The next sovereignty crisis is already here

In 2023, I wrote that the UK needed its own home-grown foundation research labs if it was to have a hope of controlling its destiny in the world post the AI revolution. As I remarked in the opening paragraph, I was already 5 years late to this thinking.1 In April 2026, the UK government launched a Sovereign AI fund, accepting that foundation AI is critical sovereign infrastructure.2 We are now building compute, investing public capital and trying to attract and retain frontier AI talent, to work in UK startups. I had nothing to do with the steps the UK government took and I can take no credit for what has happened. It took the country 8 years from Ian Hogarth's original warning to respond but fair play, it did respond. Huge round of applause for everyone involved. It's not perfect, it might not work, but we are doing something about it.

So it seems like a good time for me to write about the next sovereignty crisis: Robotics.

Since 2023, we've seen huge adoption of machine intelligence operating in the digital world. In 2026, we are seeing AI transition to becoming embodied in the physical world. AI is moving from the production of words, images and software into the realm of physical control of vehicles, drones, medical devices, industrial machines and domestic robots. While we are busy congratulating ourselves for waking up to the first transition, we are already falling behind on the second.

Why should we care about who controls physical AI? Isn't it an accepted fact that all the best machines get made in China and cut-throat competition means we will always be able to buy them at prices lower than we can manufacture ourselves?

If Sovereign AI was about retaining control over knowledge work, then Sovereign Robotics is about retaining control over physical work.

The unit of sovereignty in robotics is not simply an AI model that runs the robot's brain, or the hardware that makes the robot move. It is the closed loop from intelligence, to embodiment, to manufacturing, to deployment data, and back again. That's a flywheel that once started will compound for the players in the space. These robots will be in every home, office and hospitality venue in the country. They'll be serving on the front lines of our armed forces and healthcare systems. They'll be running the manufacturing lines inside our factories, picking orders from shelves in our warehouses and digging up our potatoes in our fields.

The question we need to ask ourselves today is this: who will supply Britain’s physical labour over the next fifty years and who will own the machines through which that labour is delivered?

The emerging world order of Physical AI

The startup and capital race around Physical AI already has two dominant poles in the US and China. America has a concentration of frontier AI talent, venture capital and technology infrastructure. Latterly it has combined this with trade restrictions to slow down Chinese entry into its markets. For its part, China made robotics and automated machine tools one of the ten strategic priorities in its Made in China 2025 plan, albeit in an age before the current advances in AI.3 It has an enormous advantage in terms of advanced manufacturing capacity and component supply chains. It already has a huge robotic industrial deployment footprint, engineering depth and an increasingly broad robotics startup base. And while China is currently broadly competitive with the US on industrial electricity costs (and far cheaper than the UK and Europe), it plans to bring about 1.5 TW of additional power-generation capacity onstream by 2030, roughly six times as much as the US over the same period.4

Both these countries are investing seriously in the companies that will automate physical work. Britain has pockets of excellent research and a tiny number of well-capitalised companies at the embodied-AI frontier, but nowhere near the same breadth, capital depth or ability to get machines into the real world.5 The default British position will therefore be to become a customer. Without deliberate action, British physical labour will be automated by American and Chinese robotics platforms. Britain risks owning neither the intelligence layer nor the manufacturing capacity.

Sovereign Robotics does not mean making every actuator, sensor and processor in the UK. It means having enough domestic capability that Britain retains meaningful choices over who supplies essential machines, who controls their software and data, whether they can be operated, repaired and updated independently, whether another country can interrupt their operation and (here's the geopolitical kicker) where the resulting economic value accumulates.

So let's imagine that we decide to do nothing at all. What could the UK look like in a future with robotic dependence on other countries?

