Tag: data centers

  • How Much Water Does a Data Center Actually Use?

    Somewhere right now there is a town meeting where a guy in a folding chair is standing up to say that a data center is going to drink his well dry, poison the aquifer, and leave his grandkids fighting over a garden hose. I am not making fun of the guy. (I have been the guy in the folding chair, about other things.) But before your town votes on one of these buildings, and towns all over this state are voting on them right now, it is worth knowing what the water numbers actually are. Because the real answer is smaller than the panic, bigger than the brochure, and weird in a way nobody at the microphone is saying.

    The question is not fringe. “Data center water usage” is searched about 9,900 times a month in this country, and it is up more than 600 percent in a year. People are not idly curious. A building the size of a small city is going up near their house and they want to know what it drinks.

    The bottle of water claim, taken apart

    The version flying around says every AI prompt you type costs about a bottle of water. That exact question, is it true that one AI prompt uses a bottle of water, gets searched roughly 8,100 times a month. Eight thousand people a month typing a rumor into Google to find out if it is real. That tells you something about where we are.

    Here is the honest version. Cooling a building full of computers takes water, and some of that water evaporates, and evaporated water is gone the way your money at the track is gone. But the per-prompt math behind the bottle claim takes the thirstiest kind of facility, in the hottest kind of place, and spreads it across your one question about a casserole recipe. Most facilities are not that facility. The claim is not a lie exactly. It is a worst case wearing a trench coat, pretending to be the average.

    What the one serious state study found

    Virginia is the data center capital of the world, so Virginia’s own legislative watchdog, JLARC, went and looked at the actual numbers and published them in December 2024. Their finding is the least exciting sentence in this whole fight, which is usually a sign it is the true one: most data centers use about as much water as a large office building.

    Sit with that. Not a reservoir. Not a river. An office building. The kind with a fountain in the lobby that has not worked since 2019.

    Now, the watchdog said most, and most is doing real work in that sentence. A few facilities use far more, and the design is the whole ballgame. A building cooled mostly by air uses very little water. A building cooled by evaporation uses a lot, especially in August. Same company, same logo on the fence, completely different drinker. Which is why the only number that matters for your town is not a national average. It is the number in that project’s permit, the ceiling the state lets them draw, in writing, before anybody breaks ground. If the people pitching the project cannot hand you that number, the meeting is not over, no matter what the clock says.

    One more fact that surprises people: according to the Congressional Research Service, about 97 percent of the water American data centers use comes from municipal systems. It is not a secret well in the woods. It is the same public water system your kitchen tap is on, which is exactly why it is a public question and not a private one.

    The water is not even the biggest bill

    Here is the part that gets lost. The fight is usually about water because water is easy to picture. But the heavier number is electricity. The International Energy Agency estimates a typical AI data center uses about as much electricity as 100,000 homes. One building. A hundred thousand homes. When a town argues about a data center, the water debate fits in a headline. The power debate is the one that shows up later, in grid upgrades and rate cases and a substation nobody budgeted for, long after the ribbon cutting photos fade.

    So no, the building is probably not going to drink your town dry, and yes, it is absolutely going to lean on your grid like it owns the place. Both things can be true. Around here, they usually are.

    What to ask before your town votes

    Ask for the permitted water draw, in gallons, in writing. Ask what kind of cooling it uses, and what it uses on the hottest week of the year, not the average week. Ask where the water comes from and what happens to it after. Ask what the power draw is and who pays for the upgrades when the grid needs them. And ask how many permanent jobs the building actually keeps once the construction crews leave, because that number has a way of shrinking between the announcement and the anniversary.

    None of that makes you anti-progress. It makes you a resident reading the fine print on a building that will outlive everybody on the council that approved it.

    The flat version

    A data center is not a vampire and it is not a toaster. Most of them drink about like an office building, a few drink like a small town, and every single one of them eats electricity like a city. The rumor about the bottle of water is a worst case dressed as an average. The real questions are local, they are in the permit, and they have numbers attached. Get the numbers before the vote, while getting them is still free.

    Related reading on Exit Desk:

    Sources:

    • GridSight, Community Questions (Google Keyword Planner data, refreshed November 1, 2026): “data center water usage” about 9,900 searches per month, up 654 percent year over year; “is it true that one AI prompt uses a bottle of water” about 8,100 searches per month: https://thegridsight.com/community-questions
    • JLARC, Data Centers in Virginia (December 2024): most data centers in Virginia use about as much water as a large office building, though a few use far more; a project’s water permit sets the ceiling.
    • Congressional Research Service, R49057: about 97 percent of US data center water is municipally supplied.
    • International Energy Agency: a typical AI data center uses as much electricity as 100,000 homes.
  • The Absurd Economics of the Future: Silver, Plywood, and the Power Crunch Facing AI

    December 28, 2025

    I can’t get this comparison out of my head.

