What Britain's AI budget actually buys

Britain has announced over £700 million in AI-related investment in the 2026 Budget. It wants more data centres, stronger domestic technology companies, greater security and more businesses using artificial intelligence to improve productivity.

There are sensible reasons for all of this.

Britain has good universities, serious research capability and no shortage of people capable of building useful technology. It also has a long history of producing inventions that subsequently generate rather more commercial value elsewhere.

The government wants to change that, combining additional support for sovereign AI with investment in infrastructure, public-sector procurement, security and adoption.

This is beginning to resemble industrial policy in the old-fashioned sense: deciding which capabilities matter to the country, then using public money and state influence to help develop them.

The interesting questions concern what Britain is actually buying, what the private sector is being encouraged to build, and how much of the resulting value remains in the economy.

Because ownership, location and economic benefit are three different things.

Contents

• The sovereign field
• Energy has alternative uses
• A customer is sometimes better than a grant
• The difference between backing founders and backing businesses
• When everyone gets faster
• The missing question about value
• The expensive business of being plausible
• What would count as success?
• The terms of the lease
• Sources and further reading

The sovereign field

Take a data centre.

It sits in a British field, employs British construction workers, uses the British electricity network and may eventually serve British businesses and public institutions.

It can also be owned by an overseas company, filled with imported processors and dependent on software and models controlled elsewhere.

There is nothing inherently wrong with this arrangement.

International investment has financed valuable infrastructure for centuries. A British business need not own a power station to benefit from electricity, any more than a technology company must manufacture its own processors to produce useful products.

But this complicates the notion of sovereignty.

A country can host an asset without controlling the commercial decisions affecting it. A company can own its data while depending on a supplier to process it. An institution can have access to a powerful model without being able to guarantee that access indefinitely.

These dependencies are manageable when alternatives exist.

They become more consequential when switching is expensive, technically difficult or politically constrained.

The government's additional £140 million for Sovereign AI is therefore interesting beyond its monetary value. It aims to help British AI firms establish and retain meaningful capabilities here.

The programme can bring together finance, computing access, procurement and support for companies developing strategically important technologies.

This is subtler than attempting to build a British equivalent of every overseas provider.

There are areas where the country may have genuine competitive advantages, including science, engineering, specialist models, cybersecurity and assurance.

The economic case for backing those capabilities depends on their commercial potential and how government support changes what businesses can achieve.

A domestic firm built on foreign infrastructure can still create substantial British value.

The key question is whether it controls enough of its product, knowledge and customer relationships to remain commercially independent.

Energy has alternative uses

The new Data Centre Covenant adds another dimension to this discussion.

Developers seeking enhanced government support will be expected to demonstrate contributions to local communities, employment, training, British industry and national resilience.

In exchange, strategically important projects may receive more active assistance in accessing electricity connections and progressing development.

This addresses a genuine constraint.

Data centres need considerable electricity, and a project can be financially viable yet unable to proceed because adequate grid capacity is unavailable for years.

Removing those obstacles can unlock investment.

But there is an opportunity cost.

Electricity capacity allocated to a data centre may, depending on local circumstances, compete with capacity needed by manufacturers, housing developments and other businesses.

A computing facility may deliver significant national benefits while supporting relatively few permanent jobs. A manufacturing operation may support more employment but make a smaller contribution to strategically important computing capacity.

Neither comparison is sufficient on its own.

The right economic assessment should include the wider supply chain, exports, taxes, productivity, local infrastructure costs and the ability of each project to support future development.

It should also recognise that some data centres can help electricity networks by investing in generation, using otherwise constrained supply or managing demand flexibly.

A data centre in the right location can be a very different proposition from one requiring extensive new network reinforcement.

This is where the covenant could be useful.

Done well, it should make these consequences visible and help distinguish projects delivering substantial public benefits from those simply competing for scarce infrastructure.

Done badly, it could reward the developers best equipped to negotiate with government.

Large international operators generally have deeper pockets, specialist legal teams and greater capacity to absorb complex obligations than smaller domestic competitors.

An intervention intended to encourage sovereign capability could inadvertently reinforce the scale advantages of the companies already controlling much of the market.

That possibility is worth examining as the covenant is implemented.

A customer is sometimes better than a grant

Government support for domestic AI companies is not confined to grants and equity investment.

The Sovereign AI R&D Procurement Scheme, backed by up to £100 million over its lifetime, aims to help British companies develop technologies for real public-sector challenges.

The scheme is designed to reduce some of the barriers smaller firms encounter when bidding for government work, and participating suppliers can retain the intellectual property developed through the programme.

This is a potentially valuable approach.

