The UK needs to do more to support its own AI companies

The writer is a Labour MP and former foreign secretary

As foreign secretary, many of my counterparts would come to me saying they wanted to “do AI with Britain”. At the time I was only able to direct them to the AI Security Institute, a state research organisation. This was in sharp contrast to how the Americans or the French handle such a request, which is by pointing people straight to companies like Palantir or Mistral.

It was thus an astute move by Andy Burnham to announce at the United Nations last week that he wanted to put AI at the heart of Britain’s chairing of the G20 next year.

But to do so will be a challenge. Despite the smiles at the recent US-China state dinner in the White House, geopolitical tensions are rising. One looming flashpoint comes from reports that the US plans to ban cutting-edge, open Chinese AI models, citing security concerns and IP theft, and thus force its allies to follow suit.

This would knock the legs out from under the European AI start-up scene and foreclose the faint hope of a genuine breakthrough in diplomacy on the subject at the G20. Martin Casado, a partner at Andreessen Horowitz, has said that when entrepreneurs walk through the VC firm’s doors there is a high chance they are using a Chinese open-weight AI model. The reason is cost. According to Vercel’s AI Gateway Production Index, these models cost roughly one-tenth of their average American counterparts.

China is also aggressively promoting them. Knowing full well that every technology of tomorrow, from autonomous cars to automated factories, will be running on AI, the Chinese want to become the default model provider around the world. This will allow them to escape US export controls on critical components like chips by binding allies more closely to the Chinese stack.

There are genuine security concerns about these Chinese open-weight models such as DeepSeek’s or the Alibaba Qwen series, and US agencies allege that many Chinese companies have inappropriately trained their models on American AI at industrial scale, using proxies and fake accounts to breach their terms of service.

But they are also a commercial threat to the US. Why pay for an expensive American AI company subscription if you can just use a cheap Chinese model, as Airbnb does, or adapt one, as Thomson Reuters has? According to OpenRouter, a service used to access AI models, Chinese models have sat above 30 per cent of US companies’ token usage since February 2026.

The risk is mounting. All the White House has to do is decree that American software and cloud providers cease working with Chinese AI models and every British company that relies on some aspect of Microsoft, Google or Amazon — which is pretty much all of them — and is also trying to use Chinese AI will struggle to operate. Such a ban would be an immense setback to the European tech sector, where many companies use Chinese open-weight models, including frontier AI developer Mistral.

That is why Burnham needs a plan to avoid a US ban on Chinese AI that would divide the world. This requires a three-part strategy.

First, the prime minister should develop a diplomatic plan with our allies to present to President Donald Trump, as we did in developing the Coalition of the Willing for Ukraine. Taking the Americans at their word about security and IP concerns, such a coalition should propose turning the UK’s AI Security Institute into a collective allied regulator that can assess whether Chinese open-weight models pose a threat, with a transparent, mutually designed process about the repercussions of a ban.

Second, we need to invest more in British AI so that we are not vulnerable to external pressure. There are extraordinary companies like Fuse, which is working on our energy grid, Era4, which is building data centres, or Fractile and Olix, which are building chips. The government needs to do more to support existing companies.

We’ve made huge strides in launching the UK’s Sovereign AI fund this year, which has invested in 18 companies, including the British AI giant Demis Hassabis’s AI drugs venture, the efficient-compute company Callosum, and the new AI lab Ineffable Intelligence. These companies are building sovereign British AI capacity and the Foreign Office must promote them.

As I saw in government, when the US abruptly banned the use of Anthropic’s latest models, the White House can take us by surprise. A ban on Chinese open-weight models would be far more than a temporary suspension of access. Burnham can get ahead of this by proposing a way to address security concerns while also building up a British technology stack that leaves us less vulnerable to coercion by either friend or foe.