AI for Public Sector: Self-Hosted

UK public sector AI procurement is increasingly residency-bound: G-Cloud and Crown Commercial Service frameworks specify UK / EU residency for many sensitive data classes. Self-hosted on UK dedicated GPU aligns cleanly with this.

Workloads

• NHS clinical decision support: triage, summary, KB Q&A (with GMC / RCN safeguards)
• Council / local government: citizen Q&A, application processing, document drafting
• Central government: policy analysis, internal document Q&A, FOI request handling
• Education (state-funded): tutoring, curriculum content, marking
• Justice / police: case summary, evidence search (with strict access controls)

Procurement

• G-Cloud framework: hosting providers must be listed; check supplier match
• Crown Commercial Service: AI tools procurement increasingly through CCS frameworks
• Cyber Essentials Plus: required for handling sensitive public-sector data
• NHS DSPT: required for NHS data handling
• Open standards: government policy preference for open-weight models + standard APIs

Residency

UK residency is increasingly mandatory for public-sector AI:

• National Data Strategy — UK data should generally remain UK
• NHS Digital guidance — UK preferred for clinical / patient data
• Schrems II concerns — US transfers add SCC complexity
• Sovereignty discourse — AI models trained on UK data, hosted in UK, controlled by UK organisations

Verdict

For UK public sector AI, self-hosted on UK dedicated GPU is the dominant path. The procurement framework alignment + residency + sovereignty all favour it over hosted-API alternatives. Pair with comprehensive audit logging + open-standard APIs + open-weight models for maximum framework compatibility.