Robotics dependence

In our default future a foreign intelligent machine could end up responsible for moving goods through British ports and warehouses. Could perform NHS logistics, diagnostics, surgery and care. Could inspect our nuclear energy infrastructure and patrol our territorial waters. Could determine how food is produced in British fields. Could manufacture goods in our factories... or make it uneconomical to manufacture goods in our factories because we don't have access to the best versions of them. Could have eyes and ears in every British home and office in the country, receiving software updates from overseas servers.

None of this requires malign intent. Each machine may be cheaper, more capable, better tested, supported by global organisations, already certified and integrated. They will be the default choice. No one will bat an eyelid when a Samsung washing machine gets augmented by a Unitree domestic robot that loads washing into it. Every individual purchasing decision will be reasonable but the national result will still be dependence. We will not wake up one morning and decide to outsource Britain's physical labour. We will do it gradually, through thousands of perfectly reasonable free-market purchasing decisions made in the absence of a domestic alternative.

This is not an argument against importing or adopting robots. Refusing automation would make the UK hopelessly uncompetitive and leave Britain operating in the 2nd tier of economic activity. No one wants that. The choice is not foreign robots or no robots. The choice is whether the UK adopts robotics while creating domestic capability, or becomes exclusively dependent on foreign suppliers. In 2016 that dependency didn't look like such a bad bargain. In 2026, we've learnt that we need to have more control over our destiny. And the consequences are not simply strategic. There will likely be a very real economic impact of this dependency.

Who captures the robotics dividend?

Automation converts work purchased as labour into work purchased with capital assets. When a worker is replaced or supplemented by a machine, some of the wage bill is converted into spending on hardware, licences, compute, spare parts, maintenance and support. Where does that value flow? Who is the ultimate beneficiary when we displace labour with capital assets? A human worker in Britain is a taxpayer and a consumer but a foreign-supplied machine is not.

If the UK owns some share of the robotics companies and productive assets, the robotics dividend can return as British salaries, profits, taxes, IP, supplier revenue, and yes, exports (I appreciate the irony).

If the machine, software, servicing and financing are all foreign, much more of that value leaves the country through imports and foreign profit. British firms may become more productive and export more, but they won't necessarily capture the highest-value layers of the transition. Physical AI will shift an enormous share of income from labour to capital.

If we have no domestic capability, the transfer will not merely be from British workers to British capital; it will be from British workers to foreign capital.

With the rise of LLMs, some spending previously directed towards knowledge work is becoming spending on tokens and compute. You can debate how much worker impact has been felt to date in the UK. What you can't debate is that an enormous and rapidly growing pool of global spending is now being directed towards tokens and chips supplied by Anthropic, OpenAI and Nvidia. At the same time, the UK’s share of global private AI investment has fallen from roughly 4.1% in 2020 to 1.7% in 2025.6 AI investment is not the same thing as value creation, but those investments are a proxy for market capture. It is difficult to argue that the UK has been a global winner in the knowledge work phase of the AI transition.

The second, physical AI phase could end up being considerably more impactful because physical work exists across almost every part of our economy. Imported robots can still make British businesses more productive and leave substantial value with their British operators. We've benefited enormously from international trade historically and international trade is not automatically extraction. The concern is what happens if we fail to capture the high-value substitution layer across such a large part of our economy. Transitioning away from labour dependence to capital dependence across such a large part of the economy could have an enormous impact on GDP. And this outcome becomes harder to reverse with every non-sovereign deployment. We risk a massive transfer of value to China via the backdoor by not having domestic robotic capabilities.

Why not just wait and see again?

Waiting eight years again will not merely delay progress; it will probably mean we never recover. Every deployed robot strengthens its supplier through a flywheel of data, learning and expertise. The robots that emerge into the market in the next few years will lay the foundation for the next generation of improved robots from their suppliers and will build a robotic supply chain that matches the manufacturing supply chain advantage that China has today.

