Can AI Adoption Succeed Without Digital Infrastructure?

The case for infrastructure investment over application development

The AI incubator has built dozens of working prototypes across government departments over the past year, including AI tutors for schools, chatbots for consultations, and automation for planning applications. Each addresses a department’s immediate problem, identifying practical applications of how departments will need to change to keep up with the wider progression of AI adoption in the private sector. These use cases are critical, as the civil service does need proof points of where new tools can change how government is delivered.

The government and civil service like to point towards AI tools, such as the ones designed by the incubator, as proof of AI adoption. But in the same way good software cannot improve bad hardware; good front-end tools are far less important than cross-departmental digital infrastructure. This is not a critique of the incubator’s work. It is an acknowledgement that no single programme, however well-executed, can substitute for the foundational transformation government requires. The incubator operates within structural constraints it cannot overcome – departmental silos, fragmented data architectures, procurement barriers that resist coordination.

If the government had a joined-up AI adoption strategy, it would solve its capacity and systems issues before generating tools – think of it as fixing the operating system before building apps.

So, when should the government build systems capacity, and in what form?

There is a persistent instinct among reformers, myself included, that building state capacity means building applications. That sovereign AI requires government to develop its own tools, cultivate its own technical expertise, and demonstrate its own delivery capability. This instinct is not wrong, but it may be incomplete.

When the state competes on applications, it distorts the market, making it unclear where startups should build, rather than enabling further innovation. While an AI incubator should exist, as a smaller experimental unit, it should complement the Government’s ambitions of digitalisation and AI-enabled delivery. The Government’s unique strength lies not in competing with markets on applications, but in its ability to convene, compel, and coordinate across constitutional boundaries that private actors cannot cross

It should clean, prepare and provide the flows for data to flow seamlessly between departments. If the government only builds tools within departments, it could further entrench siloed and disconnected public sector digitalisation. Simply developing proprietary AI tools (that could be replicable) individually in each department, when these tools already exist in replicable formats in the private sector, is not ambitious enough.

Markets are efficient at delivering applications once infrastructure exists. Startups will build AI tutors, planning assistants, and consultation tools - better ones, faster ones - if given the platform. What markets cannot do is compel cross-departmental digitalisation, mandate interoperability, or break down the constitutional and cultural barriers that prevent data from flowing where citizens need it.

There are likely some civil servants shouting at the screen of their computer, reading this, that engaging different departments is monumentally impossible for various technical and cultural reasons.

Firstly, on the cultural issue. Sovereignty and control over the fiefdom of each department is significant, with permanent secretaries’ control and protectionism over their own remit often giving little incentive for them to want to risk losing institutional knowledge and capacity if they do engage with a cross-departmental digitalisation drive.

Gov.uk provides an example of the successful delivery of a government digitalisation effort, which did succeed. Between 2010 and 2012, the Government Digital Service consolidated 1,882 organisational websites into GOV.UK¹, saving over £70 million annually in hosting costs². The real value came in creating a single publishing platform that hundreds of public bodies could use, eliminating the need for each to build and maintain separate infrastructure.

This required political will at the highest levels. Ministers working in the Cabinet Office will have spent significant political capital persuading fellow ministers that closing their departments’ websites served their interests. GDS was given joint control with HM Treasury over departmental IT spending and the power to stop new services that didn’t meet standards. These levers meant that the transformation which we take for granted today could take place.

This leads to the frustrating reality where experimental initiatives like the incubator build tools for willing departments, but while structural barriers remain intact. This is Phase 2 work, valuable once Phase 1 exists, premature without it.

The Government does have an ambition for what this phase 2 transformation looks like, but it remains vague. As seen in the Blueprint for Modern Digital Government, which outlines six critical priorities. In this, three directly require infrastructure transformation: joining up public sector services, strengthening digital and data public infrastructure, and reforming procurement. Yet, the Government does not have a true plan to link up data flows between departments with the ambition of creating seamless flows for AI applications to be built upon. As long as this is the case, AI adoption in departments will be disconnected from the transformation which is required. Departments can get chatbots (which they could have procured from the private sector), but then citizens still cannot get their hospital to communicate with DWP.

On the technical issues of delivering transformational AI and digital transformation. There are various models which we can look at how we can deliver this infrastructure. The first is the NHS Federated Data platform, delivered by Palantir. The platform, which offers standardised, interoperable infrastructure, which then means that startups can build on top of the FDP and then sell bespoke products to NHS trusts. The government builds the infrastructure, along with preparing the data layer, which then enables applications to be possible.

Estonia’s X-Road, developed in 2001 under an MIT license, provides secure, distributed data exchange across government services. Open source and interoperable, it has since been adopted by many different countries and by some corporations in Japan. Its open source infrastructure, in particular, allows building trust through developer use and reuse of the code to build their own projects on.

Sovereign AI in the context of informing a government’s strategy ought to be holistic and transformational. It means using state power to build the data infrastructure that becomes the operating system of modern government, federated architecture enabling cross-departmental service delivery, standardised APIs allowing private sector innovation, and identity systems giving citizens control over their data.

Government exists to do what only government can do. Building AI tools itself is not on that list. Building the data infrastructure that allows a thousand AI tools to flourish, including potentially better ones from startups, is. This is not an argument against the government building AI capability. It is an argument for deploying that capability where only the government can act. The state’s power to mandate coordination, enforce standards, and override departmental sovereignty represents its comparative advantage. Whether it also has a comparative advantage in building user-facing applications remains uncertain

The question is whether we will expend political capital on marginal improvements to isolated services, or on the foundational transformation that makes comprehensive improvement possible. What we need is a government willing to tackle the constitutional and cultural barriers that have resisted decades of reform.

1. https://www.gov.uk/government/publications/state-of-digital-government-review/state-of-digital-government-review

2. https://www.gov.uk/government/publications/state-of-digital-government-review/state-of-digital-government-review