China's push to integrate artificial intelligence into its healthcare system has highlighted bottlenecks in its rollout, from elite urban hospitals to rural clinics.

The application of the technology in the medical field has so far proved uneven, with caution coming to the fore in its use.

"That has put imaging — radiology in particular — at the front of the line," said Wu You, an associate professor at the School of Healthcare Management at Tsinghua University. "In these specialties, AI systems help process the flood of CT and MRI scans, improving both speed and consistency."

Digital pathology and hospital quality management have followed. "Tools that review prescriptions, audit medical records, and flag irregularities in disease-specific protocols are increasingly part of the daily workflow at major hospitals," Wu said.

But these applications remain concentrated in large, well-resourced institutions — the kind of places that already have the infrastructure and expertise to absorb a new technology.

When AI moves closer to the patient, and especially to elderly patients, things get more complicated.

One underappreciated issue is the uneven capacity for working alongside machines. Research has shown that less experienced doctors sometimes lean too heavily on AI outputs, which can lead to more false positives or a kind of passive decision-making.

"How well they work depends heavily on who is using them. In primary care settings — often understaffed, undertrained, and far from the top-tier hospitals where these tools are first tested — that dependency can limit what AI actually delivers," Wu said.

A second challenge is the gap between lab conditions and the real world. Regulatory approval in China ensures that AI systems meet baseline standards under specific, controlled circumstances. But once a tool lands in a busy rural clinic or a community health center with aging equipment, outcomes can look quite different.

"How the system is integrated, how long it has been running, and how comfortable the staff is with it all shape performance in ways that no test environment can fully predict," said Mo Kai, a senior consultant at the Singapore-based regional association APACMed.

Some experts have begun arguing that governance needs to extend well beyond the moment of approval, into ongoing monitoring and adjustment.

Data governance adds yet another layer. Medical records are sensitive, and hospitals — understandably — are cautious about sharing them with outside technology companies. Encryption, data isolation and contractual safeguards help, but institutional trust builds slowly.

"For elderly care, all of these structural issues converge with a more basic question, that is 'can the people who need these tools most actually use them'," Mo said.

"Looking ahead, the next chapter of AI healthcare in China will probably be defined less by dramatic leaps in what models can do and more by the hard, unglamorous work of integration," he said.

"The important questions are: How do you fit AI into existing clinical routines? How do you make it work across a healthcare system that ranges from world-class urban hospitals to bare-bones village clinics? And how do you keep it accessible to people who did not grow up with smartphones?"

For a country where the population is aging at a pace that few systems have ever had to absorb, these are not abstract questions. The answers will shape not just the future of technology in medicine, but what it means to be cared for in old age.