Many countries have national AI strategies that do not address domestic AI compute capacity, leaving an open question about whether they have what they need to achieve their goals. To remedy this, policymakers could systematically take stock of existing national compute capacity and review the needs of their AI ecosystem for now and the future.

At the OECD, countries are studying the best way for governments to assess AI infrastructure needs and develop AI compute plans. The primary consideration is AI compute availability and use. Effectiveness requires an understanding of the right people and policies to deploy, where innovation can help and who will need access to AI compute. AI compute systems will need to be resilient, with high security and sustainability standards. And strategies must ensure national sovereignty for long term goals.

As policymakers ensure that AI is part of the solution for global sustainability targets, they must minimise its harmful effects. To reduce adverse effects, countries should first devise the means to measure the environmental impacts across the AI system life cycle, from production to end-of-life. To be effective, governments will need a holistic view of AI’s environmental footprint beyond energy and carbon dioxide emissions to include factors such as critical mineral extraction and freshwater use.

AI underpins some of the most promising solutions to today’s global environmental challenges. In fact, AI innovations enable sustainability in nearly every sector of the economy: buildings and cities, heavy manufacturing like shipbuilding and steel production, farming and forestry, green financing and more. Applications like digital twins and Internet-of-Things are increasingly applied to climate predictions, forecasts, and environmental modelling to help meet net-zero policy objectives.