Good Shepherd International Foundation

for their work combating child labour in cobalt-mining communities through education, healthcare, livelihoods and community support. The hardware behind every AI system, including ours, most green energy initiatives depends on cobalt those communities are exploited to extract.

Water Witness

for their work improving water security, protecting water resources and supporting communities affected by pollution, scarcity and climate change. Data centres can consume significant quantities of water for cooling, and the communities nearest to raw resource extraction sites are often those with the least access to clean water.

Artists Futures Fund

for providing fellowships, bursaries, mentorship and career support to emerging artists facing economic and social barriers. Generative AI was trained on the work of artists who were never asked and never paid. We cannot undo that, but we can support the people it affects, by helping fund opportunities to foster human creativity and creation.

Client Earth

for using the law to protect the environment and hold governments and corporations accountable for pollution, environmental damage and greenwashing. Which is crucial working in fighting the impact of the climate crisis and giving incentives for businesses to conduct themselves in a way that puts the planet first.

Raid

for investigating corporate abuses in the extractive industries and supporting affected workers and communities in seeking accountability and remediation. A more fair and sustainable supply chain can go a long way towards reducing everyone’s impact from the usage of new technologies.

Generation UK

for providing intensive training, mentoring and employment support to people facing barriers to work, helping them transition into careers in technology, sustainability and healthcare. It is an initiative more important than ever because every new tool like AI has displaced jobs, and created new ones that require specialized training.

Tier 1. What your provider discloses.

We document what your current AI provider publishes about energy use, carbon emissions and water consumption, and what they do not or chose to omit. As of August 2026, only one major provider has published a detailed per-query energy methodology. One other has offered a single figure without supporting data for a model released in 2023. The rest have published nothing.

Where the figures do exist, the variance between a short query and a long one can exceed 100x, so a single "per-query" number tells you very little about what any specific workload actually costs. For any business trying to understand the environmental impact of the AI it already uses, this is an ongoing issue.

Tier 2. Right-sizing.

Not every task needs the most powerful model. A flagship model like GPT-5.6 Sol or Claude Fable 5 can use 20 to 100 times more energy per request than a smaller one like GPT-5.6 Luna or Claude Haiku. Most business AI work, drafting, summarising, classifying, extracting, does not require a flagship model. When every request goes through the most expensive model regardless, the waste is both financial and environmental at the same time. Matching the model to the task is the single largest lever most businesses have for reducing their AI energy use. With locally hosted open-weight models that control goes further, because the model, its configuration, and the hardware it runs on are all known, chosen, and measurable rather than decided for you by a provider who chooses not to disclose that information.

Tier 3. What our impact is.

Because we control the model, the hardware, and the data centre, we can measure what a workload actually costs in energy and carbon rather than estimating it. Energy per task and CO₂ per task, measured on the hardware we run, at the carbon intensity of the facility we host in. This is possible because we know the variables. Most cloud providers cannot give you the same figure because they refuse to publish the data. For clients considering self-hosting, the same calculation applies to their hardware, their power source, and their local grid. Local is not automatically greener than cloud. We commit to honestly reporting everything we measure, and let the numbers decide our stance on the environmental impact.