What is green AI?

By Kevin Naderian · Updated 17 November 2026

Short answer: Green AI means treating energy, carbon and water as core metrics when building and using AI, not afterthoughts. It covers efficient models, right-sizing, clean power and transparent reporting.

Where the term comes from

In 2019, researchers Roy Schwartz, Jesse Dodge, Noah Smith and Oren Etzioni published “Green AI”, contrasting it with “Red AI”: research that buys accuracy with ever more compute. They argued papers should report efficiency, such as the compute used, so progress isn't only measured by bigger models. The same year, Strubell and colleagues estimated the energy and carbon cost of training large NLP models, which helped bring the issue into the mainstream.

The building blocks of green AI

LeverWhat it means
Efficient architecturesMixture-of-experts models activate only part of the network per token
Distillation and smaller modelsTrain compact models to match bigger ones on common tasks
QuantisationLower-precision numbers that need less memory and energy
Right-sizingRouting easy tasks to small models and hard ones to big models
Clean energy and sitingData centres powered by low-carbon electricity, in places with less water stress
Efficient hardwareNewer chips doing more work per watt
TransparencyPublishing energy, carbon and water per query with methodology

Training vs inference

Early green AI debate focused on training giant models. Now that hundreds of millions of people use AI daily, inference, the energy used every time a model answers, is a major share of the total. That's why per-query figures matter: Google reported 0.24 Wh for a median Gemini text prompt, and independent estimates for ChatGPT are around 0.3 Wh. See how much energy AI uses.

Why it matters

The IEA estimates data centres used about 415 TWh of electricity in 2024, roughly 1.5% of global electricity, and projects that to roughly double by 2030, with AI the main driver. Efficiency doesn't automatically cut totals, since cheaper AI tends to get used more, but it's the difference between that growth being manageable or not.

The rebound effect

Economists call it Jevons' paradox: when something becomes cheaper and more efficient, people often use more of it. AI per query has become far more efficient since 2023, yet total demand keeps rising. Green AI therefore needs both efficiency and thoughtful use, such as not reaching for a huge model or video generation when a short text answer will do.

Green AI vs greenwashing

Be sceptical of:

  • “Carbon neutral AI” claims with no published data.
  • Offsets presented as cancelling emissions today.
  • Vague rules (“we plant trees!”) with no count, partner or receipts.

Look for published numbers, a methodology and a clear rule anyone can check. Our tree-planting checklist applies here too.

What you can do as a user

  • Use efficient models for everyday tasks and save reasoning modes for hard problems.
  • Start fresh chats rather than dragging very long contexts.
  • Be intentional with image and video generation.

Full list: how to reduce your AI carbon footprint.

How SproutGPT approaches it

SproutGPT applies green AI ideas to everyday chat. It runs efficient open-weight models (DeepSeek, gpt-oss and Qwen, including mixture-of-experts designs) via Vercel AI Gateway, keeps answers tight and recent context short, and makes Deep Think reasoning an opt-in Pro mode rather than the default. Usage is counted openly: typically 1 to 2 leaves per message, shown on every reply. Every 100 leaves earns a real tree from SproutGPT's revenue, published on the ledger. We don't claim this makes AI carbon neutral; read our impact method.

Try SproutGPT free.

FAQ

What does green AI mean?

Building and using AI with energy, carbon and water treated as core metrics, through efficient models, right-sizing, clean power and transparent reporting.

Who coined the term green AI?

It was popularised by the 2019 paper 'Green AI' by Schwartz, Dodge, Smith and Etzioni, which contrasted it with compute-hungry 'Red AI'.

Is green AI just greenwashing?

It can be used that way. Real green AI comes with published numbers and methods; be sceptical of 'carbon neutral' claims without data.

Keep reading

Start growing your first tree 🌱

Free every day, no sign-up. Your first answer is a few seconds away.