Analyze infrastructure demands, water usage, noise levels, and civic sentiment trade-offs
A one-minute painted explainer of the simulator's model, followed by real-world figures on data-center electricity, water and noise.
A new AI data center wants to move to town. This simulator weighs what it asks of the community, power, water and quiet, against the taxes it pays. Power first. The default campus draws one hundred fifty megawatts, but the local utility has only one hundred twenty to spare. That's a hundred twenty-five percent load, so the grid would need an upgrade. Water next. The tool assumes evaporative cooling uses twelve thousand gallons per megawatt each day. That's one point eight million gallons daily, sixty percent of the aquifer limit. Direct-to-chip liquid cooling cuts it eightfold. Then noise. The model starts at eighty-five decibels near the cooling plant and loses six decibels each time distance doubles. At a five hundred foot setback, neighbors hear about fifty-seven decibels. At a thousand feet, fifty-one. With half its property tax abated, the campus still pays one point two five million dollars a year. The tool weighs every strain into an opposition score of sixty-one out of a hundred: moderate friction. Try a different cooling choice or a wider setback, and watch which pressure eases first.
The tool divides the campus's draw by the utility's spare capacity: 150 MW against 120 MW of headroom is 125%. Nationally, Lawrence Berkeley National Laboratory estimated data centers used about 4.4% of U.S. electricity in 2023 and could reach 6.7% to 12% by 2028, which is why local headroom has become a siting question.
The simulator uses 12,000 gallons per megawatt per day for evaporative cooling, 3,000 for closed-loop and 1,500 for direct-to-chip. 12,000 gallons a day for a megawatt running around the clock is roughly 1.9 litres per kilowatt-hour, in the range of commonly reported on-site water use. At 150 MW that is 1.8 million gallons a day, versus about 225,000 with direct-to-chip.
The noise model assumes 85 dB at 20 feet and treats the plant as a single point source, so sound falls 20·log10 of the distance ratio: 71 dB at 100 ft, 57 dB at 500 ft, 51 dB at 1,000 ft and 45 dB at 2,000 ft. Real sites have many fans and chillers, low-frequency hum, terrain and barriers, so measured levels can differ a lot. The WHO recommends keeping average road-traffic noise below about 53 dB, and night levels lower.
The tool starts at 20 and adds points for grid overload (+20 at default), water draw above half the limit (+4), noise above 50 dB (+10.5) and tax abatement above 30% (+6), giving 61 out of 100. The friction tags on the page show fixed labels like '+30 Opp' that don't match these actual contributions.
U.S. data-center electricity use in 2023 and projections to 2028, plus water-use estimates.
Global data-center electricity demand and how AI is expected to change it.
Health-based limits for long-term environmental noise exposure.
Figures in the video come from the simulator's built-in coefficients at its default Wisconsin Small Town preset; they illustrate the trade-offs rather than describe any real facility.