As a side note, I don't believe the robotics supply chain and China's current manufacturing supply chain in electronics are necessarily the same thing. China is actively building a new capability here. Yes, it leverages many of the companies and processes that they have built their economic success story on, but it is not a given that they will win in robotics. The massive advantage China has enjoyed in the cost of labour will carry less weight in the robotics age as direct labour becomes a smaller part of the cost of production. This discontinuity is an opportunity for us. Energy, expertise and a new generation of components will come to dominate, and those are the things China is trying to build right now.

There is also a school of thought that says we don't need to compete in manufacturing the machines; we can just win in the intelligence layer and implant our own models into Chinese-made robots. My take on this is that the body cannot be separated cleanly from the intelligence. Sensors determine what the model perceives, kinematics feeding into the world model determine what actions it can learn, and actuators and control systems determine what actions the robot can execute. Frontier models can transfer useful knowledge across robots and draw on internet data, but they still need robot data and adaptation to the perception, kinematics and control systems of a specific body.7

We also know that robotic simulation does not translate into real-world performance perfectly without embodied testing. Real-world failures from the whole stack generate the information required to improve the model. In physical AI, the body is not a peripheral attached to the intelligence; it's part of how we actually produce the intelligent system. This holds true across humanoids, drones and medical robots. We already see American and Chinese robotics companies benefiting from this learning loop. In 2024 China installed 295,000 industrial robots. By contrast, the UK installed 2,500 and ranks 19th in the world.8 In 2026, China fielded 1,975 of the 2,056 robots at the World Humanoid Robot Games. The compounding loop has begun. Imagine where we will be in another 8 years.

So if we think it's a problem worth solving, and we think we shouldn't wait, can we kickstart Sovereign Robotics in the same way we just kickstarted Sovereign AI?

The approach we used to recover in AI won't work in robotics

The UK woke up late to Sovereign AI but when it did, DeepMind and the wider UCL diaspora had already spent years concentrating and training frontier talent. Demis Hassabis made it possible for the policy to be actionable because we had a ready-made credible cohort of researchers available to seed several companies. When I wrote my original article I thought DeepMind was a classic case of the one that got away. In retrospect, Demis' decision to take Google's money but keep the company firmly centred on London was the gift that kept on giving. Google bankrolled our Sovereign AI strategy.

Robotics requires a wider set of skills in which we have no comparable reserve. Robotics knowledge is accumulated through building, deploying and manufacturing machines. It requires competence across AI and world models, computer vision, embedded systems, mechanics, manufacturing and deployment. It cannot be recreated quickly once dependence becomes obvious.

Wayve, Humanoid and Odyssey show that Britain can create serious Physical AI companies. But they do not show that Britain can reliably create them at scale. Wayve has proved that Britain can build and finance a world-class embodied intelligence company, but it is not building the vehicles its software controls. Humanoid is trying to build the whole machine, which is much closer to what I'm talking about here, but it is very young and currently working with Bosch on industrialisation. Odyssey is betting on the world-model layer. That may become enormously valuable, but it also raises the question at the heart of this article: how much of that value can you hold if somebody else owns the body, the supply chain and the deployment? None of them have yet produced anything resembling a DeepMind-scale diaspora of robotics founders, and it would be unreasonable to expect three companies alone to supply Britain's national robotics talent base. We cannot wait for a robotics mafia to emerge from the sale of one of them.

In Sovereign AI we woke up late and found a reservoir of talent. In Sovereign Robotics, waking up late may mean discovering that the reservoir was never built.

Nobody knows which precise architectures or companies will win in robotics. The Government should not be looking to crown a national robotics champion at this stage - it would simply be a single point of failure. We need to put more irons in the fire, ideally several properly capitalised companies across the five sectors of defence, healthcare, industrial, agricultural and domestic robotics. Britain needs enough companies that a) normal startup failure does not eliminate domestic capability; b) customers have genuine British alternatives; c) knowledge and talent can be recycled across the sector.

