This OECD Policy Brief, "Securing AI compute: government policy approaches" (October 2026), examines how governments are securing access to AI compute as a strategic asset, outlining the policy challenge of uneven access, rising costs and the lack of standardised needs assessments, and aims to provide a comparative framework and actionable guidance grounded in the OECD AI Principles and an AI stack lens, with a call to develop national AI compute plans, commission needs assessments and strengthen international knowledge sharing.
Key Insights
Governments are directly securing AI compute for research and innovation through public provision and subsidised access, including Canada’s AI Sovereign Compute Infrastructure (SCIP) aiming to invest about CAD 890 million beginning in 2026–2027, the United Kingdom’s AI Research Resource offering up to 10 000 GPU hours to eligible users, and the European Union’s AI in Science Strategy allocating EUR 600 million and leveraging EuroHPC JU’s AI Factories.
Large-scale capital mobilisation and procurement are accelerating domestic capacity, with Saudi Arabia’s HUMAIN backed by over USD 40 billion in AI ventures, the United States’ Stargate project committing USD 100 billion initially and planning to scale to USD 500 billion by 2029 while announcing five sites totalling nearly 7 gigawatts and over USD 400 billion in planned investment by September 2025, and public procurements such as IndiaAI’s deployment of more than 18 000 GPUs and Korea’s purchase of 13 000 GPUs under a KRW 1.46 trillion initiative.
As shown in the charts on page 8, analysis of 46 GPAI member countries and the European Union finds 44 have adopted AI strategies, with 21 treating compute as a dedicated pillar, 10 as explicit commitments and 13 as generic references, while Canada (December 2024) and the United Kingdom (July 2025) are the first to adopt dedicated national AI compute strategies and 72% of strategies updated in 2025–2026 elevate compute to a Tier 1 pillar.
Regional and programmatic approaches are expanding shared compute through EuroHPC JU’s network of 19 AI Factories across 16 EU countries plus 13 AI Factory Antennas, alongside the EU’s InvestAI initiative to mobilise EUR 200 billion, including a EUR 20 billion fund for three to five AI Gigafactories and plans to triple EU data centre capacity within five to seven years.
The brief’s call to action urges policymakers to ground investments in national objectives, commission domestic AI compute needs assessments, and pursue dialogue and voluntary knowledge sharing with partners at different development levels, with OECD and GPAI offering analytical frameworks, data and tools such as the OECD.AI Policy Navigator to support targeted national AI compute plans.
