AI Investment and Sovereignty: Rethinking What Counts

The UK needs to find some growth, you’ve probably heard. One of the government’s key hopes is that investment in AI, and the infrastructure that enables it, will act as a catalyst for improving living standards and a central pillar of our defence strategy. But current priorities - notably the focus on data centres and the surrounding AI Growth Zones - risk missing the mark. Two factors are distorting the debate.

The first is a concept of “sovereignty” inherited from a decade of privacy arguments. It overemphasises where computers are located and underplays who benefits from controlling them. Technology, harnessed well, offers economic returns in peace time and strategic leverage in moments of international competition. The ability to make decisions in these moments is a function of far more than where data is located, though existing arguments for sovereignty often overlook this. The second is a well-intentioned focus on Value for Money (VfM), which encourages public bodies to buy what they need at the cheapest available price, typically from established suppliers, few of them British. The result is a procurement environment that suppresses investment in domestic capability while doing little to build the long-term foundations for economic or strategic returns. A longstanding example is the UK’s ageing and costly energy infrastructure, which drives up the price of high-performance computing.

This tension is coming to a head. The Strategic Defence Review has already highlighted the urgent need to bolster the UK’s AI capabilities. The AI Opportunities Action Plan commits to publishing an infrastructure roadmap shortly. And the latest Spending Review includes billions of pounds earmarked for technology investment - much of it related to AI.

Over the coming months, government will have the chance to shape investments that will underpin both our economic resilience and national security for years to come. To do that wisely, ministers and officials will need to expand what counts as an “investment” and reframe what we mean by “returns.” That means finding ways to fund capacity-building projects, not just those that deliver immediate, measurable outputs. Happily, there are many policy tools to do this: advanced market commitments, subsidised access to compute for UK-based firms, Strategic Value Multipliers in procurement assessments, a preference for home-grown software even where it costs more, and support for the development and retention of skilled, security-cleared workers. All of these would represent a significant strategic shift. But if we want resilience, not just efficiency, we must be prepared to invest for the long term, even when the benefits are intangible or delayed.

Much of the current difficulty arises from a deceptively simple question: what counts as an investment?

Whitehall departments operate under two broad budget limits: Capital (CDEL) and Revenue (RDEL). Revenue budgets cover day-to-day operations and are often pre-committed to paying staff. Capital budgets, by contrast, offer more discretion, but must be used to create or acquire an asset that can be owned and pointed to. Business case approval is typically tied to that asset producing a measurable return on investment (ROI). The intention is sound: to avoid wasting money. But in practice, this system encourages investment only in things that benefit the spending department directly - not the wider UK economy or state.

The result is a structural blind spot. Externalities - like the growth of a domestic supplier ecosystem, or the upskilling of a British workforce - don’t show up in departmental balance sheets, even when they’re critical to national capability. Yet these are exactly the sorts of gains that matter if the UK is to exercise meaningful control over AI technologies in the coming decade.

If the UK were a company, perhaps it would make sense for budget holders to ignore the growth of intangible assets like skills or IP at other firms. But we are not a PLC. We are a country. And those intangible benefits are real. If we allow them to accrue here, they will reduce costs and improve outcomes across the public sector. If we don’t, they will accrue overseas; we will be dependent on foreign suppliers to build and maintain the infrastructure on which our economy and security increasingly rely. Those suppliers will collect the profits, wield the influence, and retain the strategic leverage. Over decades these advantages reinforce each other, compounding to create high-skill and - wage clusters which the UK sorely lacks. 

The development of sovereign AI capabilities is not something that will show up on a department’s balance sheet. But it is something we can - and should - choose to invest in. That means accepting that some high-ROI projects won’t emerge from the usual cost-benefit logic. For example, strategic reserves of assets such as oil, or emergency monitoring systems such as pathogen monitoring in wastewater, sit idle for long periods. They do not generate a ROI much of the time, but are strategically important by virtue of being exactly what you need when you need them, at moments which emerge unpredictably. It’s the main reason we use public money to support a domestic steel industry, to maintain a skilled labour force and industrial capacity for times when markets fail. 

