Sovereign at last?

Why the Tech Prosperity Deal is a Big Deal for AI sovereignty - but there's still more to do.

£150 billion in AI investment for data centres and supercomputers. Closer cooperation with the US on quantum computing, energy and AI. A new AI Growth Zone in Blyth and a new supercomputer in Essex.

On the face of it, the US-UK Technology Prosperity Deal announced last week should be great news for the UK. And yet, Nick Clegg has warned this is – in his own words - a “sloppy seconds” deal, that will cement the UK’s place as essentially a technological vassal state. A mere satellite of the US.

I think this is overly pessimistic. The infrastructure will be UK-based, supporting British research and businesses under our regulatory framework. UK companies like Nscale (who provide AI cloud services) are participating in the buildout, creating genuine partnership. And realistically, only the US is able to deliver investment at this scale and speed (as security considerations rule out China); European alternatives like SAP or OVH operate at much smaller scales. For the most sensitive national security workloads, the UK should definitely maintain alternative options - but this doesn’t preclude leveraging US investment for broader AI capacity.

That said, I also don’t think that securing this enormous investment package is a case of “job done” for building up the UK’s AI sovereignty. There’s a lot more that needs to happen to make sure this infrastructure buildout actually benefits the UK and truly secures its place as a leading middle power in AI. In this post I’ll explain why the investment is necessary but not sufficient for the UK to develop AI sovereignty in a real sense.

Why this investment matters: a window of opportunity

Although the UK is certainly not short of ambitions in AI, the main constraint to these is AI infrastructure. In a nutshell, the UK has extremely limited ‘hard sovereignty’ in AI – ownership over the building blocks needed to develop models or secure a strategic position in the AI value chain. The UK does not exert significant market share in any of the specialised elements for the chips that underpin AI: we are not a hub for wafer production, lithography tools, or a producer of GPUs. For obvious reasons, having significant market share in these parts of the AI “stack” (the set of technologies, tools and processes needed for AI systems) give you significant leverage – chips are foundational to anything AI-related.

To be fair, no one country has – or even could realistically achieve right now – complete domination over the entire ‘hard’ tech stack for AI. So this kind of AI sovereignty does not imply full self-sufficiency. But limiting the amount of control others can exert on you through key bottlenecks in the AI supply chain is an important consideration to avoid excessive dependence on any one failure point.

On this reading of AI sovereignty, the situation is not great. The UK is pretty much reliant on the goodwill of key allies and strategic partnerships (though in fairness it is building these up competently, through its diplomacy with the US and recognition in the Compute Strategy of the need for international agreements) for components such as GPUs. But there is another reading of AI sovereignty that focuses more on capabilities – how you can use AI. As the UK Compute Strategy puts it:

“For the UK, sovereignty means ensuring we have the ability to act independently and effectively where it matters most – to allocate compute to national priorities, protect sensitive data, and support UK research, innovation, and public services on our own terms.

In practice, this means ensuring that programmes like the AI Research Resource and AI Growth Zones result in capacity that is physically located in the UK and strategically aligned with our national goals – not just capacity rented from abroad”

This is where the importance of compute (basically, the amount of raw computing power available in your country) comes in. Compute gives you options: you need it to train AI models, conduct research, develop lucrative applications and so on. It gives you resources that are needed in order to carve out an economic niche in AI, for example to train specialised models, conduct innovative research, or build up valuable companies.

The UK is struggling in this respect too: in terms of ‘high performance compute’ (HPC), needed for scientific research and intense workloads, the UK ranked third in the world in 2005. Today, it does not even break into the top ten countries by amount of HPC.

In terms of general compute the UK fares better, having the largest market in Europe for commercial data centres. But here, too, the outlook until recently looked challenging: I have written before about the UK’s insanely high energy costs and sclerotic planning regime putting a significant check on the UK’s near-term ability to maintain its position as the leading data centre market in the region.

Building out this capacity now is important. Given how rapidly AI is moving, and how competitive the field is going to be, we don’t have 15 years to wait if we want the UK to stay relevant and influential in AI.

The newly announced investment is a significant step change in the UK’s pace here. Alongside that, the expansion of the AI Research Resource with the recent opening of the Isambard-AI supercomputer in Bristol is another massive step in the right direction. The announcement of a new supercomputer in Essex (looks like it will be modelled on the Jupiter supercomputer in EU) is very welcome too.

