In a training facility on the outskirts of Beijing, more than a hundred humanoid robots stand in rows, each repeating a single task under the eye of a human operator holding a controller. A few miles away, at a purpose-built exhibition centre, a robot in a basketball jersey sinks free throws for visiting delegations while another performs a lion dance. The gap between these two rooms captures something that Britain's health service, mid-way through its own infatuation with automation and artificial intelligence, would do well to sit with.

China's humanoid robot industry has become a genuine spectacle. Rental businesses now number in the tens of thousands, hiring out androids for weddings, product launches and shopfront promotions at a few hundred pounds a day. State media tout the numbers. Beijing has set targets running into the hundreds of thousands of units by the end of the decade. Yet the companies making these machines concede, when pressed, that even their most advanced models reach perhaps eighty per cent of human productivity, and only on narrow, repetitive tasks such as sorting parcels. Fewer than one in ten of the robots sold by the sector's largest maker have gone anywhere near a factory floor. The rest sit in showrooms, universities and government demonstration projects.

There is a familiar shape to this story for anyone watching how automation and AI have been sold into the NHS over the past two years. The service has no shortage of vivid pilots: ambient voice technology capturing consultations, AI triage tools flagging deteriorating patients, diagnostic algorithms reading scans faster than radiologists. Each generates its own showcase moment, a trust chief executive quoted on the productivity gains, a minister citing the pilot in a speech about digital transformation. What tends to receive less attention is the distance between a successful demonstration and something that functions reliably across a service under the kind of operational strain the NHS currently carries, with waiting lists still elevated, workforce vacancies concentrated in exactly the specialisms these tools are meant to support, and procurement systems that struggle to move technology from trial to scale even when the trial works.

The Chinese robotics sector's own diagnosis of its limits is instructive. Engineers there point not to some grand failure of ambition but to unglamorous constraints: a shortage of real-world interaction data, the physical difficulty of building a robotic hand that is durable, compact and dexterous all at once, the absence of established engineering solutions to draw on. These are the same categories of constraint that quietly determine whether health technology succeeds in Britain. An ambient voice tool trained on clean pilot data behaves differently across the accents, comorbidities and clinical shorthand of a real outpatient clinic. An AI triage system validated in one trust's population does not automatically generalise to another with a different case mix. The MHRA's evolving framework for software as a medical device exists precisely because regulators have learned, sector by sector, that a demonstration and a deployment are different products requiring different evidence.

The more pointed lesson concerns who is actually driving demand. Chinese analysts now acknowledge that most orders for humanoid robots come from government rather than industry, sustained in part by the political value of showing the world that China leads in strategic technology. NHS leaders should ask themselves an equivalent question honestly. When a trust adopts an AI tool, is the primary driver clinical need and evidenced productivity gain, or the reputational and political value of being seen to modernise, at a moment when the government's reform agenda under Wes Streeting's successors and the drive toward a digitally enabled NHS make technological adoption a visible marker of progress in itself.

None of this argues against investment in health AI, where the underlying case, easing workforce pressure and improving diagnostic accuracy, remains sound. It argues for scepticism about the gap between the demonstration and the deployment, and for procurement and evaluation processes robust enough to tell the difference. China's robot rental entrepreneurs have already discovered that novelty fades faster than capability improves. The NHS, with far less room for expensive false starts, cannot afford to learn that lesson the same way.