In the context of building a strategic supplier and skills reserve this will mean allocating capital to departments who would otherwise buy from abroad, to enable them to buy British instead. It means funding projects that may be marginal in their own right, but which strengthen the industrial base in a way that pays dividends over time. This is not an abstract idea. China’s dominance of global electric vehicle markets was built using precisely this kind of strategic investment, not just in products, but in ecosystems. 

So what does sovereign control over AI actually mean? In short, it means retaining the ability for the government to make meaningful decisions at moments of geopolitical stress (such as war or international competition). It means exercising the power to capture value for the UK and the people who live here, whether on an ongoing basis or with which to trade at high-stakes moments when the smooth functioning of markets can’t be relied upon. 

It can refer to several distinct but overlapping goals, each of which we should be willing to make trade-offs to pursue. These include:

• Output Control: The ability to shape the responses given by models, either to align more closely to British values or to give locally relevant guidance and advice, such as by responding in line with NHS guidance.  
• Access Guarantee: Assurance that you will have continued access to models and tools in times of stress, which cannot be unilaterally removed by another country or foreign supplier.  
• Level Playing Field: Certainty that British firms are not disadvantaged or disfavoured when competing for scarce resources, such as GPUs or data assets, either by preferential partnerships or on price terms.   
• Repurposability: Retaining the legal basis and technical capacity to redirect resources - such as compute or engineering talent - toward urgent national priorities in times of stress.
• Economic Rewards: Accrual of the economic benefits to the UK, from the sale and exploitation of models and AI-powered software and creation of high paying jobs which keep highly skilled workers in the UK despite international competition for them. 
• Data Localisation: Storage of sensitive data on devices which are physically within the UK, so they can be safeguarded.
• Deterrence:The ability to make credible claims to be self-sufficient or to control levers for offensive action - military or economic - so that other states lack (or believe they lack) any strategic advantage which they can use in negotiations or conflict. 

These are all legitimate forms of sovereignty. But many of them involve benefits that are diffuse, interdepartmental, or long-term. Consequently these are often neglected by individual buyers working under pressure to minimise spend. Some of them are best achieved by supporting a private sector company to succeed (e.g. a National Champion, such as Rolls-Royce) rather than by building state capacity directly. Doing this is valuable for the UK, but not a recognisable benefit for departmental budgets.  

Take the NHS. When it invests in developing AI-powered tools, that investment doesn’t just improve patient services. It also creates new supplier capacity, helps retain skilled developers in the UK, and contributes to the IP base needed for future defence or security applications. The MoD might benefit enormously from the existence of such tools and suppliers. These benefits are not booked to the NHS’s account. As a result, they go unrecognised - and unrewarded. The result is a public investment regime that ignores some of the most strategically important outcomes we could be pursuing.

There are ways to fix this. We can create a new model of ROI that allows departments to access top-up capital when they deliver whole-of-government or national benefits, not just internal efficiencies. This is not a call for everything to deliver on every priority (a trap Joe Hill has aptly called “Everythingism”). It’s a call for clarity. If sovereignty matters, let’s be explicit about paying for it. If strategic ecosystems matter, let’s subsidise them. If flexibility matters, let’s be willing to accept some inefficiency in the name of resilience.

A genuinely sovereign AI capability - one that is secure in times of stress and generates returns in times of peace - won’t emerge from isolated projects, each judged on narrow terms. As the UK prepares to deploy billions in AI-related infrastructure, we must not only choose the right technologies: we must also choose a better framework for investment. That means valuing flexibility, skills, and long-term revenue potential. And it means putting estimated value on things we can’t touch: trust, talent, and time.

The UK has many natural advantages: a world-class university sector, internationally respected regulators, and a strong record of institutional innovation. But without the right investment model, those advantages will leak away. A real national strategy would spend not just on machines, but on the conditions for people to succeed with them - and to succeed here.

The future of British sovereignty in AI will not be decided by where our servers sit. It will be decided by who builds, controls, and profits from them. The investment decisions that determine this are being made now.

‍

An abridged version of this article was published in Chamber Magazine in July 2025

‍