Grid Expectations
AI uses a lot of energy. Here's why nonprofits are using it anyway.
If this topic is on your mind and you’re in San Francisco, join us during SF Climate Week. On Saturday, April 25 from 10–11:30 AM, we’re hosting a peer conversation on AI’s climate impact for nonprofit leaders navigating these tradeoffs. We’ll share what the data shows, then hand the mic to the room. RSVP to save your spot.
Here’s an uncomfortable truth about AI for humanity: the technology we’re using to solve big problems is also creating new ones. AI is an energy hog. Data centers are thirsty. So why are the nonprofit founders working hardest on climate, health, and human dignity reaching for AI tools despite these issues?
Because for the builders we work with, it’s worth it — not in a vague, “ends justify the means” way, but in a specific and measurable one. AI lets resource-constrained organizations do more with less, reach people they couldn’t reach before, and take on problems that once felt unsolvable. I wanted to understand that balance better, so I asked some of the smartest people working directly at this intersection for their take.
Watt Are We Talkin’ About Here?
To set the scene, let’s look at energy usage. U.S. data centers, driven almost entirely by AI, could consume up to 12% of total U.S. electricity by 2028, according to a Department of Energy report. That doesn’t even touch the growing concerns about water consumption and the community impact of data center construction. At the same time, the efficiency of data centers is also improving rapidly. AI models are getting leaner — smaller, more specialized models now routinely outperform landmark models from just a year or two ago.
And critically, all AI is not the same when it comes to the environment. As Gavin McCormick, founder of WattTime and Climate TRACE, points out: some AI models are far more polluting than traditional software, but others are actually less polluting. The variable that matters most? Energy source. “Instead of avoiding all AI, we make sure we only use AI powered by surplus free clean energy,” Gavin says. And, Gavin adds, this solution is growing very fast: many AI companies have now invested heavily in running their systems on clean energy.
To add some perspective, we’ve seen alarming numbers related to fast-growing, new technologies before. In 2010, Netflix was eating up 20% of Internet download bandwidth, and Slate wondered if it would destroy the Internet. It didn’t, but not because the concern was overblown. It’s because infrastructure improved, compression got better, and the industry invested in solutions. The AI story isn’t there yet, but it’s trending in the right direction. Dismissing AI entirely may mean walking away from tools that could help solve the very problems we’re worried about.
Energy In, Impact Out
For resource-constrained nonprofits, this calculus plays out in very concrete terms. AI can replace travel, scale programs without increasing headcount, and do work that otherwise simply wouldn’t get done. The alternative to using AI isn’t “no energy use.” It’s less impact.
Patrick Gold of The Change Climate Project wrote the clearest articulation I’ve seen of how a climate nonprofit can use AI in good conscience. His framing has stuck with me:
“A leaf blower and a bus both use gasoline. But one is a terrible use: loud, inefficient, and exists primarily to move a problem from one yard to another. A city bus full of passengers, on the other hand, moves forty people to work, to school, to doctor’s appointments — efficiently, safely, and without the congestion that would come from 40 individual cars. The question isn’t whether energy is being used. The question is whether the energy is doing beneficial work.”
Patrick’s organization lives this principle. The Change Climate Project measures their AI energy use and reports it the same way they’d report a commercial flight — as a real operational cost that needs to be justified. Their standard: if using AI accelerates the mission and allows them to accomplish more than they would without it, they’ll use it. But they’ll use the least amount necessary to get there.
Caroline Spears, Executive Director of Climate Cabinet, sees AI as both a tool and a responsibility. “At Climate Cabinet, we use AI to accelerate our impact — sorting through hundreds of thousands of climate policies each year and building game-changing tools,” says Caroline. “But we can’t ignore the serious climate implications of the AI-fueled data center buildout: rising electricity prices, water stress, and pollution that’s driving real public backlash.” The answer isn’t to abandon AI, she argues, but to demand better. “With a BYONCE — bring your own [new] clean energy — mindset, AI can pair growth with clean energy contracts that actually lower electricity prices. The fact that we use AI makes us more committed, not less, to advocating for a better path forward.”
Staying Current
The key question nonprofits should ask when using AI is Patrick’s bus question - Am I using AI for beneficial work that justifies its costs, or do I have a leaf blower in my hands?
A few practical guidelines:
Right-size the tool. Smaller models are more efficient and often excellent for summarization, classification, and drafting. Save the flagship models for tasks that actually need them.
Know your provider. Review their commitments to clean energy and their track record for meeting them. (Climate TRACE is tracking emissions from all AI data centers globally, and will release free tools later this year so anyone can see the data for themselves.)
Time it right. If you want to make your AI use pollution-free, tools exist to help. WattTime offers a service that shifts compute to surplus clean energy, and Microsoft’s Carbon-Aware Kubernetes does similar work.
Track what you spend. If your AI costs are going up, your energy use is going up. Treat AI emissions like travel emissions: justify them, account for them, and be transparent about them.
The builders we work with aren’t waiting for a perfect tool. They’re using what exists right now to move the needle on problems that can’t wait. Yes, AI uses a lot of energy. But so does solving humanity’s hardest problems.
Quick Bytes
Other Sector Stories
Good things come in seven: Wonders of the Ancient World, days of the week, and the AI principles Center for Humane Technology just dropped. Their report, “The AI Roadmap,” offers a framework for how AI should be built, deployed, and governed, covering everything from corporate accountability to international guardrails.
Funders want to see impact before they invest. Nonprofits need investment before they can show impact. Classic catch-22. In a new Candid piece with Elisha Smith Arrillaga and Nick Cain, we dig into why the AI funding gap persists and what it takes to close it.
After seeing how effective its unreleased model, Mythos, can be at finding high-severity security issues, Anthropic has announced Project Glasswing. The Project gives partners the opportunity to use Mythos defensively, with Anthropic’s financial support, prior to its release.
Following last year’s 2025 AI for Humanity Report, Fast Forward is developing the 2026 edition to understand how nonprofits are building, funding, and deploying AI for social impact. If you’re a founder or CTO at a nonprofit using AI, fill out the interest form to be notified when the survey goes live.
APN Opportunities and Funding News
Applications are still open for the $30M Google.org Impact Challenge: AI for Science. Apply by April 17.
UNICEF Venture Fund is offering up to $100,000 in equity-free funding for early-stage startups in UNICEF program countries using AI, ML, or blockchain for climate and children’s health. Must be open-source. Apply by May 17.
For weekly funding updates for your AI-powered nonprofit, subscribe to Funding Forward.
Let’s Talk
I am living and breathing AI for humanity these days. If you are too, drop a comment!
Thank you for approaching this topic in such a practical, clear way!