Sovereign AI
The capacity to run AI systems on infrastructure you control — independent of third-party cloud providers, proprietary model vendors, or foreign jurisdictions.
Overview
Sovereign AI is AI capability that cannot be switched off, degraded or monitored by parties outside your control. For nations it is strategic autonomy; for organisations it is operational independence; for regulated sectors it is a compliance requirement.
Sovereign AI is the capacity to operate AI systems on infrastructure under your own control — within your own jurisdiction, on your own hardware, with your own data, without dependence on third-party cloud providers or model vendors. The concept operates at multiple scales simultaneously: nations seeking strategic autonomy over AI capabilities; organisations in regulated sectors that cannot route sensitive data through external systems; and individuals and communities asserting that the inferences made about them should not be processed by entities they did not choose and cannot monitor.
The national dimension gained profile when Jensen Huang of NVIDIA, speaking at the World Government Summit in 2024, articulated the case for every country owning its own AI infrastructure, training data reflecting its culture and values, and AI systems running on domestically controlled compute. The argument mirrors the logic of energy sovereignty: strategic dependence on foreign-controlled critical infrastructure creates vulnerability. A country whose AI capabilities depend on foreign cloud services and foreign model providers can have those capabilities degraded or revoked by geopolitical events outside its control. Several nations — France (with its Mistral investment), the UAE (with Falcon), Singapore, Japan — have responded with explicit sovereign AI investment programmes.
The deeper challenge is the concentration of the compute stack. AI inference and training at scale requires semiconductors manufactured primarily by TSMC in Taiwan, on equipment made by ASML in the Netherlands, designed by Nvidia and AMD in the United States. Kate Crawford's Atlas of AI (2021) traces the full supply chain — mines, labour, energy, data centres, chip fabrication — that subtends every AI inference. Tim Hwang's Artificial Unintelligence (2020) documented the infrastructure dependencies underlying AI capabilities and their brittleness. True compute sovereignty, at the national level, requires addressing this supply chain dependency — a decades-long challenge that no nation has yet solved.
For organisations deploying AI in regulated or sensitive contexts, sovereign AI means something more tractable: running inference on on-premises or private-cloud infrastructure, ensuring that data and model weights never leave the organisation's control, and maintaining the ability to audit and modify the systems in use. This is increasingly required by regulatory frameworks in healthcare, legal services, financial services, and government — and by data protection obligations that prohibit sending certain categories of data to external processors without adequate safeguards. For these organisations, sovereign AI is not a geopolitical preference but an operational requirement.
Key Texts
Foundational works in this research tradition.
The address that brought "sovereign AI" into mainstream policy discourse: every country should own AI infrastructure, data, and capabilities that reflect its own culture and values, rather than depending on foreign providers. Framed AI sovereignty as the analogue of energy sovereignty — strategic rather than merely commercial.
The material supply chain of AI: mining, labour, energy, and the semiconductor dependencies that make AI sovereignty a physical as well as a software challenge. Every AI inference depends on a specific geopolitical and material supply chain — sovereignty means controlling that chain, not just the software that runs on it.
Infrastructure dependencies in AI: how compute concentration and cloud dependency create strategic vulnerabilities for organisations and nations dependent on third-party AI providers. Documents the brittleness of AI capabilities that rest on infrastructure controlled by others.
The EU AI Act's high-risk category and transparency requirements create de facto sovereign AI requirements for regulated sectors: certain data cannot be processed by unvetted external providers; conformity documentation must be available to national authorities. Data localisation and sovereign deployment are governance requirements, not just preferences.
Digital sovereignty as a policy goal: what it means for nations to maintain strategic autonomy over digital infrastructure including AI. Reviews national approaches to digital sovereignty across OECD member states and identifies the policy levers available to governments seeking to build indigenous AI capability.
Related Research
Connected areas of inquiry.