The Ugly Truth About Data Centers — and the Plan to Move Them to Space

Every time you ask a chatbot a question, stream a movie, or back up a photo, that work happens in a giant warehouse full of computers running around the clock. Those warehouses — data centers — now use about 1.5% of the world's electricity, drink billions of gallons of water for cooling, and pile up fast-growing mountains of e-waste. The AI boom is pouring gasoline on all three. This is the honest version: what's real, what's overblown, what it means for your power bill in South Georgia, and the genuinely wild plan to move the whole mess into outer space.

I'll be straight with you. I make my living on this stuff — I set up the computers, the cloud backups, the AI tools that save local shops real hours. I'm not here to tell you to feel bad for using technology. But when a customer asks me "is all this AI actually hurting the planet?", they deserve a real answer with real numbers, not a scare headline and not a Big-Tech press release. So let's talk about it plainly.

If you want the friendly primer on what a data center even is and why it matters, I wrote that one first: Data Centers: The Digital Backbone and Its Environmental Impact. This post is the tougher follow-up — the ugly truth, and where it's all headed.

How much electricity are we really talking about?

A lot, and it's climbing fast. According to the International Energy Agency's 2025 Energy and AI report — the gold-standard source on this — data centers used about 415 terawatt-hours of electricity in 2024, roughly 1.5% of the world's total. That consumption has grown around 12% every year since 2017, more than four times faster than electricity demand overall.

Here's the part that matters for the next few years: the IEA projects data-center electricity use will more than double to around 945 terawatt-hours by 2030, and says plainly that "AI is the most important driver of this growth." In the United States specifically, the U.S. Department of Energy's Berkeley Lab found data centers already ate 4.4% of all U.S. electricity in 2023 — and that could reach anywhere from 6.7% to 12% by 2028. The IEA figures data centers will account for nearly half of all the growth in U.S. electricity demand between now and 2030.

Think about that. Nearly half of the new electricity demand in the country isn't factories or electric cars or new houses — it's computers thinking.

Does AI really use ten times more than a Google search?

You've probably seen that claim. It's outdated, and I'd rather give you the honest number. The "ten times a search" line traces back to a 2024 estimate from the Electric Power Research Institute that pegged a ChatGPT query at about 2.9 watt-hours versus 0.3 for a traditional Google search. It got repeated everywhere.

But newer, more careful analysis has cut that way down. In 2025, the research group Epoch AI estimated a typical chatbot query actually uses around 0.3 watt-hours — about ten times less than the old figure — because the original estimate assumed slower hardware and unusually long answers. Google, publishing its first measured numbers that same year, put the median AI text prompt on its Gemini app at just 0.24 watt-hours — roughly what an LED bulb uses in a couple of minutes — with 0.26 milliliters of water and a tiny puff of carbon.

So one question is basically nothing. The catch is the word "one." Multiply a tiny number by billions of queries a day, then add the staggering amount of power it takes to train these models before you ever type a word, and you're back to those 945 terawatt-hours. The problem was never your single question. It's the scale.

The thirsty part: data centers and water

This is where the scary stat lives, so let me defuse it first. You may have heard AI "drinks a bottle of water per question." That's a misquote. The University of California, Riverside team that produced it estimated 500 milliliters for an entire conversation of 20 to 50 back-and-forth exchanges on an older model — and they've since revised a single modern prompt down to roughly 15 milliliters, about a tablespoon. OpenAI's Sam Altman claimed an even smaller figure. Per-question, water use is small.

The real issue isn't the per-question number — it's the totals and the location. Data centers use water two ways: directly, to cool the servers, and indirectly, through the power plants that feed them. The Department of Energy's Berkeley Lab found U.S. data centers directly consumed about 17 billion gallons of water for on-site cooling in 2023, and that could climb to 38–73 billion gallons a year by 2028. A couple of real examples that stuck with me:

That's the honest water story. A single prompt is a tablespoon. But build a mega-campus in a drought-prone county and have it quietly draw down a quarter of the town's water, and you've got a real fight on your hands.

The e-waste and materials nobody talks about

Electricity and water get the headlines. The stuff that gets forgotten is everything the industry builds and throws away. A peer-reviewed study in Nature Computational Science projected that the generative-AI boom alone could add 1.2 to 5 million metric tons of electronic waste between 2020 and 2030 — servers and chips that get scrapped as they're constantly replaced with faster ones. The one bit of good news: the same study found smarter reuse, repair, and recycling could cut that by as much as 86%.

For context, humanity already generated a record 62 million tons of e-waste in 2022, according to the UN's Global E-waste Monitor, and only about 22% of it was properly recycled. AI hardware is now stacking onto that pile. And building the chips takes scarce materials: the IEA projects that by 2030, data centers could consume about 2% of the world's copper and up to 11% of its gallium — a metal China refines about 95% of. Even one high-end AI server carries a hidden cost: NVIDIA's own datasheet puts the "embodied" carbon of building a single HGX H100 system at about 1,312 kilograms of CO₂ before it's ever switched on, most of it from manufacturing the memory and chips.

