UK AI Growth Zones: compute needs a local adoption system

The policy test is diffusion, not designation

AI Growth Zones (AIGZs) are now a material part of the UK’s AI and growth strategy. By January 2026, government had designated five zones across Great Britain, reporting £28.2 billion of associated investment, more than 15,000 jobs and up to £5 million per zone for local AI-adoption activity. The programme is intended to coordinate the difficult inputs required for advanced compute—power, land, planning, connectivity, capital and technical capability—and, in England, enables eligible business-rate growth to be retained locally over the long term (Department for Science, Innovation and Technology, 2026) ↗ (Department for Science, Innovation and Technology, 2025) ↗.

There is a credible national strategic case for this concentration of effort. Compute infrastructure is capital-intensive, energy-dependent and slow to deliver; fragmented planning and uncertain grid access can deter investment. The UK Compute Roadmap also frames AIGZs as more than data-centre sites, linking them to applied research, innovation, adoption, collaboration and workforce development (Department for Science, Innovation and Technology, 2026) ↗.

But this breadth creates an accountability problem. A zone can succeed at enabling nationally valuable infrastructure while producing limited change in its immediate economy. A secured grid connection, a construction programme and an investment announcement are all meaningful milestones. They do not show that local firms have improved a workflow, that public services have delivered better outcomes, or that residents have gained durable access to better work.

Sanctuary’s judgement is that every AIGZ should therefore be assessed through two distinct lenses. The first is national strategic value: resilient compute capacity, investment mobilisation, research capability and infrastructure delivery. The second is local diffusion: whether organisations in and around the zone repeatedly adopt, retain and improve useful applications. These objectives can reinforce one another, but they are not interchangeable. Treating infrastructure inputs as evidence of local productivity risks overstating what the programme has achieved.

Access is not the same as organisational adoption

The latest business evidence makes the gap between availability and transformation clearer. In June 2026, 35% of UK businesses with 10 or more employees reported using at least one AI technology, compared with around 12% in late 2023. Yet adopting firms used 1.6 AI technologies on average, up from 1.4, and only 10% reported extensive use (Office for National Statistics, 2026) ↗.

These are early indicators, and they should not be used to imply that every firm needs a wide portfolio of AI tools. A narrowly deployed application can generate worthwhile value. Nevertheless, the pattern cautions against equating access to compute, experimentation with software or a first use case with organisational transformation. Reported business use has been directed principally towards improving existing operations rather than developing new products, services or markets (Office for National Statistics, 2026) ↗. That may be the rational starting point for many smaller firms, but it also means that the local-growth dividend will depend on implementation quality rather than technology presence alone.

Management capability is one important complement. ONS analysis found that firms with stronger management practices were more likely to adopt advanced technologies and more likely to act on intentions to adopt AI. This does not establish that management quality alone causes adoption: better-managed firms may also have more capital, skills or clearer commercial opportunities. It does, however, support the practical conclusion that leadership, process design and workforce engagement are part of the adoption mechanism, not peripheral considerations (Office for National Statistics, 2025) ↗.

For an SME, the immediate issue is usually not whether it is ‘doing AI’. It is whether it can improve one defined workflow without compromising accuracy, accountability, confidentiality or customer trust. That may require better operational data, staff involvement, a human-review process, vendor assessment, data-governance decisions and a way to judge whether the intervention worked. These are implementation challenges before they are questions of model choice. They are also reasons why awareness campaigns, one-off demonstrations and undifferentiated training offers are unlikely to produce sustained local adoption on their own.

How infrastructure value can fail to become local value

Three mechanisms explain why an infrastructure-led programme may generate strategic national value without enough local economic diffusion.

First, the labour markets are different. Construction, data-centre operation and applied-AI adoption require overlapping but distinct capabilities. A facility can create construction work and specialist technical roles without automatically creating broad routes into AI-enabled employment for local residents. A place needs deliberate bridges: employer-linked training, local supplier-development support, progression from foundational digital skills into data, cloud, cybersecurity and facilities roles, and measurement of job entry and progression rather than course completions alone.

Second, the organisations that may benefit most from implementation help are least likely to be reached by infrastructure investment by itself. Research for the Department for Business and Trade finds that SMEs value reliable, personalised support throughout their technology-adoption journey, rather than information about available technologies alone (Department for Business and Trade, 2025) ↗. A grant, showcase or short course may create interest, but a smaller business may still need help to select a use case, assess a supplier, involve staff, set data boundaries, test outputs and decide whether to stop, revise or scale. This matters for high streets and local supply chains as much as for technology businesses: the potential gains may be operationally modest at firm level but cumulatively significant if they are retained and replicated.