If we could get to the stage where we had five viable companies competing in each of those five categories (a five-by-five target), it would represent a level of redundancy and resilience that would transform our economic sovereignty. There is nothing scientific about five-by-five. It is simply my test of whether we are making enough attempts to be serious. To produce five durable contenders, considerably more than five attempts must be seeded. Failure is both unavoidable and completely acceptable, provided the resulting people, knowledge and intellectual property remain available to the next attempts. Creating many more and diverse robotics startups is the only way to give us meaningful sovereign choice further down the line.

Private UK capital will not fund Sovereign Robotics at the required scale

Robotics requires more capital, over longer periods, than UK venture funds generally tolerate. Hardware takes longer to iterate and revenue typically arrives later than with software. Regulation and procurement extend the timeline even further in sectors like healthcare and defence. Our capital markets are still focused on fast-traction AI startups at a time when Chinese state investment and US venture investment have both moved on to investing heavily in robotics hardware capabilities. Overseas investment can and does fund some of the UK's category outliers, but UK private capital will not finance the breadth of attempts or the time horizons Britain needs on its own. A serious public response must also finance the brutal step from prototype to repeatable production, otherwise we will fund the inventions and import the resulting machines.

Frankly, private investors aren't in the business of underwriting the national value of retaining independent defence, healthcare or industrial capability. The country captures that sovereign value, not the investor, so why should the private market pay for it? The independent Smart Machines Strategy mentioned previously proposed a £100m venture fund.5 Compare that to Humanoid's latest round of $152m, Odyssey's of $310m, and Wayve's of $1.2bn - it doesn't sound like a serious response. A Sovereign Robotics fund cannot be smaller than the financing required by one meaningful robotics company. I don't know what the answer to that is, but it feels like the response will need to be measured in billions over a decade to seed growth and manufacturing capital in a serious way.

Government has three responsibilities if it cares about sovereign capability

1. Supply capital

Treat Sovereign Robotics with at least the priority given to Sovereign AI, with more patience because physical systems take longer to develop and deploy. The objective is to give Britain multiple shots at success, not keep companies alive when they cannot perform. The capital needs to be there from company formation through seed, Series A and growth rounds. The money must go into creating startups, not institutions. Public capital should bring private investment with it, while paying for the national value that private investors have no reason to price in.

2. Become a UK-first customer

Use areas in which the state already purchases physical work to give our startups an advantage:

  • defence
  • nuclear decommissioning
  • transport and infrastructure
  • healthcare and social care delivery
  • agriculture

Government defines the problems and buys outcomes from competing UK companies. Make it UK-only in areas of national security and UK-startup-led elsewhere, using first-customer and R&D programmes that fit within our procurement obligations.9 Establish trusted-supplier rules before foreign machines become embedded inside hospitals, defence, energy and critical infrastructure. Ban Chinese robot suppliers from sectors you want to protect. Yes, that is a risk, but without it, we won't get revenue, deployment experience, operational data and reference customers from those sectors. It is inevitable that Chinese robots will be subsidised by the Chinese state to help them win in these markets. We need to push back on the impact of that distortion.

3. Make domestic deployment easy

Create the globally fastest pathways for testing and approval:

  • medical, surgical and care robots through the MHRA
  • drones and autonomous aviation through the CAA
  • autonomous vehicles and mobile systems through CCAV-style mechanisms
  • collaborative industrial machines with HSE involvement
  • controlled trials across public infrastructure and agriculture

Alongside acceleration, give UK companies an advantage in these pathways. Again, the government does not need to design the machines. It needs to give UK companies enough advantage that they get financed, tested, purchased and deployed.

Companies also need somewhere credible, domestic or trusted, to turn a prototype into a product. Otherwise Britain will finance the invention and somebody else will accumulate the industrial capability. Not every component must be British. Trusted international supply chains will remain necessary, but a British company should not have to hand industrialisation to somebody else simply because there is nowhere credible to manufacture here. Humanoid working with Bosch is a rational choice today but the strategic question is whether future UK robotics companies will have access to a credible domestic alternative.