    This weekend, as spot silver hovered around $76–$80 an ounce amid thin holiday trading. With intraday volatility briefly pushing prices higher earlier in the week, I checked the cost of something far more ordinary: a sheet of 3/4-inch plywood.

    At New Jersey big-box stores, standard 4×8 construction plywood typically runs $35–$65, depending on grade, with higher-quality panels priced above that.

    In other words, an ounce of silver is now priced roughly on par with a single sheet of plywood.

    Let that sink in.

    Plywood is a building material.
    Silver is a critical input for the electrical systems shaping the future economy.

    And yet the market is treating them as near equivalents.


    Why Silver Matters More Than Most People Realize

    Silver is not a decorative metal or a speculative novelty. It is the most electrically conductive metal on Earth, and it sits quietly at the center of modern infrastructure:

    • Solar panels
    • Electric vehicles
    • Power grids and substations
    • Semiconductors and advanced chips
    • Data centers
    • Telecommunications equipment

    Every major push toward electrification relies on it.

    Plywood supports structures.
    Silver enables systems.


    The Energy Constraint Behind the AI Boom

    Artificial intelligence continues to scale rapidly. Compute capacity keeps expanding, and each new generation of models requires more processing power than the last.

    But there is a hard constraint software alone cannot overcome:

    Electricity.

    Training and operating large-scale AI systems already consumes power equivalent to small cities. Industry projections suggest data centers could account for roughly 8–10% of global electricity demand by the end of the decade.

    The challenge is not ambition.
    It is infrastructure.

    • New power plants often take a decade to permit and build
    • Nuclear expansion shows promise, but timelines remain long
    • Natural gas helps bridge gaps, but it is not infinite

    The fastest path to new capacity remains:
    solar generation, battery storage, and grid upgrades

    All of which depend heavily on silver.


    Solar, Silver, and a Supply Reality Check

    Each solar panel uses approximately 15–20 grams of silver. Multiply that across the terawatts of solar capacity planned worldwide, and demand rises quickly into the tens of thousands of metric tons.

    That demand stacks on top of:

    • Electric vehicles
    • Consumer electronics
    • Industrial electrification
    • AI and data center hardware

    At the same time, silver supply is not accelerating.

    • Global mining output has largely plateaued
    • New mines take years to develop
    • Recycling helps but cannot close the gap
    • Refining remains highly concentrated

    Beginning January 1, 2026, new export licensing and controls from major refining hubs, including China, are expected to further tighten availability. For markets already running persistent deficits, that matters.

    This is not hype.
    It is a physical supply issue.


    What This Means for New Jersey Residents

    These forces are already visible in New Jersey.

    The state is seeing a surge in data center development, particularly in South Jersey. In Vineland, Cumberland County, Nebius and DataOne are building a 300-megawatt, AI-focused data center campus, phased over time and spanning millions of square feet.

    Projects like this bring:

    • Construction jobs
    • Technical and operations roles
    • Local tax revenue

    But they also bring enormous electricity demand.

    Regional grid operator PJM has warned that data center growth is now one of the largest drivers of future demand, contributing to recent capacity auction outcomes. Utilities such as PSE&G and JCP&L have already signaled that higher capacity and transmission costs could translate into double-digit percentage increases in future rate adjustments.

    For households, this shows up as:

    • Higher monthly electricity bills
    • Greater peak-load stress during summer and winter
    • Slower timelines for grid upgrades and new connections

    When critical materials and generation capacity tighten, costs tend to flow downstream to consumers first.


    Why This Price Disconnect Is Unlikely to Persist

    Markets often price stories quickly and constraints slowly.

    Silver is still treated like a traditional commodity. In reality, it is increasingly critical infrastructure.

    If electrification, grid modernization, and AI expansion continue on their current trajectory, the physical demand for silver will not align with prices that treat it as interchangeable with construction materials.

    When that realization fully sets in, repricing tends not to be gentle.

    Because:

    • Electrification is continuing
    • AI development is accelerating
    • Power demand is rising

    The materials that make those systems possible will be pulled along with them.


    The Bottom Line

    An ounce of silver, essential to the electrical systems powering tomorrow’s economy, is priced roughly the same as materials we use to frame floors and decks today.

    That is not equilibrium.
    It is a mismatch between financial pricing and physical reality.

    And mismatches like that do not last forever.

    The future will be electric.
    It will be power-intensive.
    And it will depend on materials the market is only beginning to reprice.