All of this will significantly bolster the UK’s compute capacity, hence giving it more options to act and create value. That’s why I see this injection of capacity as a vital first step to fleshing out the idea of UK AI sovereignty.

Why it’s not the last word on AI sovereignty

As I said on Times Radio last week, there’s so much more to do to translate this promised upside into tangible benefits for AI sovereignty. Compute is really a stepping stone along the way to the real objectives: as a country, becoming too valuable to ignore in the AI era, and ensuring AI has broad social and economic benefits.

That is to say that what will really build up the UK’s AI sovereignty is using compute in order to build up capabilities in useful areas, and in so doing develop an economic niche in the AI stack. There’s a lot of uncertainty in what these areas could be for the UK, but there seem to me to be two broad approaches the UK could look to.

One is innovation in chip design and hardware orchestration. It seems like some opportunity exists at this layer of the stack: although the current hardware paradigm (underpinned in Taiwan by TSMC, and in the Netherlands by ASML) has got us impressively far on AI, GPUs can struggle with certain features such as irregular data, sparsity, or conditional choices that often appear in complex AI workloads. So now, novel computing architectures are starting to emerge – new kinds of chips that are more naturally suited to the demands that AI inference will place on them. At the same time, a proliferation of various different kinds of specialised components creates a new problem: how to effectively orchestrate all these different hardware components for maximum performance. This ‘orchestration layer’ is the major control point in a world where we move past GPUs to a more federated compute architecture for AI. If you want more detail, there’s a fascinating piece about this on Artificial Intelligence Made Simple.

Given the dominance of other countries in fabrication, as well as the extraordinarily high set up costs, the UK would in all likelihood be competing at the design stage rather than hard fabrication here. But it wouldn’t be a case of competing completely from scratch. The UK’s semiconductor industry accounts for 2% of the global market, and is attractive for sector-specific talent: according to DSIT research, “72% of the dedicated, internationally headquartered semiconductor companies identified through this study undertake research, development, design and IP activity in the UK”. At the company level, there are already UK-based firms such as Graphcore and ARM Holdings that are involved in AI-specialised chip design. Many of these run a “fabless” model, where they design chips and systems but outsource the manufacturing.

There could also be opportunity at the orchestration layer, the software design that harnesses new specialised components. That would work well if the UK goes for a fabless approach – as it would want to do given the difficulty of directly competing on the fabrication stage. This orchestration layer itself might end up open-sourced, which would change the value proposition for the UK, although countries that contribute significantly to open-source software reap other benefits in terms of influence, talent, and indirect economic spillovers.

The second broad area is specialised AI application. This encompasses AI assurance, enterprise and public sector adoption, specialised small-scale AI models and innovative regulatory frameworks. There also seems an opportunity here; although businesses have been keen to adopt AI, in most cases we are not yet seeing this reflected in bottom lines. There are still data, privacy and regulatory concerns around further adoption, too.

Given the UK’s existing advantages in talent and institutions it seems reasonable to think that the UK could create comparative advantages in this kind of area. Consider that the UK created the world’s first state-backed AI Safety (now Security) Institute. We also have strengths in sector-specific datasets, with the NHS being the main (perhaps controversial) example. London is a major hub for relevant specialist sectors such as FinTech and legal services. These specialisms could play an important role in pioneering enterprise AI adoption through fine-tuned models. Although this is still nascent, the Government’s apparent desire to carve out a “third way” on regulation between the EU and US models, plus the arrival of bodies such as the Regulatory Innovation Office hint at the potential for UK influence in regulation, standard-setting and model evaluation.

We would not be building frontier models (which is in any case not plausible given the sheer amount of scale needed for this – the US and China are realistically the only countries able to throw sufficient resources at the problem to compete at the frontier) or low-level hardware, but using our existing expertise and soft power to develop effective and trustworthy AI solutions to business needs is a real option for the UK.

The next steps

So what does the UK now need to do to really make the most of this shiny new investment?

On a strategic level, the Government needs to have a clearer vision about what impact it wants AI to achieve for the UK – beyond just “bigger and better”. How does it think AI can contribute to national policy missions? How will we use the AI Research Resource and AI Growth Zones (AIGZs) to make sure our startups and universities have the compute they need to grow, or innovate in the public interest? Where could the UK be looking to develop a global niche in AI – are there ‘upstream’ areas like quantum computing, AI-specialised chip design or ‘downstream’ application-layer areas where the UK could take a lead?