Will this raise my power bill in South Georgia?

This is the question that actually hits home, so let me be careful and local about it. You've seen national headlines about data centers spiking electric bills. A lot of that comes from the PJM grid up North, where the independent market monitor blamed data centers for 63% of a huge jump in "capacity" prices, and the Natural Resources Defense Council estimates the average family there could pay about $70 a month more by 2028.

Here's the honest local caveat I always give: South Georgia is not on the PJM grid. We're Georgia Power territory, under the Georgia Public Service Commission. So those specific PJM numbers don't land on your Valdosta bill one-for-one. (I dug into this distinction in my write-up of the tech-leader White House dinner and the data-center electricity build-out.) But don't relax too much, because Georgia is right in the thick of it.

The Georgia PSC's own fact sheet lays out the scale: Georgia Power's projected need for new power generation exploded from about 400 megawatts in 2022 to 8,500 megawatts by 2025 as data centers poured in — against roughly 22,000 megawatts the company runs statewide today. In December 2025 the Commission certified nearly 10,000 megawatts of new generation, about 80% of it expected to power data centers. To feed that, Georgia Power's plan keeps coal plants running longer and adds new natural gas.

The fair-to-everyone question is: who pays for all that? To their credit, Georgia regulators have taken steps to keep the cost off regular households:

That's genuinely encouraging, and I'll give credit where it's due. But here's the reality check: Georgia Power residential customers have already absorbed six rate increases since the start of 2023 — for fuel, base rates, and the Plant Vogtle nuclear expansion — adding about $43 to the average monthly bill (Georgia Recorder, citing Atlanta Journal-Constitution tracking). So the protections matter, and they're worth holding regulators to.

One myth to put down while we're here: you may have seen a viral claim that people near data centers saw bills jump 267% in five years. Fact-checkers at PolitiFact rated that "Mostly False" — the 267% was a wholesale market price, not anyone's actual retail bill. The real national increase in average residential electricity cost over five years was about 42%. Still a lot. Just not 267%.

The wild fix: put the data centers in space

So if data centers are straining our power grids and drinking our water, here's the idea a handful of very serious, very rich people are now chasing: stop building them on Earth. Build them in orbit instead.

The pitch is actually clever. In the right orbit, the sun never sets, so a solar panel can be up to eight times more productive than on the ground — unlimited, round-the-clock clean power (Google's own research). There's no neighborhood to anger, no farmland to pave, and no river to drain, because in the vacuum of space you cool the machines by radiating heat away instead of evaporating water. On paper, space fixes the electricity problem, the water problem, and the land problem all at once.

On paper. Here's who's actually chasing it, and how far along they really are — because the gap between a press release and a working orbital data center is enormous:

The Elon Musk piece is the one people ask me about, so let me lay it out. In early 2026, SpaceX and Musk's AI company xAI merged in a deal reported at around $1.25 trillion. SpaceX then detailed "AI1," a compute satellite the width of a jumbo jet designed to run Grok, after filing with the FCC for a constellation of up to a million satellites. Musk's line is that space-based solar data centers will undercut anything on Earth within "two to three years."

My honest take as a tech guy who's seen a lot of "revolutionary in two years" promises? Be excited, but keep your feet on the ground. Right now, exactly one refrigerator-sized satellite has ever run a real data-center chip in orbit, and it trained a model small enough to fit the complete works of Shakespeare. Going from that to a five-gigawatt city of computers in the sky means solving brutal problems: the cost of launching millions of pounds of hardware, cooling that actually works in a vacuum, radiation frying the chips, and — the one nobody likes to mention — what happens when a server breaks and there's no technician for 300 miles straight up. It may well happen someday. It is not happening this year.

So should you feel guilty about using AI?

No. And I mean that. Your personal ChatGPT questions and Netflix nights are a rounding error next to the industry as a whole, and these tools save real people real time and money — I watch it happen in local businesses every week. Guilt-tripping regular folks was never the point.

The honest story is about scale, sources, and fairness: how much we build, whether we power it with clean energy or new gas plants, whether we build it where the water can spare it, and whether everyday ratepayers or the tech giants foot the bill. And there's real reason for optimism — efficiency is improving fast. Google reported the energy footprint of its median AI query dropped 33-fold in a single year. The machines are getting dramatically leaner even as we use them more.

What this means for South Georgia

You don't run a data center, but this still touches you two ways. First, your power bill — keep an eye on Georgia PSC decisions, because whether data centers pay their own way is being decided right now, and it affects every Georgia Power customer. Second, your own tech: the smartest, greenest, cheapest thing most homes and small businesses can do is simply run efficient equipment and use the cloud wisely instead of over-buying hardware that ends up as e-waste.

That last part is exactly what I do. I help Valdosta and Hahira folks pick right-sized computers, set up cloud backups and AI tools that don't waste money, and retire old gear responsibly instead of tossing it in a landfill. If you want a hand making your tech leaner — or you're just curious how AI can help your business without the hype — take a look at our AI setup & training and IT services, or just call me.

Frequently Asked Questions