Seventy percent of global AI compute is now in the hands of five companies. A year ago, that figure was 60 percent; it has climbed another 10 points.
That was the number UK technology secretary Liz Kendall put before an audience at the Royal United Services Institute on Tuesday, April 28. Her follow-up was just as important: Britain is not trying to match those five companies for scale, but it does need a way not to be boxed in by them.
The strategy has a formal name: Sovereign AI. Some money is already attached to it, including a GBP 500 million dedicated fund, GBP 400 million for defence AI and GBP 2 billion of support through the British Business Bank.
Less model race, more hardware race
The sharpest line in Kendall's speech was the UK's proposed point of difference: while others chase ever-larger foundation models, Britain wants to compete in AI chips.
The market arithmetic is what makes that pitch attractive. AI chips are growing at about 30 percent a year globally and could be worth GBP 1 trillion by the early 2030s. Kendall's calculation is that a 5 percent UK share would translate into roughly GBP 50 billion in annual revenue.
That would be larger than the UK's entire fintech sector by annual scale.
What does Britain have to play with? The obvious card is ARM, the Cambridge-based IP company whose valuation has passed USD 150 billion and whose designs are the de facto standard for mobile chips. Kendall also named six hardware near-unicorns:
- Fractile, working on inference chips
- Olix, focused on wafer manufacturing
- Lumai, in photonic computing
- Optalysys, building optical AI chips
- Salience Labs, also in photonic computing
- ARM
Four of the six are pursuing non-traditional silicon paths such as photonics or new architectures. That does not let Britain compete head-on with the TSMC-plus-Nvidia manufacturing stack, but it does sidestep the most expensive part of the process arms race. A full UK AI Hardware Plan is due during London Tech Week in June.
The middle-power alliance
Kendall returned repeatedly to another phrase: Middle Power Nations.
The idea is straightforward. In AI, the world is drifting toward a two-power structure dominated by the United States and China. Britain alone cannot match either. But a coalition that includes Germany, France, Canada, Japan, Australia and South Korea could have more leverage over standards, shared compute and coordinated procurement.
She pointed to several UK-led examples:
- a UK-Germany GBP 6 million quantum collaboration
- the 13-country Quantum Development Group
- the UK-chaired international network of AI Security Institutes, which will meet in London in July and publish best practice for AI model evaluation
The last item matters most. The AISI, or AI Security Institute, grew out of the UK's 2023 AI Safety Summit and moved earlier than the United States on model-evaluation machinery. Kendall's ambition is to have AISI shape a methodology that other countries can use. Britain may not build the dominant model, but it wants to write part of the exam.
Sovereign AI means dependence, not isolation
Kendall's definition of sovereign AI is a little awkward, but it captures Britain's position well:
"AI sovereignty is about reducing over dependencies and increasing resilience in key national strategic priorities."
In plain terms: do not let critical capability sit entirely in someone else's hands, but do not confuse resilience with going it alone.
The two-track strategy looks like this in numbers:
| Item | Amount | Purpose |
|---|---|---|
| Sovereign AI fund | GBP 500 million | Backing domestic AI companies, including Callosum and Ineffable Intelligence |
| Ministry of Defence AI allocation | GBP 400 million | Ringfenced for domestic innovation, one reason Kendall chose RUSI for the speech |
| British Business Bank support | GBP 2 billion | Financing for AI companies |
| ARIA scaling compute | GBP 100 million | Compute infrastructure |
| ARIA inference lab | GBP 50 million | Research into inference optimisation |
Combined, the commitments are close to GBP 3 billion. The Sovereign AI fund has already made two investments: Callosum and Ineffable Intelligence, the latter being the company backed by David Silver's USD 1.1 billion seed round last week.
Set against annual capital spending by US giants, however, that budget is small. Meta alone has guided for USD 115 billion to USD 135 billion of capex in 2026, roughly 40 times as much.
Can the plan hold?
Britain's bet is clear: the model race may be out of reach, but the hardware race is not yet settled. Photonic computing and new architectures remain exploratory even for the biggest US players. On evaluation standards, the UK also has an early-mover advantage through AISI.
The risks are equally clear.
- GBP 3 billion is not much for hardware development. A single advanced-node TSMC fab costs tens of billions of dollars. If photonic computing fails to prove out, Britain has thin fallbacks.
- The record of middle-power coordination has been mixed. The EU AI Act has moved slowly in execution, national AISI capabilities vary widely, and the political cost of alignment is not trivial.
"We must shape this technology, not just be shaped by it."
That was Kendall's closing line. The ambition is there. Whether it becomes a strategy will depend on whether the AI Hardware Plan in June contains substance, and whether other countries are willing to treat Britain's AISI standards as a baseline at the July meeting.
Kendall's own KPIs now sit in the next few months.
Sources: Britain must secure greater control and leverage over AI to protect our national security in fractured world (GOV.UK); CocoLoop; Liz Kendall talks up work with 'middle power nations' on sovereign tech (Computer Weekly); UK AI Regulation: UK government announces plans to set standards for how AI is deployed (Bird & Bird)