Of course, the Government will not be able to pre-empt everything that happens in AI; its role is not to ‘pick winners’ but to create a supportive environment for UK capabilities in AI to develop. But it should nevertheless have a more well-rounded story to tell about what AI success for the UK looks like. This is particularly important given that the public are a long way yet from being won over by AI, as several surveys but most recently a new study from the Tony Blair Institute has found.

This challenge is unlikely to abate any time soon – consider for example the likelihood that in the short term, gas will be needed to power some data centre clusters. This is clearly in tension with the government’s net zero ambitions, most notably its 2030 target for electricity decarbonisation. Rather than pretending this trade-off doesn’t exist, it would be better to get out ahead of this with a clear account of where the opportunities are for the UK and how communities can benefit.

On a practical policy level, there needs to be much more synergy between different areas to maximise the benefit that new infrastructure will have for the UK. Compute is one building block of making AI, but there are others. Data is another fundamental input, but the UK’s approach here is still unclear in 2025. We still don’t know what the government is planning on the National Data Library, which would join together public datasets to support research and innovation, more than a year after the election. Nor what the Government intends to do on copyright, after it retreated from imminent reform to a drawn-out process of technical working groups. This is important: even with compute available, if data is difficult to access and use for training or AI-related use, it will remain difficult to build AI models in the UK compared to other jurisdictions.

We also need to keep up the momentum on planning and energy policy. After all, these are needed to make sure that we can power AI infrastructure and do so quickly – and also avoid the data centre boom putting excessive strain on our grid. There has been some recent progress on nuclear Small Modular Reactors (SMRs), with commercial deals such as a proposed SMR near the old Cottam power station afoot, as well as an intention from the Government to cut the approval time for nuclear projects from 3-4 years to 2.

But policy could still be stronger on this front. We could go further in recognising decisions on reactor designs from trusted regulators, to speed up approval. There’s much more to do around reforming grid connections to make it commercially easier to set up SMR/data centre colocation by better facilitating more ‘private wires’ setups and flexible grid connections. There could also be something around permitted development rights to give colocation sites more flexibility in adding capacity over time. Britain Remade recently came out with an excellent report going into some more detail on how we can get nuclear energy policy truly firing on all cylinders.

Principally, the government could be stronger in designating AIGZs. To date, this process has relied on applications of interest, with a (very) slow drip-feed of new AIGZ announcements. Indeed, we do not know all that much more about where AIGZs will be or how they will be utilised than we did at the start of 2025. But the Government already has planning powers to go further and faster: it could use Special Development Orders, for example, to significantly increase the speed and/or scope of development.

I also think there are cleverer ways to use planning and tax policy to build buy-in from local communities. Something simple the Government could do is allow local areas to retain 100% of the business rates from data centres and allied developments such as SMRs, giving them an economic stake in these projects – something they could use to keep Council Tax low or invest in public services, without having to go with a begging bowl to Whitehall.

The UK is still a leader in AI talent and it remains an attractive place for skilled workers to locate. But we should not assume that the UK will remain in this favourable position indefinitely. Our visas are expensive, and the talent pipeline from universities to AI spin-outs is not always as solid as it could be. Other countries are stepping into the arena and making offers to attract AI talent. Though the UK has made a raft of good announcements on attracting talent recently, this is going to remain a competitive space.

Ensuring a supportive business environment for UK-based AI startups will also be important. The UK has a persistent problem of startups – most notably Wise - spurning London and listing in the US instead. More companies, such as Monzo, Revolut and Cleo could follow suit. If there is still employment and economic activity happening in the UK because of newly US-listed companies, this isn’t necessarily the end of the world. But without market reform and more efforts to make London a more attractive place to list, UK tech success stories could continue to end up looking more American than British, with much of the value uplift flowing to the other side of the pond.

So yes, compute is necessary. This tranche of investment in it is a good thing. But you need a vibrant ecosystem around that compute to really capture value. To put it more bluntly: what good is it having a massive data centre in the North East if nobody is interested in building companies or innovative AI applications in the UK?

Now the investment is flowing in, a whole-of-government approach is needed across data, energy, planning, tax, skills, migration policy and more besides to translate options into value. That’s where the real AI sovereignty will come from.