No community access tier for public compute and open-weight AI

AIRR/Isambard-AI allocate compute via structured calls favouring researchers, startups and the public sector; the £500m Sovereign AI Unit (opened April 2026) makes venture investments. No programme funds charities, community organisations or local newsrooms to fine-tune or run open-weight models for local services (advice triage, translation, casework), and no public inference endpoint exists for accredited civil-society organisations. Running Llama/Mistral-class models locally is technically feasible on consumer hardware today, but the skills, hosting and support layer for communities is absent; Nesta and CAST offer advice, not infrastructure.

Without a community tier, 'sovereign AI' means sovereign for the state and investors only. Civil society will rent US frontier APIs instead, sending sensitive community data offshore and forfeiting the privacy advantage that local open-weight inference uniquely offers.

A 'community compute' allocation within AIRR plus a grant programme providing hosted open-weight inference, fine-tuning support and deployment assistance to accredited civil-society organisations.

// Build together: Counterparty: DSIT/UKRI for an AIRR community allocation; open-weight models and hosting are ready today.

Sensitive community data drifts offshore without a public inference tier and a grant instrument is straightforward, but the stakes are nascent and no deadline forces action.