Conscious compute

AI production runs on real infrastructure. At Ford Media Lab, we estimate our usage each month and use those estimates to guide support for carbon removal, water restoration and biodiversity research. We contribute to Climeworks, the Bonneville Environmental Foundation and the Earth Species Project. Conscious Compute makes that effort part of our practice, without claiming that every asset is net-positive or that AI’s environmental costs are solved.

The environmental cost belongs in the production conversation.

What is conscious compute?

Conscious Compute is our self-directed approach to considering the environmental footprint of AI production and supporting work that addresses it.

Each month, we estimate our AI production usage and set contribution targets based on those estimates. Our aim is to contribute beyond the impacts we estimate. Those targets guide our purchases; they are not measured results or a third-party audit of our footprint. This is an ongoing effort with limits we need to name clearly.

What does AI content actually cost?

The tools behind AI content use energy and water. A production’s impact depends on the models, infrastructure and workload involved. Our estimates help us act, but they are not a complete lifecycle assessment of every asset. We cannot turn uncertainty about that footprint into a guarantee that a client’s content has no environmental cost.

Where our support goes

Carbon

We support carbon removal through Climeworks. A purchase supports that work; it is distinct from evidence of removal already delivered. Delivery records establish the amount, method and timing of completed removal. Provider verification applies to the activity covered by those records, not to our own estimate of a production’s footprint.

Water

Through the Bonneville Environmental Foundation, we purchase Water Restoration Certificates to support freshwater restoration. BEF describes its certificates as independently verified, tracked and registered. That verification concerns the restoration program. It does not independently establish how much water our AI production uses or prove that an individual asset is water-positive.

Nature

Through the Earth Species Project, we support open-source research using AI to understand animal communication. This is a contribution to research intended to benefit biodiversity, rather than a quantified offset for our production. The natural world that inspires our work deserves support of its own.

What does verification cover?

Our monthly estimates, purchases and provider-reported outcomes are different kinds of evidence. A payment record documents a purchase or contribution. A delivery record or certificate documents the provider’s stated activity and scope. Neither independently verifies FML’s estimation method or the full footprint of a client’s content.

We describe this initiative as support guided by internal estimates. We do not present our targets as achieved removal or restoration, and we do not promise that each asset returns more to the environment than it takes.

The distinction matters: supporting a solution is meaningful without overstating what it proves.

Why we keep doing it

Uncertainty is a reason to be precise and keep working. We want environmental questions to stay visible alongside questions of quality, authorship and production. Conscious Compute gives that intent a recurring place in how we run the studio, while leaving room to improve the estimates and learn from the organizations we support.

Fascination is the outlier. Responsibility is part of the work.