What UK Government AI Spending Reveals - And What It Misses

A recent Sky News report about UK government AI expenditure tells a revealing story, not just about where public money is being allocated, but about a fundamental misalignment between ambition and execution. The £3.35 billion spent on AI contracts since 2018, while substantial, masks a more troubling reality: the departments with the greatest potential to transform citizens' lives through AI remain conspicuously underfunded, whilst infrastructure projects dominate the budget.

Following the Money: Infrastructure Over Impact

The spending breakdown is striking. The Met Office's £1.2 billion supercomputer contract with Microsoft and Transport for London's agreement with Init account for the lion's share of government AI investment. These are worthy investments in computational infrastructure and operational systems. But here's what should concern us: the Department for Work and Pensions, which processes benefits for millions of vulnerable citizens and has an annual IT budget exceeding £1 billion, has spent less than £100 million on AI cumulatively since 2018. The Treasury, overseeing the nation's tax system, similarly features in the bottom three of departmental AI spending.

This pattern suggests the government has prioritised AI for computational and operational infrastructure over AI for human-centric service transformation. It reveals a fundamental misunderstanding of where AI can deliver the greatest value, not just in processing speed and technical capability, but in improving outcomes for real people facing real problems.

The Preventative Opportunity We're Missing

At Datnexa, our work developing Hey Geraldine, an AI assistant supporting social care and health professionals across Peterborough City Council, has shown us what's possible when AI is deployed close to the frontline. Built using the knowledge of experienced occupational therapists and offering 24/7 support in over 50 languages, this tool doesn't just automate processes; it augments human expertise to deliver better care outcomes.

The contrast with government spending priorities is instructive. Whilst the NHS is beginning to pilot AI for predictive care, identifying patients at risk of frequent A&E attendance or predicting falls with 97% accuracy, these initiatives remain small-scale and fragmented. The transformative potential of preventative AI in healthcare and social care is widely acknowledged, with 80% of non-communicable diseases thought to be preventable and obesity alone costing the NHS £6 billion annually. Yet the infrastructure for scaling these solutions across government remains inadequate.

Our work on the National Frailty Index, which has analysed data for over 3.1 million households to identify individuals at highest risk of falling within the next two years, demonstrates the strategic value of predictive AI. This is not about replacing human judgement, it's about directing human expertise where it's needed most, before crises occur. The economic argument is compelling: prevention is invariably cheaper than intervention. The human argument is even stronger.

The Foundation Problem: Legacy Systems as the Silent Blocker

The Sky News article's observation that "up to 60% of some bits of government are running on legacy, older versions, of IT" understates the severity of this challenge. Recent government assessments reveal that 28% of central government IT systems are now classified as legacy technology, up from 26% in 2023. More concerning still, 15% of surveyed organisations cannot even quantify their legacy estate.

This isn't merely a technical debt problem; it's a strategic blocker to the government's AI ambitions. Legacy systems limit integration capabilities, creating barriers to the data sharing and interoperability that effective AI requires. The £45 billion in potential productivity savings identified by government reviews cannot be realised whilst these foundations remain fragmented and outdated.

As Palantir's UK boss noted when discussing the government's AI journey, "there is a lot of fear that tomorrow I'm going to have to do a different thing to what I was doing yesterday". This cultural resistance to change is amplified when the underlying systems make innovation appear risky and complex. Without addressing legacy infrastructure, AI adoption will remain piecemeal and inefficient, pilots that never scale, proof-of-concepts that never become production systems.

The Procurement Paradox: Innovation at the Edges

The spending data reveals another troubling pattern. Alphabet, one of the world's leading AI investors, holds just two contracts with the UK government via the Cabinet Office and Ministry of Justice. Meanwhile, SME participation in public sector procurement remains stubbornly low, only 20% of procurement spend went directly to SMEs in 2024, with the median spend per SME at just £31,000.

This matters because innovation often happens at the edges, not in the centre. Small, agile companies like Datnexa can move faster than large suppliers, with deeper domain expertise in specific problem areas. We work iteratively, testing and refining solutions in close partnership with frontline users. The Procurement Innovation Hub and new Procurement Act 2023 aim to address these barriers through problem-led procurement, but cultural change in procurement practices lags behind policy ambition. 

