Inspiration

In January 2025 Richmond’s water plant failed and Mechanicsville lost water for up to a day and a half. About 80% of Hanover County’s water supply comes through that one Richmond contract. A mile and a half from here, the Iron Horse data center is on record at up to 1.2 million gallons a day. Counties approve data centers using average-day or yearly water numbers, but water systems run out on the single worst day.

What it does

Headroom takes a data center’s size and cooling design and calculates its water use for every hour of last year, using real weather measured at Hanover’s airport. It finds the campus’s worst day and checks whether the county’s water system can cover it, including if the Richmond plant goes down again. Then it finds a better way to run the cooling and writes the permit rule a county can adopt.

Challenges we ran into

Iron Horse’s power use isn’t public, so we worked backwards from its 1.2 million gallons a day to get about 100 MW. Our first optimizer claimed a 78% cut, but it secretly raised peak power on the hottest afternoon, so we added a power limit and report the honest 42%. A weather service blocked us mid-build, so we switched to measured airport data stored in the app.

Accomplishments that we’re proud of

On its worst day, the “water-efficient” hybrid still uses 0.89 million gallons, 7.8 times a normal day. Headroom brings that day down to 0.52 million gallons: 42% less, or 370,000 gallons the county keeps on its hardest day. The data center uses the same total water for the year, its peak power doesn’t go up, and it uses a little less electricity. It also writes the rule a county can adopt: no more than 520,000 gallons on any day.

What we learned

Averages hide the day that matters. A “closed loop” system can still evaporate water. And water and power peak on the same hot afternoons, so a fix has to protect both.

What’s next for Headroom

Share it with Hanover planning staff and Virginia’s environmental agency (DEQ), add pipe-level detail for each part of the county, and test it against real utility meter data.

Built With

Share this project:

Updates

Submission history