How does AI make you feel? Excited? Ready to “vibe-code” a smarter future? Or anxious? The anxiety comes from the bill. Data centers guzzle billions of gallons of water. They spew pollution. They burn energy.
Dig deeper into the hype and you hit a hard question. Are these massive leaps in generative AI actually worth the cost?
Right now, hundreds of thousands of GPUs are crammed into facilities worldwide. Nvidia calls its 1999 launch the world’s first GPU. Some trace the tech back to arcade games in the 1970s. The hardware drives everything from smartphones to autonomous cars. It also powers the AI boom that made Nvidia the most valuable company on Earth.
Catherine Flick, professor of ethics at the University of Staffordshire, notes a pattern.
“There are massive hardware developments that start because of Games,” Flick says.
Whether it is virtual reality or AI, the ethical questions start with games. What is the impact on how we interact with reality?
Manufacturing mess
The damage begins long before the GPU powers a chatbot. Mining raw materials leaves scars. Semiconductor factories expose workers to toxic chemicals.
In data centers, GPUs burn water and energy. The result? More greenhouse gases. More air pollution. Climate change accelerates.
When the chip dies, it becomes e-waste.
Is it worth it? Do AI-driven breaks in weather forecasting justify the footprint? What about the GPU in your iPhone? Should it face the same scrutiny?
The local burden
AI has thrust GPUs into the spotlight. For many Americans, it hits close to home.
The US has more data centers than any nation. Tech firms are racing to build hyperscale facilities. Communities are waking up to hulking warehouses next door.
As an environmental journalist, I hear the pushback. Are we making AI the bogeyman? Fast fashion is bad. Consumer tech is bad. Why single out AI?
Some venture capitalists claim bad press on environmental impact hurts AI adoption. Flick laughs at that.
“Isn’t it nice to have environment as a scapegoat?”
She’s not alone. On TikTok, product strategist Ashley Striblet pushed back on VCs blaming consumers. People know fast fashion pollutes. They still buy. They get value.
“I think most people know that ordering clothes from [Shein] or Amazon is not the most environmentally conscious things to do,” Striblet says.
Cost savings justify the purchase for many. Flick agrees. But the benefits of AI? Many consumers aren’t seeing them yet.
“It’s just rubbish what they say their technology can do,” Flick says.
We don’t have personalized butlers yet. The gap between promise and reality is wide. Meanwhile, the costs are real. Utility bills rise. Pollution increases. This isn’t happening in Asia anymore. It’s happening in affluent US neighborhoods.
Environmental injustice
Data centers are also landing in low-income areas. Communities of color have long fought polluters setting up nearby.
The NAACP sued xAI (doing business as SpaceXAI). The lawsuit targets air pollution from gas generators powering data centers. xAI did not respond to requests for comment.
The civil rights group warned tech companies to stay alert. Local campaigns are mounting.
The human cost of progress
Shaolei Ren remembers the black carbon in his childhood home in northern China. He grew up in a coal-mining region. Windows stayed shut. Water was rationed in tanks.
Now an associate professor at UC Riverside, Ren studies the environmental toll of data centers. He looks beyond climate change. He focuses on air quality. Water scarcity.
These costs are invisible to users typing into a chatbot.
Ren sits in his office. A whiteboard behind him is covered in blue equations. He wants a “community-integrated data center.” It is doable. Facilities should not harm residents.
For now, they do.
Powerful GPUs demand more energy. They create pollution. They use vast amounts of water for cooling. This happens around the clock. The lifespan of a GPU in a data center is just a few years.
A 2024 study by Ren and colleagues estimates training Meta’s Llama 3.1 model creates air pollution equivalent to 10,000 car trips between LA and NYC. Meta declined to comment, pointing to its sustainability reports.
The study warns of public health costs exceeding $20 billion by 2038. 1,300 premature air pollution deaths annually by 2030.
Energy and water spikes
Gaming once led GPU energy use. In 2019, US gaming consumed 34 TWh. That was CO2 equivalent to 5 million cars. Newer consoles use more. Grid dirtiness matters.
AI is surpassing that.
According to Lawrence Berkeley National Lab, GPU-accelerated AI servers went from 2 TWh in 2017 to over 40 TWh in 2023. By 2028, consumption could hit 165 to 326 TWh. The lower end equals 8.7 million US homes.
In 2025, Alex de Vries-Gao’s study suggests AI exceeded Bitcoin’s power use. AI likely accounted for nearly half of all data center electricity globally. Carbon emissions reached 32.6 to 79.7 million tons. Compare that to New York City’s annual 50 million tons.
Thirsty infrastructure
Southern California sees the boom. Warehouses define the landscape. Goods from Asia arrive at the Port of Los Angeles. They move to “dry ports” with fulfillment centers. Big rigs buzz around the clock.
Data centers add a new layer. Retirees see quiet neighborhoods transform. Noise complaints rise from generators and cooling systems.
Powering AI is thirsty. It uses water for generation and cooling. de Vries-Gao estimates AI used 312.5 to 764.6 billion liters in 2025. That is the global annual water bottle consumption.
Ren’s 2023 study predicted up to 600 billion liters by 2027.
Annual totals hide the danger. Water use is “spikey.” Peaks hit during heat waves. This stresses local water districts. Drought risk spikes.
Homes use 1.5 to 2 times normal water during peaks. Data centers? 6 to 10 times. Some projects need 30 times more.
US data centers could need 1,451 million gallons of new peak water capacity daily by 2030. The cost: up to $10 billion.
Small community systems struggle with this. They are often underfunded. Infrastructure upgrades are expensive.
Ren argues tech company pledges to recycle or replenish water miss the point. We need transparency. We need planning. Communities should help build infrastructure. Residents shouldn’t foot the bill.
Material footprint
Sophia Falk at Bonn University and David Ekchajzer at Université Paris-Saclay are taking a harder look. They threw thousands of dollars of GPUs into industrial blenders.
They are looking for more donated chips. They want to study the materials. The footprint.
When we use AI, we forget the real-world impact. It is out of sight. Out of mind.
Falk studied the Nvidia A100. 90% of it is heavy metals. Copper. Iron. Tin. Nickel. Silicon. Copper conducts electricity. It requires massive extraction.
