The internet's physical backbone is being rebuilt around artificial intelligence, and it is happening fast enough to reshape power grids, water supplies, and even how law enforcement treats the people who object. A wave of giant AI data centers is going up across the United States, and the same companies that route your everyday web traffic are re-architecting their networks around AI. Here is a plain-language rundown of what is being built, what it costs the communities living next to it, and the one part of this story you actually control.
Last updated: June 2026
Key takeaways
- AI is driving the largest internet-infrastructure buildout in decades. Single training clusters are now being wired to link more than a million chips, and the first gigawatt-scale campuses are coming online through 2027.
- The cost lands on a small number of communities: higher utility bills, heavy water use, and air and noise pollution, much of it concentrated in the US rather than placed in naturally cooler, water-rich regions.
- Federal agencies have begun tracking some data-center opponents under a new, undefined "anti-tech violent extremism" label, and the deeper question is who ends up controlling all this compute, and what you can do about it.
What is the "Internet of AI"?
For years, data centers were warehouses of servers that stored websites, email, and streaming video. The new generation is different. They exist mainly to train and run AI models, and they are enormous.
At its Cloud Next conference in April 2026, Google introduced the Virgo Network, a "campus-as-a-computer" fabric built specifically for AI. It links about 134,000 chips inside a single facility and is designed to connect more than one million chips across multiple sites into one training cluster, effectively turning many buildings into a single giant computer. Other operators are racing to build "gigawatt-scale" campuses, each capable of drawing as much electricity as a mid-sized city. Industry trackers expect US AI data centers to need roughly 20 to 30 gigawatts of combined power by late 2027.
At the same time, internet providers and cloud companies are embedding AI into the networks themselves, using it to spot congestion and reroute traffic automatically. The internet is getting smarter and far more power-hungry at once, and control over it is concentrating in fewer, larger hands.
The real effect on your home connection
Here is the honest answer most coverage buries: a nearby data center will not slow your connection. Your modem, router, and download speeds work the same way. The real stakes show up elsewhere.
- More AI in everything. Search, customer service, and the apps you use increasingly run through these data centers, so more of your online activity now travels to and from them.
- Smarter, more centralized networks. Providers are using AI to manage traffic, which can improve reliability, but it also concentrates control of core infrastructure in fewer, larger systems.
- Local infrastructure strain. If a large data center moves into your region, the noticeable impact is far more likely to appear on your power and water bills, and in your local politics, than on your internet speed.
The local costs communities are facing
Higher electricity bills
Data centers run around the clock and draw huge, steady loads. In Virginia, the world's largest data-center market, these facilities already consume a large share of the state's electricity, by some estimates around a quarter. A state legislative study by the Joint Legislative Audit and Review Commission (JLARC) warned that unconstrained data-center demand could push Virginia's electricity use up by as much as 183 percent by 2040, against just 15 percent with no new data-center demand, and that residents could end up sharing the cost of new power lines and plants. Regulators have since created a special rate class to make large data centers pay more of their own infrastructure costs, but residential bills have still been rising.
For balance: utilities such as Dominion Energy say data centers largely pay their share, and some recent bill spikes are also tied to unusually cold weather.
Heavy water use, with limited transparency
Many of these facilities use large amounts of water to stay cool. A single large data center can use up to several million gallons per day, roughly the daily consumption of a town of 10,000 to 50,000 people, according to the Environmental and Energy Study Institute. The bigger problem is that no one knows the full picture. Reporting is voluntary and inconsistent, fewer than a third of operators track their water use at all by some industry estimates, and companies like Google and Meta often estimate rather than fully measure their consumption, sometimes leaving leased or third-party sites out of their reports entirely.
The most visible flashpoint is in Georgia. Residents near Meta's Stanton Springs campus, on the Newton and Morgan County line, reported brown tap water and failing wells. In May 2026, Representative Alexandria Ocasio-Cortez held up jars of discolored water at a congressional hearing, prompting the EPA to agree to review data-center water impacts. Meta says it commissioned an independent groundwater study that found no link to its operations, and that all its water comes from the local utility rather than wells.
For balance: efficiency is improving, with newer closed-loop cooling systems using far less water, and some viral "bottle of water per AI query" claims have been overstated.
Air, soil, and noise pollution
To power these sites quickly, utilities are building new natural gas plants, and most facilities keep large diesel generators for backup. Both add air emissions to surrounding areas. Neighbors near data centers also frequently complain about a constant low-frequency hum from cooling equipment that runs day and night, which can affect nearby homes and property values.
Why build them in the US, and in the wrong climates?
