More Data Centers Are Coming. Here’s How Communities Can Come Out Ahead

The headlines have settled into a single story: AI is coming for our water and our power grid. In communities where data centers are being built fastest, like Northern Virginia, the Phoenix suburbs, and parts of the Midwest, residents understandably ask who will bear the cost of any strain on local water and power. The narrative of data centers' water and energy use has taken hold because the numbers involved are genuinely large and these utilities are critical for all of us. That said, few of us have an instinct for how much water and electricity ordinary life already uses. As a read on what industry and communities need from shared infrastructure undergoing major changes, the numbers deserve a hard look. So does a better question: not whether data centers get built, but what the communities that host them get in return.

The Numbers

Water

In 2023 (the most recent year with comprehensive national water estimates), data centers in the United States collectively consumed about 17 billion gallons of water on-site for cooling, according to the Lawrence Berkeley National Laboratory. That on-site figure understates it: most of a data center's true water footprint is upstream, at the power plants generating its electricity. LBNL estimates those plants consumed roughly 211 billion gallons in 2023. This is net of what returns to the rivers, since 97% of what those plants pull goes right back according to the USGS. Scaling both figures by the growth in data center electricity use since 2023, roughly 70% higher today, puts current consumption at something like 29 billion gallons on-site and 360 billion gallons upstream: around 390 billion gallons a year all in.

Which water they use matters as much as the quantity. The power plants are using mostly river and reservoir water, but on-site is different. The Congressional Research Service reports tap water supplied roughly 97% of data centers' on-site needs in 2025, about 0.3% of the nation's drinking water. It is a small share of what people drink, but it is the same water, which is why communities notice. The strain is also uneven: Bloomberg reported that roughly two-thirds of U.S. data centers built or planned since 2022 sit in water-stressed regions. Those are real local concerns, and we will get to the fixes in a bit. First, some context on scale.

So is 390 billion gallons a lot? Golf courses use about 531 billion gallons a year, according to the United States Golf Association, and California's almond orchards use somewhere between 1.2 and 1.8 trillion. What about more everyday uses?

Car washes process roughly two billion vehicles a year at about 40 gallons each: around 80 billion gallons a year, even with closed-loop recycling. The average American uses about 82 gallons a day at home, according to the U.S. Geological Survey, and roughly 1,800 gallons a day once you count the water embedded in the food, goods and electricity they buy, according to the Water Footprint Calculator. Across 340 million people, that comes to roughly 10 trillion gallons a year at home, and around 220 trillion all in.

Then there is the water we simply waste: U.S. distribution systems lose roughly 2.1 trillion gallons of treated drinking water every year to leaks and aging pipes, according to the American Society of Civil Engineers.

Electricity

The Lawrence Berkeley National Laboratory's most recent report, published in June 2026 and covering 2024 actuals, puts U.S. data center electricity use at 192 terawatt-hours, or 4.7% of total U.S. electricity consumption. Globally, the IEA puts data centers at around 1.5% of world electricity demand, on track for roughly 3% by 2030. In the United States, where the buildout is most concentrated, the share is already three times the global figure, and LBNL's reference case has it reaching roughly 12% of U.S. electricity by 2030. That's a big uptick, although the sector has absorbed this kind of pressure before: Between 2010 and 2018, the world's data centers took on six times more computing, with essentially flat energy use, as reported in Science.

Driving Momentum

Yes, the build-out is big. Big enough, it turns out, to be one of the largest single drivers of American economic growth. Investment in data centers, hardware and networking reached about 1.4% of GDP in early 2026, double its level a year earlier, and roughly half of that quarter's GDP growth traced to the equipment category the build-out dominates. Spending on the physical facilities is up 4x since 2020, propping up a construction sector that has otherwise been soft. Regardless of what we think of it, an economy will not walk away from that.

The demand side has changed shape too, and most people haven't noticed: AI has become its own biggest customer. In early February 2026, agentic token usage passed human token usage for the first time on OpenRouter, the platform that routes traffic across hundreds of AI models. Six months later, agents were consuming more than five times what people were.

That shift changes the math for infrastructure. Human demand for computing is capped by human hours, by how many prompts a person can read and answer in a day. An agent running autonomously for hours has no such ceiling.

The Strain

In the places where data centers are being built fastest, the strain on power is increasingly visible. Bloomberg reported wholesale power prices within 50 miles of major data center activity up as much as 267% in five years. Those are wholesale rates, not household bills. Retail bills are up too, about 42% nationally over five years, according to the Energy Information Administration. Most of that increase, though, traces to weather, fuel costs and long-deferred grid spending rather than data centers.