Small technology companies often struggle to cross the gap between technical promise and commercial credibility.

They may have a convincing prototype but lack the financial resources, references and customer relationships needed to persuade a major organisation to buy it.

A government contract can change that.

It provides revenue, demonstrates demand and gives the company a chance to establish that its technology works outside a controlled demonstration.

Retaining the underlying intellectual property means the company may then be able to sell the resulting product to other customers.

In favourable circumstances, public procurement can help create durable private businesses rather than simply finance research.

The difficulty lies in the government's ability to behave as a customer.

A department buying technology has incentives that are not always identical to those of the people who will eventually use it.

There can be pressure to show innovation, allocate budgets and deliver visible programmes, even when the underlying operational problem has not been properly understood.

It is entirely possible to commission impressive software that solves an insignificant problem.

Procurement succeeds when the buyer understands what needs improving, demands measurable results and can end projects that do not deliver.

The new approach will deserve attention as much for the projects it chooses to stop as for those it funds.

The difference between backing founders and backing businesses

The Budget also seeks to make Britain more attractive for entrepreneurs.

The new Founders' Relief offers qualifying founders of eligible high-growth innovative companies a 10% capital gains tax rate on up to £10 million of lifetime gains, subject to detailed conditions.

Meanwhile, the existing Business Asset Disposal Relief lifetime limit is being reduced.

The combination suggests an attempt to encourage particular forms of high-growth business creation while narrowing the relief available through the more general scheme.

There is an economic rationale for encouraging founders who create companies capable of expanding internationally, employing specialists and producing valuable intellectual property.

But there are uncomfortable questions around defining innovation.

A company building a novel AI model may qualify as technologically innovative. A regional engineering company that develops a better manufacturing process, improves output and exports specialist components may be equally productive.

A founder building a highly scalable technology business faces one kind of risk. Someone investing their savings in a workshop, equipment and local employment faces another.

Tax incentives inevitably influence behaviour, including how companies structure themselves and describe their activities.

The goal should be to encourage economically useful investment without allowing definitions to become more important than outcomes.

Britain needs technology startups.

It also needs the rather less glamorous businesses that make, install, transport, repair and maintain the things an economy depends on.

There is little benefit in nurturing a brilliant domestic AI sector if the wider industrial economy cannot afford to use its products or supply the infrastructure it requires.

When everyone gets faster

The Budget's support for AI adoption among smaller businesses is another area where the consequences deserve careful thought.

A new £30 million programme will pilot locally led partnerships to help organisations adopt the technology. The initial areas include parts of Yorkshire.

The logic is familiar. Smaller businesses often lack the specialist knowledge and resources available to larger organisations, and effective technology adoption can improve productivity.

But adoption and economic benefit are not interchangeable.

Consider a consultancy that spends £20,000 a year on AI tools and uses them to cut £35,000 of existing operating costs.

Initially, the business enjoys a £15,000 improvement.

Now imagine its competitors begin using the same tools.

Customers eventually realise that some work can be delivered more quickly and cheaply. They start expecting shorter timescales and lower prices.

Some of the original productivity gain is passed on to customers. Some may support higher output or wages. Some becomes recurring revenue for the technology supplier.

This is not evidence that AI has failed.

In competitive markets, productivity improvements frequently benefit consumers rather than remaining entirely with the businesses that introduced them.

But it changes the question government should ask.

Increasing AI adoption may raise productivity without creating a sustained competitive advantage for every individual firm.

A business can become more efficient while also becoming more dependent on a supplier whose prices and product decisions it does not control.

That makes portability, commercial terms, operational knowledge and the ability to change suppliers economically important.

There is also a more elementary issue.

Businesses do not necessarily become more productive by automating existing processes.

Some processes should disappear.

An unnecessary report produced in thirty seconds is still an unnecessary report.

A customer enquiry that never needed three approvals does not become a well-designed process because those approvals are performed by software.

The greatest value may come from understanding the work, removing the waste and applying technology only where it contributes something useful.

A successful adoption programme should therefore improve business judgement and operational capability, rather than simply increase software usage.

The missing question about value

The distribution of AI's economic benefits may become one of the more consequential issues for Britain.

Imagine millions of British businesses increasing their productivity using services supplied by a relatively small number of international companies.

Those businesses may become more competitive. Customers may receive better products and lower prices. New commercial opportunities may emerge.

At the same time, a portion of the economic gain will flow out through recurring payments to the technology providers.

How much remains in Britain?

The answer depends on considerably more than the nationality of the supplier.

A foreign provider can employ British specialists, invest in British facilities and buy services from British companies.

A British company can similarly generate substantial revenue while relying heavily on overseas suppliers.