Third, local legitimacy is an operating condition, not a communications exercise. Data-centre development raises place-specific questions about energy, water, emissions, land use and pressure on local infrastructure. The Local Government Association has called for transparent information on these matters and highlighted the specialist-capacity pressures councils face in determining complex digital-infrastructure proposals (Local Government Association, 2026) ↗. Better disclosure will not eliminate difficult trade-offs, but it gives planning authorities and communities a basis for credible scrutiny.

There is a legitimate counterargument. Requiring an extensive local programme before investment is firm could delay nationally important infrastructure. The answer is sequencing, not a veto. A proportionate local-benefits model should be established while planning, design and grid decisions are made, then deepen as the investment pipeline becomes more certain. Leaving adoption, skills and local-value work until facilities are operational would make them an underpowered add-on.

A diffusion compact should sit beside every infrastructure plan

Sanctuary recommends that each AIGZ publish a local diffusion compact alongside its infrastructure delivery plan. This should be a concise, accountable agreement between local government, mayoral bodies where relevant, employers, education providers, anchor institutions, investors and community representatives—not another brand or an uncosted list of social-value ambitions.

First, the compact should identify a limited set of demand-led use cases. Rather than promise economy-wide transformation, a zone should select three to five local sectors or public-service problems with repeatable workflows, a credible service owner, manageable risk and a plausible route to measurable value. Maintenance planning, energy management, logistics administration, export documentation, adult social-care administration and customer-service workflows are illustrative possibilities, not a prescribed list. The selection criterion should be user need and implementation feasibility, not technical novelty.

Second, it should provide an SME implementation service rather than only a training catalogue. A practical offer could combine a diagnostic, workflow mapping, data-boundary templates, vendor-neutral procurement support, staff-engagement guidance and a post-pilot review. The objective is to help firms make one narrow application dependable before encouraging expansion. This follows the evidence that smaller firms value tailored support over the full adoption journey (Department for Business and Trade, 2025) ↗.

Third, public-service testbeds should create useful demand while remaining tightly governed. Councils, NHS partners and other anchor institutions can expose local suppliers to real operational problems, but testbeds should not become sites of uncontrolled experimentation. Each should have a named service owner, information-governance oversight, user involvement, defined outcome measures and a decision point to stop, improve or scale.

Fourth, skills policy should be treated as a progression system. Zones should track movement into paid work, supervised application of skills on the job, task redesign and employer retention of capability after external support ends. The North East’s published proposition is notable for bringing infrastructure, innovation, skills and adoption together, rather than presenting them as separate agendas (North East Mayoral Strategic Authority, 2026) ↗. The substantive test will be whether that integration becomes a repeatable delivery model with visible outcomes.

Finally, the compact needs a public measurement framework. Delivery bodies should distinguish inputs—funding allocated, firms reached, construction activity and training places—from outcomes: deployments still active after six and 12 months, workers moving into AI-enabled roles, local supplier spend, relevant public-service performance measures, and energy, water and emissions disclosures where applicable. No dashboard can prove that every observed change was caused by an AIGZ. But baselines, clear definitions and follow-up measures would provide a much stronger account of contribution than announcements alone.

The next six months should establish the local delivery conditions

The programme is too new for claims of broad local productivity effects to be credible. Its strongest claims should remain conditional. AIGZs may become nationally significant infrastructure and innovation assets; whether they become engines of inclusive local growth will depend on the connection between compute, management capability, skills, procurement, public-service demand and local accountability.

The immediate priority is to make that connection operational before headline investment totals define the programme’s public story. Local and mayoral leaders should identify priority workflows and employer demand before commissioning large-scale skills provision. Anchor institutions should identify common barriers in procurement, assurance and data governance that prevent smaller suppliers from participating. Developers and investors should provide sufficiently granular, verifiable information for planning and public engagement. And programme leaders should report infrastructure milestones, capital commitments and realised local outcomes separately.

This does not require every zone to adopt an identical model. Their industrial structures, institutional capacity and local priorities will differ. It does require a common discipline: infrastructure supply must be linked to mechanisms that help local organisations use technology repeatedly, learn from implementation and reinvest in capability. Compute capacity is a necessary strategic asset. It is not, on its own, a local growth strategy.