Our Environmental Approach

AI has an environmental footprint.

Running AI systems requires computing power, which uses electricity and can contribute to water consumption and carbon emissions. Fjord is built around AI, so we think it is important to understand that impact rather than ignore it.

Our approach is simple: reduce what we can, take meaningful action where we cannot yet eliminate impact, report honestly on what we know, and keep improving as better information and tools become available.

Measure
Reduce
Take Action
Report
Improve

Reduce

The most useful unit of computation is the one we do not need to use. Fjord works to reduce unnecessary AI usage through compact context, prompt caching, deterministic code where AI is unnecessary, and less compute-intensive systems where they can provide the same quality.

Take Action

Fjord allocates 1% of revenue processed through Stripe to Stripe Climate Commitments, supporting a portfolio of emerging permanent carbon-removal technologies. This is one concrete action—not a precise offset of Fjord’s estimated emissions—and it does not make Fjord carbon neutral.

Learn about Stripe Climate

Report

We track operational data such as AI requests, model names, and token usage. We use published research and infrastructure assumptions to estimate electricity consumption, water consumption, and carbon emissions, clearly treating those results as estimates rather than direct measurements.

Improve

Environmental accounting for AI is still evolving. As better provider-specific data, research, and mitigation options become available, Fjord intends to improve both how it estimates environmental impact and what it does about it.

Carbon Is Only Part of the Picture

What AI infrastructure can consume

Stripe Climate supports carbon removal. It does not address or offset Fjord’s electricity or water consumption, which we consider separately.

Electricity

AI inference runs on computing infrastructure that consumes electricity. The amount varies with models, hardware, data-centre conditions, and how requests are processed.

Water

Water can be used directly for data-centre cooling and indirectly in electricity generation. Conditions differ substantially by facility and region.

Carbon emissions

Electricity use can produce greenhouse-gas emissions depending on the energy source. Hardware and infrastructure also have impacts beyond operational electricity.

How We Think About Environmental Impact

Fjord directly measures operational data such as model and token usage. Electricity, water, and carbon figures are modeled estimates based on published research and configurable assumptions—not measurements from the data centres serving every request.

Actual results can vary with the model, hardware, request processing, electricity source, climate, and facility. We acknowledge this uncertainty internally and plan to update the methodology as better evidence becomes available.

Current methodology version: 2026.1 · Last reviewed 2026-08-13

Where We May Improve Next

These are directions under consideration, not commitments to programs that do not yet exist.

  • More accurate environmental measurement
  • Further reductions in compute per learning session
  • Evaluation of water-replenishment initiatives
  • Evaluation of support for additional clean energy
  • Increased or more targeted carbon-removal commitments

We expect our approach to evolve.

As Fjord grows, we want to keep reducing unnecessary resource use, improve the quality of our environmental estimates, and direct more of the business toward meaningful environmental action. We’ll update this page as our approach develops.