The government's focus on large-scale contracts with established technology giants may deliver computational infrastructure, but it risks missing the human-centred innovation that emerges from organisations embedded in public sector challenges. When we developed Hey Geraldine with Peterborough City Council, we didn't just build a tool, we co-created a solution with occupational therapists, iterating based on their feedback, ensuring the AI augmented rather than replaced their expertise.

Where Government Should Focus: Three Priorities

1. Invest in Preventative, Not Just Operational, AI

Government AI spending should shift towards preventative applications in health, social care, and welfare systems. The ROI of preventing falls, reducing emergency admissions, or identifying vulnerable children before crises escalate far exceeds the cost of faster weather forecasting. The NHS AI pilot identifying frequent A&E attenders demonstrates this principle, but such initiatives need systemic funding and scaling support.

AI should be deployed where it can improve lives before problems become acute. This requires data infrastructure that connects services end-to-end, enabling AI to spot patterns across health, social care, housing, and benefits systems. The government's "once only" rule for data sharing is a step forward, but implementation requires investment in the data foundations that legacy systems currently prevent.

2. Fix the Foundations Before Scaling AI

The £3.35 billion spent on AI contracts will deliver limited value whilst 28% of government IT remains legacy technology. The government should prioritise "legacy remediation and risk reduction" as part of the forthcoming Spending Review, creating tailored funding models that allow departments to modernise infrastructure alongside AI adoption.

This doesn't mean wholesale replacement of every system, that's neither feasible nor sensible. It means strategic modernisation focused on data interoperability, API-first architectures, and cloud migration where appropriate. The DWP's annual IT budget of over £1 billion should be substantially allocated to modernising platforms that enable, rather than obstruct, AI integration.

3. Embrace SME Innovation Through Problem-Led Procurement

The government should leverage the Procurement Innovation Hub to actively seek SME-led solutions for specific, well-defined problems. Rather than specifying technology requirements, departments should articulate outcomes they want to achieve, reduced A&E admissions, faster benefit processing, earlier identification of safeguarding risks, and invite innovative solutions.

This approach has worked internationally. As noted in recent discussions, Ukraine's emergency procurement processes allowed rapid mobilisation of digital capacity. The UK's traditional procurement culture is slower and more risk-averse, but the new Procurement Act provides powers for competitive, flexible procedures. Departments must now use them.

SMEs bring not just agility but deep domain expertise. Our work at Datnexa across government regulation, healthcare, professional services, and charity sectors has shown us that effective AI requires understanding the specific context, culture, and constraints of each setting. Generic platforms rarely deliver the nuanced solutions frontline professionals need.

Human-Centred AI at Scale

The government's AI Opportunities Action Plan sets ambitious goals, £14 billion in private investment, 13,250 new AI jobs, and productivity increases of up to 1.5 percentage points per year. The Blueprint for Modern Digital Government commits to harnessing AI "for the public good". These are the right aspirations.

But aspirations require execution. The spending data reveals a government investing in computational infrastructure whilst the departments with greatest citizen impact remain digitally underfunded. At Datnexa, we believe the focus should shift to three imperatives:

Preventative AI that improves lives before crises occur - deploying predictive tools in health, social care, and welfare to identify and support vulnerable people earlier.

Legacy modernisation as an enabler, not an obstacle - systematic investment in data foundations and interoperability that allow AI to work across fragmented systems.

SME-led innovation through problem-focused procurement - leveraging the agility and domain expertise of smaller providers to deliver human-centred solutions that frontline professionals will actually use.

The £3.35 billion spent on AI is substantial, but it's being allocated to the wrong priorities. We don't need more supercomputers. We need AI that helps occupational therapists support elderly citizens, that helps social workers identify at-risk children, that helps benefits processors serve vulnerable claimants more effectively. We need AI that augments human expertise, not just computational power.

The technology exists. The methodology exists, we've proven it works. What's missing is the will to invest where it matters most: at the intersection of data, frontline expertise, and human need. That's where transformation happens. That's where AI can truly serve the public good.