The clearer connection is local and specific. Pepco customers in Washington, D.C. saw bills rise roughly $21 a month from June 2025, and the district's Office of People's Counsel traced about half of that to a spike in grid capacity prices, which PJM's independent market monitor in turn attributes largely to data center load forecasts. That attribution is an estimate, and the utilities dispute it, but the direction of the pressure is hard to miss.

The national picture cuts against the easy story: states with the fastest load growth, Texas and Virginia among them, have seen the smallest rate increases, while states with flat or falling demand, like California and New York, have seen the largest. E3's study for Virginia's legislative audit commission, in the world's largest data center market, found no evidence of costs being shifted from data centers onto households. Data centers are simply the most visible newcomer on a bill that is rising because of gas prices, grid spending, storm recovery and inflation all at once. The fight should be about who pays for the upgrades, and whether the new power is low carbon.

At the local level, what matters is whether a given grid can carry a given facility, and what that facility puts back. It's a familiar pattern with infrastructure: Airports deliver national economic benefits, but often impose local costs, and we long ago decided they owe local compensation: noise abatement, curfews, community funds. Data centers should be held to the same standard. The question is not "data centers, yes or no?" A better question is: "what do the communities that host them get in return?"

We don't have to speculate about what that looks like in practice. In July 2025, Ohio's utility regulator ordered that data centers must pay for at least 85% of their subscribed electricity capacity for up to 12 years, meaning they cannot lock up grid infrastructure and then walk away if AI demand softens, leaving ordinary households holding the bill for stranded upgrades. That ruling is being watched closely in other states. The precedent: the grid gets stronger because of data centers, not in spite of them, and ordinary ratepayers don't subsidize the margin.

A community playbook is emerging from these fights, and it has four asks:

1. Require large facilities to pay for the capacity they reserve, so stranded costs never reach household bills.

2. Make reclaimed water part of the deal, with the city owning the infrastructure afterward, as Quincy, Washington did.

3. Put the waste heat to work in pools, schools, and businesses or convert it to power rather than venting it to the sky.

4. Secure commitments to ease demand on peak days, so the facility stabilizes the grid it sits on.

None of this is hypothetical. Every item on the list has already been done somewhere.

Cause for Optimism

The same demand that's straining water and power systems is now large enough to rebuild them, for everyone. At scale, data centers become the anchor customers that make otherwise unfinanceable infrastructure projects viable. The hyperscalers bring the demand. Increasingly, it is startups that bring the fixes. Every ask on that community list is also a market.

Opportunity for Founders

This is the future worth building toward. The choice was never between stopping the build-out and building at any cost. We need to build in a way that leaves the water cleaner, the grid stronger and power cheaper.

That's an enormous, fundable, near-term opportunity, and there are plenty of startups that stand to capitalize on it: companies like Iceotope, GRC or ZutaCore, working on advanced cooling; Gradiant, WaterBridge or Membrion, looking at water re-use and treatment; Form Energy, Exowatt and LineVision in different areas of grid hardware / storage; or Zanskar, Commonwealth Fusion and many others working on clean power. This is where venture investors can support change, working alongside founders, operators, utilities and communities. Three we have backed:

• SOLARCYCLE recovers 95% of the metals from end-of-life solar panels and feeds them back into new ones. That takes cost out of the next round of solar. Cheaper panels mean cheaper clean power.

• Molg builds robotic microfactories that disassemble retired servers so the components and metals go back into the supply chain.

• ZwitterCo makes fouling-resistant membranes that recover usable water from streams too dirty to treat conventionally, including data center cooling tower blowdown.

At HG Ventures, we back the teams moving the economy forward sustainably by solving its hard, physical problems. For nearly a century, The Heritage Group's businesses have lived in the essential, unglamorous layers of the real economy: environmental services, recycling, specialty materials, and the infrastructure that keeps a country moving. That history is why we don't see the infrastructure demands created by data centers as a reason to stop building. We see them as a forcing function to build better, and the best chance in decades at a more sustainable future.

Data centers are the catalyst in this story, but they are not really the point. The point is what happens when new sources of industrial demand collide with physical infrastructure that has not kept pace. Power, water, materials and the grid are becoming constraints on economic growth, and we think that collision will define some of the most important technology markets of the next decade. More data centers are coming. The open question is whether this build-out is something that happens to our communities or something we steer, so that the investment strengthens the infrastructure underneath them and communities come out ahead. That is the opportunity we are focused on. If you're building a company that can turn this wave of infrastructure investment into lasting benefits for the communities where it takes shape, we want to hear from you.