Skip to content

Hybrid AI: scaling starts with knowing where it runs

11 November 20265 min read
Guest Insights
Glowing digital cloud illustration with interconnected nodes representing cloud computing infrastructure

The Chancellor has put AI at the heart of the Budget, with more than £600 million to scale British AI firms, strengthen AI security and spread the benefits across the country. Scaling, though, rests on something the headlines rarely mention: whether we can trust the devices and supply chains beneath the models. Cost, value and where AI runs all matter, but trust is essential.

The device is the new front door 

AI security discussions tend to focus on models and applications. Like the Trojan Horse, a compromised PC or workstation can bring risk inside an organisation before an AI tool is ever used. Every endpoint has a journey through sourcing, manufacture, transport and configuration, and as devices become gateways to valuable AI workloads and intellectual property, that history matters. 

The stakes are rising: the National Cyber Security Centre has warned that the UK should be braced for an AI-powered ransomware attack on its critical national infrastructure by 2028, as AI lowers the barrier for attackers and exposes years of underinvestment in basic cyber protections. Before investing, buyers should ask partners: 

  • Can provenance be verified? 

  • Are manufacturing and chain of custody secure? 

  • Is hardware protected through the life cycle? 

  • Is there evidence, not just assurance? 

Counting the real cost, and the real return 

Early pilots look affordable because paying by query, token or transaction makes experimentation accessible, but those economics change as usage spreads. Finance and IT need a shared baseline of tools, subscriptions and departmental usage, set against outcomes rather than prompt counts: hours saved, faster sales conversion, reduced compliance risk. The aim is not to cut spending, but to know whether it earns a worthwhile return. 

It seems businesses are hitting an inflection pointy where AI is becoming material to the cost structure. Indeed, KPMG's Q2 2026 Global AI Pulse, a survey of 2,145 senior leaders, found 49% had scaled back AI agent deployments because costs outweighed benefits.

And PwC is pointing out a paradox: Token costs are dropping, yet AI bills are surging. The reason? Seeming “cheapness” is making volume explode.

Matching the task to the machine 

Cloud models remain valuable for complex reasoning across extensive enterprise datasets, but routine tasks, including some involving sensitive information, can run on the device itself. Processing closer to users reduces latency, supports continuity when connectivity fails and limits unnecessary transfers of sensitive data. Combining capable data centres with secure Edge AI on PCs and workstations gives the UK a more distributed, resilient foundation. 

Running suitable tasks locally could provide additional benefits: 

  • Privacy and security: sensitive information can stay under the organisation's control. 

  • Cost: less reliance on subscriptions and tokens for routine tasks. 

  • Environment: less data centre energy and water use. 

  • Resilience: work continues when connectivity or cloud access is disrupted. 

Rules that keep pace with adoption 

Clear rules should name approved tools, accountable owners and budgets, and which workloads may use external models. Finance, IT, security and legal teams should agree safeguards for confidential data and review costs and outcomes regularly. Local processing still needs strong security and oversight. 

Three steps for government and industry 

Organisations cannot manage this alone. Three steps would help the UK scale AI with confidence: 

  • Procurement should reward verifiable provenance and evidence of secure supply chains, not just price and performance. 

  • Government and regulators should give clear guidance on which workloads suit local processing and which can use external cloud models. 

  • Industry should be transparent about security across the device lifecycle, and about the costs and outcomes of AI deployments. 

Secure AI on the Edge 

Hybrid AI is therefore a management discipline, not just an architecture. Government and industry should judge progress by measurable value and resilient deployment. Sustainable AI growth depends on choosing the right environment for each task and verifying the technology entrusted to deliver it. The Budget has added funding to the ambition. Whether it becomes lasting growth will depend on something less visible: confidence that every device carrying AI into the workplace can be trusted from factory to desk.