What Does One Dashboard Actually Cost the Planet? I Did the Math.

How much does a dashboard cost the planet?

Something new is happening in American town halls. In the first half of 2026, roughly 75 major data-center projects worth more than $130 billion were delayed or canceled, in part because of organized local opposition, and by summer New York had signed the country's first state-level moratorium on new hyperscale facilities. More than 300 cities and counties have passed bans or pauses. Polls show a majority of Americans, across both parties, don't want one near them. The objection that comes up first, again and again, is water.

I lead analytics teams for a living. And watching communities fight the physical footprint of the cloud, I realized I'd never once asked what my own corner of it costs. So I sat down and did the math on a single dashboard. Here's the honest version, assumptions and all.

First, the units, from the public research:

A single AI text query uses roughly 2.9 watt-hours of electricity, about ten times a traditional web search's 0.3 watt-hours. On water, estimates vary a lot and this is genuinely contested: comprehensive accounting from UC Riverside puts a short AI exchange around 10 to 50 milliliters, while industry figures for direct cooling alone are far lower, near a third of a milliliter. I'll use the higher, whole-system range and tell you exactly where it comes from, because the honest number includes the water that generated the electricity, not just what evaporated off a cooling tower.

Now the dashboard. Not a plain one. The kind everyone is building right now, with an "AI insights" panel that writes a little natural-language summary every time the page loads or refreshes. Say it's set to refresh every fifteen minutes, around the clock, which is a common "near real-time" default nobody really questions. That's 96 refreshes a day, about 35,000 a year, each one firing off an AI generation.

Run the arithmetic. That single dashboard's AI layer burns roughly 100 kilowatt-hours a year, on the order of what a home refrigerator uses in a couple of months, and somewhere between a few hundred and a couple thousand liters of water a year depending on whose figure you trust. For one dashboard.

Here's the part I want to sit with, because it's the honest finding: that's a small number. One dashboard is not an ecological crisis. And that is exactly the trap.

Because it's small, and invisible, and lives on a bill someone else pays, nobody governs it. So we set every dashboard to refresh in real time when daily would do. We bolt a frontier-sized AI model onto a summary a right-sized small model could write for a fraction of the energy. We leave up the dashboards nobody has opened in eight months, quietly refreshing every fifteen minutes forever. And then we do that five thousand times across an enterprise. Multiply my one small dashboard by five thousand and you're at roughly half a million kilowatt-hours and millions of liters a year. That's not a rounding error anymore. That's a meaningful slice of a real building, in a real town, with a real fight at the zoning board.

None of this required the biggest model or the fastest refresh. That's the whole point. Most analytics workloads are wildly over-provisioned, running huge models and real-time refreshes for questions that didn't need either. We're spending the planet's resources on precision and freshness people forgot they asked for and no one is reading.

Which makes this, quietly, a data-governance problem. And governance is my job.

The levers are unglamorous and completely within our control. Right-size the refresh: does this update daily, or does it truly need to be live? Right-size the model: a small, local, cheaper model handles most summarization at a fraction of the footprint, which is the same lesson the efficiency research keeps teaching. And run the audit nobody runs: how many of our dashboards has anyone actually opened this quarter? The greenest dashboard is the one refreshing daily instead of every sixty seconds. Better yet, it's the one you delete because nobody was looking at it.

I keep coming back to the town halls. The cloud used to feel like it lived nowhere. It lives somewhere now, and the somewhere is a specific place with specific neighbors and a specific water table. I sit on a public board in my own community. I know what it looks like when residents fill a room because a project is going to draw on the same water they drink. Leading well in this era has to include leading responsibly toward the grid and the people on the other end of it. Not as a press release. As a design decision, made quietly, in the defaults we choose.

I'm not anti-dashboard. I build them. I'm anti-waste, and most of what we're spending here is waste we could simply choose to stop. Efficiency isn't a constraint on good analytics. It IS good analytics. It always was. AI just made the cost of forgetting that a lot more visible.

One question for the data and analytics people here: when's the last time you audited how often your dashboards refresh, and how many nobody opens? I'd bet there's a small climate win and a budget win sitting in the same overlooked place.