We use cookies to understand how visitors use our site and improve the experience.
For sustainability-minded clients
An honest look at the numbers, and why a small AI-augmented business is one of the greener ways to build.
If you care about your environmental footprint, AI is a fair thing to ask hard questions about. This page looks at the data: energy, water, the resource cost of running a business, the parts AI gets right, and the parts the industry still has to answer for. Honesty over reassurance.
01 · The premise
If you've read coverage of AI's environmental impact, you've probably seen scary headlines about data center electricity, water for cooling, and emissions growth at big tech. Those stories are real, and so is the upside. The question is what the numbers actually look like at the scale a small business uses AI, and whether the trade-offs work out.
Our position: we'll show you the data with both sides accounted for. We'd rather tell you something is harder than the headlines say than promise you something easy. Where there's a trade-off, we'll name it.
02 · Per-task energy
A single AI prompt uses less energy than most people's regular digital habits. A typical workday of AI prompting still uses far less energy than an evening of streaming.
| Activity | Approximate energy use |
|---|---|
| One AI query (e.g. ChatGPT, Claude) | ~2.9 Wh1 |
| One Google search | ~0.3 Wh (typically several per task) |
| One hour of HD Netflix streaming | ~70 Wh2 |
| A 10-hour Netflix binge | ~700 Wh, equivalent to ~240 AI queries |
What that looks like at workday scale: a heavy AI user running 50 prompts a day uses about 145 Wh, roughly the energy of two hours of HD streaming. That's for a full day of AI-assisted work.
03 · Water
Data centers use water to cool the servers that run AI. Estimates of how much water a single AI query "uses" range widely, and the honest answer is that there's no single number.
| Estimate source | Per-query water cost |
|---|---|
| UC Riverside & UT Arlington3 | ~519 ml per 100-word prompt (≈ a small bottle) |
| Academic GPT-3 lifecycle study4 | ~17 ml per 150–300 word output (cooling + electricity) |
| OpenAI public statement | ~0.3 ml per query (≈ 1/15 of a teaspoon) |
Why such a wide range? The high end counts the full lifecycle (cooling water plus the water used to generate the electricity the data center runs on, which depends heavily on where the power plant sits). The low end counts only direct cooling at the data center. Different methodologies give different numbers, and both are honest answers to slightly different questions.
For comparison: a single 5-minute shower uses ~75 liters of water. Brushing your teeth with the tap running uses ~6 liters. Even at the highest per-query estimate, AI prompts are a small daily-use activity by comparison.
The aggregate picture is real, though: a typical data center uses 300,000 gallons a day, and large ones up to 5 million gallons. US data center water consumption is projected to double or quadruple by 2028.5 The individual prompt is small. The industry's growth is not.
04 · The bigger picture
Where AI's environmental case becomes genuinely strong is at the structural level. A small business running on AI agents avoids the largest categories of office-based emissions entirely.
A traditional five-person office runs HVAC, lighting, lifts, kitchens, and shared infrastructure for forty-plus hours a week, then eats five daily commutes on top. None of that exists for a solo founder running an AI-augmented business from home. Buffer's own audit measured 4.9 tons of CO₂e per employee per year as a fully remote company, versus ~5.8 tons for a comparable in-office team.8
What an AI-augmented small business avoids
The structural argument: the most environmentally meaningful choice a founder can make isn't which AI provider they pick. It's choosing to stay small and remote in the first place. AI is what makes that choice viable for many businesses where it wasn't before.
05 · The honest caveats
There are real costs we don't get to wave away just because per-task numbers look good. Here's what we acknowledge.
Training a large AI model uses orders of magnitude more energy than running it. The cost is paid once and amortised across millions of users, but it's real, and it's the largest single environmental line item in AI development. The honest framing: using existing trained models is much cheaper than training your own.
Even as individual AI queries have gotten more efficient, the industry's aggregate emissions have grown. Reports show Google's emissions up nearly 50%, Microsoft's up 23%, and Meta's up 60% over recent years. US data center electricity could nearly triple from ~4.6% to ~13.8% of total demand by 2028. The macro trend is concerning even when the micro math checks out.9
Cheaper, more efficient AI tends to lead people to use more of it (this is sometimes called the Jevons paradox). A small per-query footprint multiplied across billions of new users can outweigh efficiency gains. Using AI thoughtfully matters more than the per-query number suggests.
Most AI providers don't publish detailed water or energy data per query, and figures vary by region and time of day. We work with what's publicly available, and we're honest when we don't have a precise number to give.
06 · Practice
The choices we make at the prompt level matter, not because any single one is large, but because the habits compound across an entire engagement.
Operating principles for our work with you
07 · Provider choice
Our agent team runs on Claude, made by Anthropic. We picked them on output quality first, but their public commitments on infrastructure impact mattered to the decision too.
Anthropic has pledged to pay 100% of the grid upgrades needed to interconnect their data centers, paid through their own electricity charges rather than passed onto local ratepayers.10
Anthropic has publicly committed to deploying water-efficient cooling technologies across their infrastructure expansion.
Anthropic is partnering with local leaders on initiatives that share AI's benefits broadly with surrounding communities.
To be clear: all major AI providers are pushing toward renewable infrastructure, and the picture changes constantly. We picked Claude on a mix of output quality, public commitments, and how we like to work, not because we think any single provider has a monopoly on doing this right.
AI doesn't make a business green. The structural choices around it do: staying small, staying remote, avoiding the office, and using AI to do the work a larger team would have done. That's the model we offer. If it fits what you're building, we'd love to talk.
Let's talk[email protected] · WhatsApp +971 55 252 6779
08 · References