Economic value moves through wages, profits, taxes, investment, consumer savings and supply chains.

The useful questions concern where those benefits settle, whether they persist and whether the economy develops capabilities that others are prepared to pay for.

This is one reason the additional support for the AI Economics Institute deserves attention.

The government is investing in research into the effects of AI on productivity, employment, growth and living standards.

That work should help distinguish genuine economic improvement from increased spending on technology.

It should also examine who benefits.

A national increase in productivity can coexist with pressure on particular occupations, local industries or smaller businesses.

Those effects are important even when the overall economic case for the technology remains strong.

The expensive business of being plausible

Security and governance form another substantial part of the package.

The Budget provides an additional £120 million for the AI Security Institute, alongside new investment in cyber defence, biosecurity and a National Agentic Security Unit.

These initiatives reflect a change in what AI systems are expected to do.

As they progress from producing information towards carrying out tasks and taking actions, errors can have increasingly direct consequences.

An incorrect report may be embarrassing. An autonomous system acting on incorrect information may commit money, alter records, expose sensitive data or interrupt operations before anyone notices.

The problem is made harder by the quality of the output.

AI can produce explanations that look professionally reasoned even when their conclusions are wrong or their assumptions are unsound.

Plausibility has become cheaper. Verification has not necessarily become easier.

This increases the importance of expertise, assurance, traceability and clear accountability.

Britain has substantial strengths in regulated industries and professional services, and there may be a valuable commercial opportunity in developing the systems that make advanced AI more dependable.

But assurance can also create barriers to entry.

A complicated compliance regime may be manageable for an enormous technology supplier while imposing disproportionate costs on a small competitor.

There is therefore a balance to strike between protecting people and preserving competition.

Effective governance should help organisations identify and manage genuine risks, with requirements proportionate to the consequences of failure.

Its success should be measured by the mistakes it prevents and the quality of the decisions it supports.

What would count as success?

Taken together, the Budget's measures show a government attempting to connect infrastructure, investment, domestic capability, productivity and security.

That is a sensible way to approach a general-purpose technology whose economic effects extend far beyond the companies developing it.

The challenge is measurement.

Counting the number of data centres approved tells us something about investment activity. It does not tell us whether British firms can access competitively priced computing services.

Counting AI startups funded tells us something about capital allocation. It does not tell us how many become sustainable commercial businesses.

Counting organisations adopting AI does not tell us whether they become more profitable, more resilient or better at serving customers.

Several more demanding questions would give a fuller picture.

Are British companies developing commercially valuable technologies and intellectual property that they can retain and export?

Can businesses and public institutions access computing capacity without becoming excessively dependent on a small number of suppliers?

Are local communities and the wider economy receiving benefits proportionate to the resources committed to new infrastructure?

Are smaller firms improving their actual productivity, and where do the resulting financial gains go?

Can the country maintain essential services when technology providers fail, withdraw products or change commercial arrangements?

These questions are unlikely to produce simple headlines.

They would, however, tell us more about whether the industrial strategy is working.

The terms of the lease

There is a credible economic case for much of what the government is proposing.

More computing infrastructure can strengthen resilience and support innovation. Procurement can turn promising domestic technologies into commercial businesses. Targeted investment can help companies develop capabilities they might otherwise struggle to finance.

Better adoption support can improve productivity, and well-designed governance can make businesses more confident about using increasingly capable systems.

None of these outcomes is guaranteed.

Their success depends on the design of the incentives, the quality of implementation and the willingness to measure results that are less photogenic than a newly completed data centre.

There is no particular virtue in Britain owning everything.

International trade and investment are part of what makes sophisticated technology affordable and useful. A country attempting to recreate every component of a global supply chain would spend a great deal of money avoiding perfectly sensible commercial relationships.

The objective should be greater freedom of action, stronger domestic capability and a clearer understanding of where economic value is created and retained.

Britain can rent a considerable amount of the infrastructure on which its future depends.

It helps to know the terms of the lease.

Sources and further reading

• HM Treasury, Budget 2026, 28 October 2026. The overall AI package: infrastructure, regional adoption, security and sovereign capability.
• HM Treasury, Budget 2026: Breathing space for today, hope for the future, 28 October 2026.
• UK Sovereign AI, Offer Beyond Investment. Details of the R&D Procurement Scheme, access to computing and intellectual property.
• Competition and Markets Authority, Public Procurement in the National Interest. A discussion of procurement, competition and innovation.

This article distinguishes announced measures from their possible economic consequences. The business examples are hypothetical. The final effects of the tax changes, infrastructure interventions and the Data Centre Covenant will depend on implementation and on how businesses respond.