Industrial Strategy names steel, chemicals, and grid equipment. The deep dive treats artificial intelligence as a capability multiplier for those sectors, not as a standalone growth story. This companion pulls that argument into one place and extends it to labour, the state, and the information environment.
Britain does not need another "AI strategy" PowerPoint. It needs rules for who controls the models, who owns the data, who gets displaced, and who profits when public systems adopt automation.
What AI is in this programme
AI is useful where it improves performance in systems the country already depends on:
- Grid and energy: demand forecasting, fault detection, balancing across a stressed network described in Energy.
- Industrial process: predictive maintenance in steel, chemicals, and grid factories, not chatbots on factory floors.
- Health: diagnostics support and operational planning, with clinical accountability retained, as Health & Social Care requires.
- Tax and enforcement: pattern detection for avoidance and fraud, where Civil Service capacity is the binding constraint.
- Defence: sensor fusion, logistics, and command support, not uncritical outsourcing of kill chains.
Treated as a vanity "tech sector," AI becomes a distraction. Treated as infrastructure with sovereignty conditions, it is part of rebuilding capability.
Sovereign compute and procurement
The UK will not outspend US frontier labs on general-purpose model training. That is not the strategic play.
The play is sovereign capacity for public and defence workloads, plus domain-specific models where Britain has data and institutional advantage: NHS pathways, tax records, grid telemetry, defence logistics.
Public contracts for high-value AI deployment should require, as standard:
- UK data residency for identifiable personal and operational data unless a named national-security exception applies.
- Model portability and open interfaces so departments are not locked into one hyperscaler forever.
- UK operational control of the deployment environment, including audit logs ministers can inspect.
- Human accountability for decisions that affect rights, benefits, housing, or liberty. No fully automated refusal of social security, no opaque planning veto.
Procurement preference for portable systems creates a domestic market without pretending the UK will build GPT-5 in a shed outside Swindon.
Labour displacement is a programme risk
Previous technology transitions displaced workers gradually while creating new job categories. AI adoption in administration, call centres, logistics planning, and routine professional work can displace faster and broader.
That matters to this series in three places:
- Skills & FE: retraining must be funded for mid-career workers, not only school leavers.
- Social Security: earnings disregards and training support must cover people moving out of automated roles.
- Industrial Strategy: automation that destroys mid-skill jobs without replacing them with skilled industrial work is not a win. Productivity gains that flow only to capital reproduce the 2008-to-2023 wage stagnation the Situation chapter describes.
A serious government does not ban useful automation. It pairs adoption with sector bargaining, retraining routes, and public investment in the jobs that remain geographically distributed.
The financial layer
Large AI infrastructure spending has created a financial expectation layer on top of real physical investment. Corporate borrowing against projected AI revenues is not the same thing as building a functioning grid or NHS.
The Fiscal Framework already treats strategic capital separately from annual programme lines. AI capex bubbles belong in the risk register: a sharp correction in tech valuations can tighten credit for unrelated public investment if ministers panic at the wrong moment.
This programme invests in sovereign public workloads and industrial AI applications. It does not bet the fiscal credibility of the state on venture narratives.
Algorithmic power is political power
Platform algorithms already organise protest, amplify false claims about migration and the NHS, and destabilise politics faster than institutions respond. That is covered in The Press and Media and Reform Counter-Narrative.
The industrial point is simpler: whoever controls ranking, targeting, and moderation controls political reality for millions of people. AI governance is therefore not a DCMS sideshow. It sits next to press reform and rapid response in the survival architecture of any progressive programme.
What this programme would do
- Public AI charter: binding standards for procurement, appeal rights, and prohibited fully automated decisions in benefits, housing, and justice-adjacent systems.
- Sovereign compute fund: capital for secure public-sector and defence AI environments, not general subsidies for US cloud bills.
- Interoperability mandate: cross-government standards so models and data pipelines can move between suppliers.
- Displacement protocol: when a public body or regulated contractor automates roles above a threshold, publish impact assessment, fund retraining, and report net job quality outcomes annually.
- Platform accountability linkage: Ofcom and Cabinet Office rapid response share intelligence on model-driven harm campaigns, not only content takedowns.
What this is not
This is not a call to pause all AI because screens are scary. It is not a call to hand NHS patient data to the lowest bidder with a slide deck about innovation.
It is the insistence that a country rebuilding industrial and state capacity must govern the tools that will run much of that capacity within ten years.
Parent chapter: Industrial Strategy. Related depth: Industrial Strategy: Deep Dive (AI as capability multiplier).