How You Can Use AI and Still Care About Climate Justice

Data centers are projected to consume up to 1,050 terawatt hours of electricity by 2026, and the environmental costs of that consumption aren’t spread evenly. They’re concentrated in specific communities, often ones that were already overburdened. You don’t need to stop using AI, but rather consider how to use it intentionally, and hopefully for good.

Data centers already consume about 1.5% of global electricity, and servers built specifically for AI workloads are expected to jump from 93 TWh in 2025 to 432 TWh by 2030, nearly a fivefold increase. In Virginia, data centers already consume about 26% of the state’s electricity, and in Ireland, they’re on track to hit 32% by this year.

Then there’s water. A large data center can use up to 5 million gallons of water per day, roughly equivalent to a town of 50,000 people, and the communities absorbing that impact aren’t getting much say in the matter.

Who’s Paying the Price

The environmental burden of AI infrastructure is landing hardest on communities that have been fighting for clean air and clean water long before anyone was talking about large language models.

• In South Memphis, a historically Black community, Elon Musk’s xAI is powering its Colossus 2 data center with 27 gas turbines installed just across the state line in Southaven, Mississippi, with no air permit. Those turbines could emit more than 1,700 tons of nitrogen oxides per year, which would make the facility the largest industrial source of smog-forming pollution in the greater Memphis area, a region already failing to meet national smog standards. The NAACP filed a Clean Air Act lawsuit against xAI on April 14, 2026, represented by Earthjustice and the Southern Environmental Law Center.
• In Amarillo, Texas, the proposed Fermi America campus would transform 5,800 acres into 18 million square feet of data centers, drawing 2.5 to 5.5 million gallons of water per day from the Ogallala Aquifer, a water source already stretched thin by agriculture and drought and being drained faster than it can replenish.
• The pattern scales globally, too. The African continent holds around 30% of the world’s critical mineral reserves needed for AI hardware, but captures only about 10% of the revenue generated from those resources. Low-income countries in the Global South are absorbing massive data center development because they offer cheap land, weak environmental regulations, and natural resources that big tech can tap with limited local oversight. Critics have called this “extractive colonialism,” and the label fits.
• And the job promises don’t hold up either. In Chile, Microsoft announced that its data center investment would generate over 81,000 jobs, but permit filings tell a different story: the largest facilities create only hundreds of full-time positions, mostly in security and cleaning. One Microsoft data center permit showed just 75 long-term jobs over 30 years of operation. The promise of economic development is being used to open doors that local communities are increasingly trying to close.

What Organizations Can Do About It

A 2025 UCL study published in a UNESCO report found that practical changes to how people use AI, like shorter prompts, shorter responses, and smaller models for specific tasks, can reduce energy consumption by up to 90%. 

If You’re an Individual Contributor

The person with the most daily AI interactions has more influence over energy use than they might realize.

• Be specific in your prompts. Vague prompts generate longer, more energy-intensive responses, and the UCL study found that halving response length cut energy use by 54%. “Summarize this report in 3 bullet points” costs a fraction of “tell me everything about this report.”
• Pick the right model for the job. You don’t need GPT-4 or Claude Opus to draft a meeting agenda, and smaller models use far less energy for routine tasks.
• Don’t default to AI when you don’t need it. A Google search still uses roughly 3 to 10 times less energy than a ChatGPT query, so if you’re looking up a fact, search first.
• Batch your requests. Instead of 10 separate prompts, give the model context and ask it to handle related tasks in one conversation, because fewer round trips means less compute.
• Ask for visibility. Push your tools and your company to surface usage data, because you can’t change what you can’t see.

If You’re a Team Lead or People Manager

You set the norms, and what you model is what your team copies.

• Build “AI hygiene” into team practices. Create lightweight guidelines for when to use AI, which models to default to, and when a simpler tool will do, and make it part of onboarding rather than an afterthought.
• Run an AI usage audit. Most teams have no idea how much AI they’re consuming, so spend a sprint tracking it, because you’ll find both waste and opportunities.
• Reward doing more with less. If someone figures out how to get the same output from a smaller model or fewer prompts, that’s worth celebrating in standups and retros.
• Make it safe to question AI use. A lot of people feel weird pushing back when the pressure is to adopt, adopt, adopt, so create space for the conversation about whether a given task needs AI at all.
• Pair AI literacy with climate literacy. When you train your team on new AI tools, include a 5-minute section on the environmental cost, because context changes behavior.

If You’re an Executive or Senior Leader

You control the infrastructure decisions and procurement that move the needle.

• Ask your AI vendors hard questions. Ask about energy sources, water usage, and carbon reporting, and if they can’t answer, that tells you something. The lack of reporting standards across the industry is a real problem, and buyer pressure is one of the fastest ways to change it.
• Set AI usage policies that include sustainability. “Use AI responsibly” isn’t a policy, so specify model tiers for different use cases, require teams to justify compute-heavy workflows, and build energy considerations into procurement criteria.
• Spend on doing more with less compute. The World Economic Forum reports that on-device AI processing can reduce energy use by 100 to 1,000x per task compared to cloud-based processing, so it’s worth exploring edge computing, model compression, and right-sizing before defaulting to the biggest model available.
• Tie AI strategy to your climate commitments. If your org has sustainability goals, your AI adoption plan should reference them, because these two strategies can’t live in separate spreadsheets.
• Advocate for industry standards. Support efforts to create consistent reporting frameworks for AI’s environmental impact, join coalitions, and use your influence.

The excitement around AI is enormous, and a lot of it is deserved, but when the cost of that excitement falls disproportionately on communities already carrying more than their share, it’s worth paying attention. Using AI and caring about the planet aren’t in conflict, they just require the same thing: being thoughtful about where the costs land, and making choices that account for more than convenience.

So use AI, ideally for good. Just don’t be wasteful about it.