Insights from Attempted Intelligence, Episode 2, hosted by Matt Duren, SVP of AI and data, KnowBe4 and Harlan Parrott, VP of AI innovation, KnowBe4.
AI data centers are one of the most debated topics in the world right now. Some people worry about the energy and water it takes to run them. Others see the foundation for the next wave of AI advances. Both camps have a point.
On episode two of the Attempted Intelligence podcast, Matt and Harlan each picked a side and debated it. Their goal wasn’t to tell listeners what to think. It was to help them weigh the facts themselves. They framed the discussion around global versus local effects and talked through four topics: electricity, water, noise, and community economics.
Electricity: Small Globally, Big Locally
Harlan made the global case. Data centers, AI and non-AI combined, used about 1.5% of global electricity in 2024. Even if that doubled, it would reach roughly 3%, which is small next to air conditioning, industry and electric vehicles.
Matt pushed back with the local picture. Data center demand grew 17% globally in 2025, against 3% overall electricity growth. In Ireland, data centers took 23% of metered electricity in 2025, up from 5% in 2015. He also cited Lawrence Berkeley National Laboratory’s updated June 2026 estimate that data centers could reach as much as 15% of U.S. electricity. Cheaper inference also tends to drive more usage, not less.
Harlan agreed that local impacts matter, pointing to rate increases of roughly 28% in some Virginia areas (though inflation played a part). His answer: careful planning, requirements that data centers own their own power supply, and support from construction unions can offset the downsides.
Water: Myth, Reality and the Problem with Averages
Harlan noted that public concern was fueled by a since-debunked claim that each ChatGPT query uses gallons of water. U.S. data centers used about 17.4 billion gallons in 2023, out of roughly 228 billion gallons of water in total. That’s far less than golf courses or irrigated agriculture, and indirect water use from power generation is much larger than direct cooling. His challenge: if you object to data centers but resent lawn-watering limits, reconsider.
Matt countered that the supply is often drinking water, and evaporative cooling doesn’t return water to its source. Places like drought-stricken Tampa and parts of Texas, where neighbors’ wells have run dry, face real strain. As he put it, you can’t replenish a watershed with a national average.
They found common ground on solutions: closed-loop and zero-water designs (such as Microsoft’s planned 2027 facilities) and published water studies before any new build. Matt also pointed to KnowBe4’s own sustainability commitments , including net zero emissions by 2040.
Noise and Community Economics
Matt described the continuous, low-frequency tonal hum of data centers. At a Vantage site in Sterling, Loudoun County, Virginia, residents complained even though the operator met a 55-decibel limit.
Harlan noted that society has long handled loud neighbors like lumber mills and airports through zoning and mitigation. Data centers with on-site generation, he argued, should be zoned as industrial, not office space.
He also highlighted the upside for communities. Loudoun County’s FY2026 data center property tax revenue was $1.2 billion, or 39% of its budget. Since 2016, the residential tax rate has dropped from $1.14 to $0.80 per $100. Add the multiplier effect of construction spending in local economies, and the benefits are real.
The Takeaway
The broader debate around AI data centers is looking at it from a global perspective and what it means for humanity overall. The right answer depends heavily on local circumstances, from a desert site on well water to one with its own power next to abundant fresh water.
Listeners are urged not to fall exclusively on one side, to weigh both the costs and the opportunities, and to advocate for responsible decisions in their own communities, states, and country.
Listen to the latest Attempted Intelligence episode here, on Spotify, Apple Podcasts and everywhere major podcasts are streamed.
