I was lucky enough to attend London Tech Week 2026 and the AI Summit last week, where UK sovereign AI took a real step forward. However, how much we benefit from the announcements will come down to how quickly we can translate policy, compute, chips and talent into a coherent national strategy, that gives Britain a seat at the global table, not just a seat in the audience.
Why this matters: a warning shot
One recent event has put everything else into context. On Friday evening, just days after launching them, Anthropic switched off its two most powerful models (Fable 5 and Mythos 5) for every customer in the world. The trigger was a US government export order banning use by any foreign national, even Anthropic's own overseas staff, and rather than block some users and not others, the company pulled the models completely. They came off the major clouds too, including AWS, and as of this weekend they are still down. Anthropic has pushed back, saying it was given little detail and saw no evidence of a serious flaw, but that argument is still playing out.
Think about what that means for us. British firms had started building on a frontier model, and overnight a decision made in another country took it away. That is the entire case for sovereign AI in a single event: if you rent your AI from abroad, someone else holds the off switch. Everything announced in London last week is, in the end, about making sure we are not left in that position.
A big week for British compute
The headline number is £1.1 billion. Technology Secretary Liz Kendall used London Tech Week to launch the UK AI Hardware Plan, anchored by a new £750m national AI supercomputer – a mixed-chip system due to go live in 2030, joining Isambard-AI to be part of the AI Research Resource.
Inside that £750m sits £400m for specialised chips, the same commitment the Prime Minister had announced earlier that day. Part of it is a £150m Advance Market Commitment for next-generation inference chips, to be awarded this summer – a real opportunity for British firms like Fractile and OLIX.
There is also a second, separate £750m, this one for the Next National Supercomputing Service in Edinburgh, the successor to ARCHER2. Add the £500m Sovereign AI Fund from April, a £120m AI Hardware Innovation Programme, £45m for skills, and a new deeptech hardware fund led by Playground Global with up to £150m from the British Business Bank (its largest ever fund investment), and for the first time this looks like a long-term plan rather than a string of one-off bets.
It was not all future tense, either. On Wednesday last week, Cambridge formally launched Zenith, now the UK's largest AI-for-science platform. Zenith is the upgrade to Dawn, built by Dell with AMD, and the result of a £36m Department for Science, Innovation and Technology and UK Research and Innovation investment that delivers roughly six times the performance of its predecessor. The headline use cases are health, energy and the environment – from cancer research and diagnosis to detailed weather forecasting for navigation in harsh conditions like the Arctic. It launched alongside a new public-private Sovereign AI Innovation Lab (SAIL) and sits next to Sunrise, the fusion AI system being built with UKAEA on the same Cambridge-designed architecture.
Sovereign AI is not isolationism
The clearest framing of the week came from Kanishka Narayan MP, the Minister for AI and Online Safety. Sovereignty, he argued, is not about owning every layer of the stack; it is about having enough leverage to make our own choices on the technologies that matter, for national security, economic progress and human dignity.
Matt Harris, Hewlett Packard Enterprise's Managing Director for the UK and Ireland, made the same point from the industry side on the sovereign AI panel, alongside George Osborne (OpenAI), Judith Dada (Visionaries Club) and chair Dan Milmo of The Guardian. The argument was simple: the UK will not out-spend the hyperscalers, but we can out-strategise them, with smart regulation, faster decisions and an open, distributed ecosystem. It was great to hear Matt (who I bumped into at the AI Summit on Wednesday) and Kanishka speak so warmly about Isambard-AI being such a critical piece of UK infrastructure.
On a separate panel later in the week, Oliver Ilott (DSIT's Director General for Emerging Technology and AI) set out a clear goal: he wants the UK to be the easiest country in which to deploy overseas models, not one that keeps them out, and he made the case that investments like the new hardware plan will earn us a seat at the global table. He also talked about work the AI Security Institute has done on Isambard-AI, including the recent data poisoning study – a joint project with Anthropic and the Alan Turing Institute. Keep an eye on the BriCS YouTube for more on this soon.
