AI data centers consume city-level power with zero transparency

WIRED’s Uncanny Valley podcast explored how AI data centers operate and their mounting environmental and economic concerns, featuring discussions on energy consumption, political battles, and whether the current AI infrastructure boom represents a sustainable investment strategy. The conversation revealed significant gaps in transparency around energy usage while highlighting growing local opposition to data center expansion despite continued federal support.

How AI queries actually work: When users submit requests to ChatGPT or similar services, the process involves multiple computational steps that require significant infrastructure.
• Text queries are broken down into “tokens” (small text chunks) that AI models can process, then sent to specialized hardware called GPUs (graphics processing units) in data centers.
• These GPUs perform parallel processing to handle multiple calculations simultaneously, with popular models like Nvidia’s H100 chips powering the inference process.
• The AI model predicts what words should come next, building complete responses token by token before sending results back to users—all within seconds.

Energy consumption concerns: Large data centers consume massive amounts of electricity, with some facilities approaching the power usage of major cities.
• Meta’s planned Hyperion data center in Louisiana will consume about five gigawatts of power—roughly half of New York City’s peak power load.
• In Ireland, data centers already consume more than 20% of the country’s total electricity.
• Energy usage varies based on grid sources, with facilities connected to renewable energy having lower environmental footprints than those relying on fossil fuels.

Transparency problems: Companies provide limited information about actual energy consumption, making it difficult to assess true environmental impact.
• Climate researcher Sasha Luccioni, who works at Hugging Face, criticized Sam Altman’s claim that average ChatGPT queries use energy equivalent to running an oven for “a little over one second,” calling the figure inadequate.
• “It blows my mind that you can buy a car and know how many miles per gallon it consumes, yet we use all these AI tools every day and we have absolutely no efficiency metrics, emissions factors, nothing,” Luccioni said.
• Most energy data comes from voluntary company reporting rather than independent verification.

Political dynamics: The AI infrastructure expansion faces different reception at federal versus local levels.
• The Trump administration supports data center expansion powered by fossil fuels, oil, gas, and nuclear energy.
• Local communities increasingly oppose new facilities due to concerns about electricity rate increases, water usage, noise, and air pollution.
• Elon Musk’s xAI facility in Memphis sparked controversy after installing unpermitted gas turbines in a predominantly Black community already facing air quality issues.

Economic sustainability questions: Despite massive investments, consumer demand for AI services hasn’t matched the scale of infrastructure spending.
• Hyperscaler companies are using accounting methods that may understate infrastructure costs while inflating reported profits from AI services.
• Much current AI revenue comes from enterprise developers rather than general consumers, raising questions about long-term demand sustainability.
• The situation mirrors historical technology bubbles where infrastructure investment preceded actual market demand.

What experts recommend: Industry observers suggest citizens become more informed about local utility operations and energy policy.
• Understanding how local electric utilities work can help residents prepare for potential rate impacts from nearby data center construction.
• Supporting renewable energy initiatives and monitoring utility rate changes provides ways to influence local energy policy.
• Learning about these systems helps people engage more effectively with decisions affecting their communities.