing loss, and sleep deprivation. One person’s dog tore out its own fur; horses panicked; wildlife fled. Electricity bills doubled or tripled.
Geographic Patterns
Anwar’s map reveals clusters in water-stressed regions—Dallas, Phoenix, El Paso, the Texas Panhandle—using the World Resources Institute Aqueduct model. Data centres also cluster near major aquifers and within roughly one mile of high-voltage power lines, since extending transmission costs roughly $1 million per mile. He notes that only about 10% of data centres have disclosed power capacity, so actual impacts are likely far higher.
Moratoriums and Local Action
- 13 states have proposed data centre moratorium bills; none have passed at state level. Maine’s 18-month pause was vetoed by Governor Janet Mills.
- 69 local moratoriums exist across counties, cities, and towns; 50 are active.
- Anwar recommends connecting with local water conservation organisations, monitoring permit submissions, using Freedom of Information Act requests, and attending rezoning hearings. He is developing an "adopt a data centre" feature on his map to help communities organise.
Flock Cameras and Surveillance
The conversation shifts to Flock cameras (automated licence plate readers). Anwar sees them as part of a "surveillance control grid" feeding data into AI systems. The cameras capture:
- Licence plates
- Vehicle make and model
- Bumper stickers
Linked to DMV records, they enable real-time location tracking. He recommends deflock.me to locate cameras and notes Sedona residents succeeded in removing them from city limits, though county-level cameras remain.
Significant Points
- No verified water recycling: Despite industry claims, Anwar has confirmed zero operational recycling programmes.
- Heat intensity: GPU heat nears solar surface temperatures, making water cooling currently unavoidable.
- Political entanglement: A former governor directly profiting from a federally named data centre project.
- Permit fraud: xAI ran 20 unpermitted gas turbines for months before legal challenge.
- Scale distortion: Single data centres can exceed entire state power consumption (Utah example: 9 GW, double the state total).
- Noise as weapon: Crypto-mining noise drove animals to self-harm and caused human hearing loss.
- Democratic reversal: Anwar, an anarchist inspired by Derek Broze and Larken Rose, now advocates local government engagement - what he calls a 180-degree turn. It is necessary to push back locally in all locations.
- Android 2027 lockdown: Google will require developer ID and master keys for app distribution, ending side-loading as currently practised.