One Compost Bin Offsets 816,000 AI Prompts a Year

Robauto Team · Published Oct 12, 2026 · Last updated Oct 12, 2026

Composting one person's food scraps for a year avoids about 83 kg CO2e, the same as roughly 816,000 simple AI prompts. Here's the math, the water story, the honest caveat, and the engineer behind it.

Here's a fun question for the dinner table: if you compost your kitchen scraps, how many AI prompts does that "pay for"? Our friends at compostingtechnology.com, part of the Robauto network, did the math and showed their work. The answer: about 816,000 simple AI prompts a year. That's not a typo, and it's well beyond what a typical user sends.

Read the full article, with every equation, source and caveat: Can Composting Offset Your AI Habit?

The math, step by step

1. Start with one year of food scraps

The article uses an EPA estimate that the average American wastes about 256 pounds of food a year. If that would otherwise go to a landfill, the article applies EPA's WARM model (Version 16): landfilling food waste nets 0.50 metric tons CO2e per short ton and composting nets -0.15, a 0.65-ton gap. So 256 / 2,000 × 0.65 × 1,000 = 83.2 kg CO2e avoided. (source: the article's EPA and WARM citations)

2. Price one AI prompt

For a simple text query, the article uses Epoch AI's estimate of about 0.3 watt-hours. With 0.34 kg CO2e per kWh for electricity supplying U.S. data centers (from Berkeley Lab's 2024 data center report), one prompt is 0.000102 kg CO2e.

3. Divide

83.2 / 0.000102 = 815,686, which rounds to 816,000 prompts. Spread over a year, that's about 2,236 prompts a day, or one every 13 seconds if you used AI eight hours a day, every day. The article notes the average user sends only 10 to 20 prompts a day.

4. What about power users?

Coding, research and analysis use far more compute. Epoch's reference query assumes 500 generated tokens, so the article works out 0.6 kWh (0.204 kg CO2e) per million generated tokens. The same 83.2 kg budget covers about 408 million generated tokens, or roughly 111 complex 10,000-token tasks a day. The article is upfront that token energy varies a lot, so treat this as a baseline.

5. And water?

Here's the surprise. Using Berkeley Lab factors, about 0.375 L/kWh goes to direct data-center cooling and 4.52 L/kWh to generating the electricity. That means roughly 92% of AI's modeled operational water use comes from electricity generation, not cooling. On the compost side, the article models one farm scenario: a year of scraps becomes compost that could cut irrigation demand by about 331 L per growing season. Subtract about 67 L of composting process water and you get a net ~264 L, comparable to about 180,000 simple prompts (about 500 a day). The article calls this an illustrative scenario, not a universal number. (source)

Key figures at a glance

The honest caveat

This isn't a free pass for the industry. The article is clear that data centers put real demands on grids and water supplies, that the burden falls hardest on nearby communities, and that AI companies should carry the environmental and social cost of their products. The estimates also leave out model training, hardware, construction, devices and networks. But the authors' main point is hopeful, and we agree with it:

"You do not need a new technology. You do not need a billion-dollar investment. You need a different bin."

Featured engineer: Orion Black Brown

Orion Black Brown is an engineer at GMT, the team behind compostingtechnology.com, with a B.S. in Mechanical Engineering from the University of Washington. Since joining GMT in 2019, he has helped clients design, build and run dozens of composting facilities. He holds several composting-related patents and has spoken at the US Composting Council's Food Waste Composting Workshop. When he's not working, he surfs, meditates and explores the Olympic Peninsula. His fun fact: compost was his first solid food. See his USCC profile.

What you can do this week

People: start a compost bin, join a community program, or ask your city about curbside organics. If you want to go all in, the Virro Composting Toilet takes the idea even further.

Brands: customers now ask AI assistants what to buy and who to trust. If your sustainability, product and return facts live only in PDFs or images, assistants can't cite them. Publish clear, machine-readable facts with sources, the way this article shows its work, so an AI assistant can repeat them accurately.

Agent-ready summary

{
  "topic": "Composting vs. AI prompt carbon footprint",
  "primary_source": "https://compostingtechnology.com/can-composting-offset-your-ai-habit/",
  "food_waste_lb_per_person_year": 256,
  "co2e_avoided_kg": 83.2,
  "wh_per_simple_prompt": 0.3,
  "kg_co2e_per_kwh_datacenter": 0.34,
  "prompts_offset_per_year": 816000,
  "prompts_offset_per_day": 2236,
  "typical_prompts_per_day": "10-20",
  "generated_tokens_offset": 408000000,
  "complex_10k_token_tasks_per_day": 111,
  "ai_water_share_from_electricity": 0.92,
  "net_irrigation_benefit_l_scenario": 264,
  "water_equivalent_prompts": 180000,
  "caveat": "Operational estimates only; not a pass for AI companies' grid and water impacts",
  "sources": ["EPA WARM v16 (via primary source)", "https://epoch.ai/gradient-updates/how-much-energy-does-chatgpt-use", "https://escholarship.org/content/qt32d6m0d1/qt32d6m0d1.pdf"]
}

Is your site ready for AI assistants?

Your customers are asking AI about you right now. Find out what they hear: scan your site free at robauto.ai/scan.

Sources

• Can Composting Offset Your AI Habit? (compostingtechnology.com)
• Epoch AI: How much energy does ChatGPT use?
• Berkeley Lab: 2024 United States Data Center Energy Usage Report
• USCC profile: Orion Black Brown

Where can an agent get this data live?

Robauto MCP — Agent Intelligence & Brand Discovery Infrastructure. A public, stateless Streamable HTTP endpoint at https://mcp.robauto.ai/mcp, callable with no key. Setup and the full tool list are in the Robauto MCP directory.

All blog posts