Private capital showed up too
Government money was the public anchor, but the corporate pledges may be the bigger signal:
- AMD committed up to £2bn over five years.
- Nebius committed roughly £1.7bn to grow UK AI capacity to 65MW by 2027.
- BT Group and Nscale are building sovereign AI data centres across three existing BT sites.
- Nvidia's £2bn investment in UK startups, first announced last September, is now flowing into hubs in London, Oxford, Cambridge and Manchester.
Combined with public money, that is comfortably over £4bn in a single week.
Why Isambard-AI matters more than ever
From where I sit at BriCS and Isambard-AI, the best part of the week was not the headline numbers – it was the proof that UK sovereign compute is already delivering real results.
A few that stood out:
- Doubleword is running inference on Isambard-AI at 90–95% lower cost than other leading providers, with 70x faster model cold starts.
- Prima Mente is using its allocation to build biological foundation models for Alzheimer's, Parkinson's and ALS.
- Cursive and Ineffable Intelligence are early Sovereign AI Fund recipients building on UK infrastructure.
5,400 Grace Hopper Superchips, running on zero-carbon electricity, with British startups doing world-class work on them – that is what "AI maker, not AI taker" looks like in practice.
A nice personal moment too: our Simon McIntosh-Smith and CTO Sadaf Alam were both named to the inaugural SCW75 list this week. Well deserved.
Simon also delivered a fantastic talk at London Tech Week, sitting down with Dr Claire Thorne for a session called 'Isambard-AI - Tackling Humanity's Biggest Challenges.' The session covered many themes from project build and funding to software stack and agentic users. It was great to hear Simon talk about the impact of some of the projects being run on Isambard-AI, and he gave an exciting insight into the future of BriCS.
The sovereignty gap we still need to close
Several speakers were honest about the gaps. The UK public sector still leans heavily on US-owned tech stacks, Europe hosts only around 5% of the world's data centres, and our energy costs are structurally higher.
The Fable 5 shutdown I opened with is the perfect example of the risk – and the encouraging news is that the UK answer is already being built. Cosine , one of the companies selected under the £500m Sovereign AI programme, is training Lumen Sovereign, a frontier coding model. Cosine has signed up a coalition of UK institutions to shape it – including BT, Lloyds, NatWest, LSEG, PwC, BAE Systems, Leonardo UK and Babcock, with the Alan Turing Institute involved too, with delivery targeted for late 2026.
The link to Fable 5 is hard to miss. Fable 5 is itself a frontier coding model, and the defence, finance and critical-infrastructure sectors Cosine is building for are exactly the ones that cannot afford to have their AI switched off by a foreign order. A sovereign coding model trained on UK compute is precisely the kind of leverage point that turns a cautionary tale into a strategy. As Cosine's CEO Alistair Pullen puts it, enterprises are waking up to the risk of depending entirely on foreign providers for the most important technology of our generation.
Neither the US nor China owns every layer of the AI stack, and neither will we. But if we do not own enough leverage points – specialised chips, sovereign datasets, national supercomputers, open models, and the talent to run them – we end up as a customer of someone else's AI, not a shaper of the global ecosystem.
Playing to our competitive advantages
Our edge is not out-spending Silicon Valley. It is:
- World-class research compute: Isambard-AI, Zenith, the new AIRR system, and a successor to ARCHER2 at Edinburgh.
- Mission-focused systems like Sunrise for fusion energy research.
- A growing chip design base plus the Scaling Inference Lab from ARIA and CommonAI.
- A dense startup cluster that can actually get on these machines, rather than being priced out by hyperscale cloud.
If we keep systems like Isambard-AI close to researchers, startups and public sector users, and line up compute, chips, skills and governance behind one clear vision, the UK can be globally relevant without trying to match US or Chinese spend pound for pound.
The announcements last week were the starting line, not the finish. The real test is execution between now and 2030.