A striking amount of this construction is concentrated in the United States rather than spread around the world, and many projects are sited in hot or drought-stricken regions where cooling needs are highest. A naturally cold or water-rich location would reduce the energy and water required for cooling, yet siting decisions are driven more by tax incentives, land, and grid access than by climate. The result is that the heaviest environmental burden often lands on a relatively small number of US communities.
| Impact area | What is happening | Representative figure |
|---|---|---|
| Electricity bills | 24/7 loads strain the grid; residents may share the cost of upgrades. | Virginia demand projected to rise up to 183% by 2040 (JLARC). |
| Water | Cooling consumes large volumes; reporting is voluntary and patchy. | A large facility can use ~5 million gallons/day (EESI). |
| Air & noise | New gas plants, backup diesel generators, and a constant cooling hum. | Can affect local air quality and nearby property values. |
| Siting | Concentrated in the US, often in hot or drought-prone regions. | Driven by tax breaks, land, and grid access, not climate. |
| Civil liberties | Protest and even observation can be logged as "suspicious." | 1,000+ pages of federal reports reviewed by Wired. |
When objecting can get you watched
This is one of the more concerning recent developments, and it sits squarely in our wheelhouse. According to an investigation by Wired, widely covered across the tech press, US law enforcement, including the Department of Homeland Security and the FBI, has circulated more than 1,000 pages of unpublished reports referencing a new category called "anti-tech violent extremism." The term does not appear in any public domestic-extremism guide, which suggests it is brand new.
The documents reportedly group a wide range of people under that single label, and the indicators are loose enough to worry civil-liberties experts. One regional intelligence center listed activities such as photographing or observing a data center as potentially suspicious. A senior counsel at the NAACP Legal Defense Fund cautioned that such reports are often built on vague or innocent behavior. Opposition is widespread: hundreds of community groups across 42 states have organized against data-center construction, mostly over power and water concerns. The worry is that lawful, constitutionally protected dissent could be monitored as a security threat.
See where data centers are being built near you
If you want to check what is planned or operating in your area, including approximate power and water use, environmental activist Erin Brockovich's crowdsourced reporting map is a useful, independent starting point, with source citations on each marker.
The bigger picture: centralized AI versus your digital sovereignty
Step back and the pattern is familiar. The same logic that makes ISP equipment rentals and cloud lock-in a bad deal is playing out here at the scale of national infrastructure: a handful of companies are concentrating the world's AI compute into rented, centralized campuses that you neither own nor control. The water fights and surveillance reports are the visible edge of that concentration.
The part you actually control is your own stack. You do not need a gigawatt campus to use capable AI. A growing class of open-weight models runs privately on hardware you own, with no data leaving your network and no monthly bill to a cloud provider. If that is the direction you want to lean, start here:
- Best hardware for running local AI models, a budget-tiered guide from $300 starter builds to prosumer workstations.
- The best open-source LLMs and the hardware they need, covering which models actually fit on consumer GPUs and Apple Silicon.
- The best mini PCs for local AI, if you want a small, low-power machine that runs models privately.
None of this makes the data-center buildout disappear. But it does mean the centralized version of AI is a choice, not the only option.
Where ModemGuides stands
To be clear, this section is our opinion, not reporting. We believe AI can be a genuinely powerful and beneficial tool, and that it should stay open and accessible, runnable on hardware people own, rather than tightly controlled by governments or a handful of private companies. We also believe the companies building this infrastructure have a responsibility to do so without harming the residents who live nearby.
Right now these data centers are going up so fast that there is little time to study their ecological, environmental, and community impacts before the concrete is poured. Our view is simple: slow down enough to do the studies, and pass reasonable, common-sense rules, noise limits, pollution standards, and fair water and electricity terms, that protect residents and property owners. Building responsibly and building quickly do not have to be opposites. And whatever the policy outcome, keeping a private, local option alive is one way individuals hold a little sovereignty in an increasingly centralized system.
Frequently asked questions
Will AI data centers slow down my home internet?
Not directly. Your modem, router, and connection speed are unaffected by a nearby data center. The bigger local effects are usually on electricity and water, not your download speed.
Do data centers raise electricity bills for nearby residents?
They can. In high-density states like Virginia, regulators have warned that residents may share the cost of new power infrastructure, which is why special rate classes are being created to shift more of that cost onto the data centers themselves.
How much water do AI data centers use?
A large facility can use up to several million gallons per day for cooling, comparable to a small town. Exact figures are hard to confirm because reporting is voluntary and many operators estimate rather than measure their usage.
What is Google's Virgo Network?
Virgo is a data-center fabric Google introduced in April 2026 that links about 134,000 AI chips in one facility and is designed to connect more than a million across multiple sites, so they work together like a single supercomputer. It is part of the broader shift toward AI-first internet infrastructure.
Why are most AI data centers being built in the United States?
Tax incentives, available land, and access to the power grid drive most siting decisions. Critics point out that many are placed in hot or drought-prone areas, where cooling demands are higher, rather than in cooler regions that would naturally reduce energy and water use.
Are people who protest data centers being labeled extremists?
A Wired investigation reported that federal agencies have begun referencing a new "anti-tech violent extremism" category in internal documents. Civil-liberties groups warn the term is vague enough to include peaceful protesters and ordinary residents raising utility or pollution concerns.
Can I run AI without relying on these data centers?
Yes. Open-weight models can run privately on a consumer GPU, a mini PC, or an Apple Silicon machine, with no cloud account and no data leaving your network. Our local AI hardware guide breaks down what you need by budget.