This is not a 2034 problem

Robotics is the infrastructure through which an increasing share of physical labour will be delivered and the companies that will control physical labour in the UK are being formed now. Without domestic capability, Britain will become dependent on American and Chinese platforms for that labour. Once Britain has a large foreign installed base, recovery becomes much harder.

The key takeaway is that we should not think of these robots as industrial IoT or smart infrastructure. These intelligent machines are the base capability of our physical workforce over the next 50 years. The Smart Machines Strategy contains much of the right diagnosis, but an independent report welcomed by government is not a funded national mission. The lesson from Sovereign AI is not that Britain can always catch up. It is that waiting eight years means you have to get lucky to catch up.

We cannot repeat the Sovereign-AI scramble with robotics. This time there may be no DeepMind alumni network to plunder, no industrial base to reactivate and no second chance to create the companies we failed to seed. Sovereign Robotics is not a problem for 2034. The best time to build the companies that will perform Britain’s future physical work was a decade ago. The second-best time is today.

We went gently into the dying of our manufacturing base. The transition to robots doesn't have to be the same. It's time to rage against the dying of our physical workforce. It's time for a UK policy on robotic sovereignty.

  1. Ian Hogarth's 2018 essay AI Nationalism made the case first.↩
  2. The government launched the £500m Sovereign AI fund on 16 April 2026. GOV.UK announcement.↩
  3. Made in China 2025 (issued in 2015) named robotics and automated machine tools as one of ten strategic industries. US Trade Representative summary.↩
  4. China ended 2025 with 3.89 TW of installed power-generation capacity and plans to reach 5.4 TW by 2030, implying about 1.51 TW of net growth across all sources. China's 2025 capacity and 2030 energy plan. The EIA projects total US electric-power and end-use generation capacity to grow by about 232 GW over the same period, including gas, nuclear, renewables and storage. EIA Annual Energy Outlook 2026, Table 9. The definitions differ slightly because China reports installed capacity while the EIA reports net summer capacity, but the gap remains roughly sixfold. Installed capacity is not the same as annual electricity output, and the mix of generation matters; this comparison is about the scale of infrastructure being built. The IEA reports that EU electricity prices for energy-intensive industry in 2025 were more than twice US levels and nearly 50% above China. IEA electricity-price comparison.↩
  5. The Smart Machines Strategy identifies 2,204 UK robotics and automation companies, but only ten had raised more than £185m and only two more than £500m. It says the US averaged 600-700 Smart Machine deals a year between 2018 and 2024, against 100+ in the UK. It also recommends creating a £100m Smart Machines venture investment fund; it is an independent strategy published and welcomed by government, not a government funding commitment. There's much to like about this report but I diverge from the authors on the need to create a centre of excellence and regional hubs. For me, funding startups trumps funding institutions. Smart Machines Strategy 2035.↩
  6. Based on comparisons of the 2021 AI Index Report and the 2026 AI Index Report.↩
  7. Cross-embodiment transfer is real, but even these models are trained on both web and robot data, and evaluated on real robots. Google DeepMind, Scaling up learning across many different robot types.↩
  8. The 2026 World Humanoid Robot Games attracted 2,056 robots from 666 teams. Chinese participants accounted for 1,975 robots and 641 teams, drawn from 157 companies and 200 universities and research institutes. Beijing Municipal Bureau of Economy and Information Technology. The International Federation of Robotics reports 295,045 industrial robot installations in China in 2024; the UK installed approximately 2,500 and ranked 19th. IFR World Robotics 2025.↩
  9. Above-threshold procurements covered by the WTO / UK trade agreements generally cannot discriminate against qualifying "treaty state suppliers". Below-threshold contracts can be reserved to UK suppliers, some R&D services are exempt, and certain defence and security procurements have wider discretion. GOV.UK guidance on treaty state suppliers.↩