From AI doctors to AI hospitals, how does Bauhinia Zhikang unlock the commercial space of medical AI?
In the past few years, the development speed of medical AI has been astonishing. AI can interpret medical images, answer professional questions, and even achieve scores higher than the human average in some medical examinations.
The technology is very advanced, but how can enterprises build a sustainable business from it?
Recently, medical AI enterprise Zijing Zhikang has completed a nearly 100 million-yuan Angel+ round of financing, with investors including Shanshi Capital, Ren'ai Capital and Inno Angel, and its valuation has reached billions of yuan.
Established in September 2024, this enterprise is incubated by the Institute for AI Industry Research (AIR) of Tsinghua University. Its technical source is the Agent Hospital project initiated by the team of Professor Liu Yang from Tsinghua University. In April 2025, Tsinghua University announced the establishment of an artificial intelligence hospital, with Agent Hospital providing core technical support, and Zijing Zhikang taking charge of the R&D and implementation of Agent Hospital.
From a single AI doctor to a complete system that can connect hospitals, doctors and patients, Zijing Zhikang is exploring a different technical path. Behind this path, there is also greater commercial opportunity for medical AI.
AI Doctor ≠ AI Hospital
However, if you actually walk into a real hospital, you will find that AI-enabled medical treatment is far more complicated than just having several rounds of conversations with AI.
A patient needs to go through multiple departments and contact different doctors from registration, triage, consultation, examination to treatment, follow-up consultation and follow-up visit, and information needs to flow between multiple systems. Even if AI can complete a certain link, it does not mean that the entire medical process has become intelligent.
The problem is not whether AI can see a doctor, but whether it can truly participate in the operation of the medical system.
Qiao Yuchen, Chief Product Officer of Zijing Zhikang, explained that healthcare is a highly interconnected ecosystem. If AI always stays at single-point applications, it is difficult to bring systematic changes. This is also the reason why Zijing Zhikang chose to build an "AI Hospital" from the very beginning, rather than a separate "AI Doctor".
In its product system, there are currently three application entrances: patient end, doctor end and hospital end. Patients can complete pre-consultation and upload examination reports before seeing a doctor; doctors can get medical record sorting, information analysis and diagnosis and treatment assistance; hospitals can carry out shift management, resource scheduling and data statistics.
However, the three-end products are only the visible surface layer. What Zijing Zhikang really wants to build is the medical AI infrastructure (Infra): to connect doctors, patients, hospitals and different medical agents at the bottom layer, so that they can share workflows and complete tasks collaboratively.
This is significantly different from traditional hospital information systems. In the past, hospital informatization mainly focused on data recording, management and circulation. In the AI era, the system has the opportunity to further understand information, analyze demands, and even actively participate in task collaboration.
Another key difference is that Zijing Zhikang does not try to bypass hospitals.
Different from general health consultation, serious medical treatment involves professional links such as consultation, prescription writing and treatment, which cannot be separated from doctors' practicing qualifications and responsibility judgment. Therefore, in the design of Agent Hospital, AI is responsible for information processing, auxiliary analysis and suggestion making, and the final medical decision is still made by real human doctors.
This human-AI collaboration method allows technology to enter the real medical process, and also gives Zijing Zhikang the opportunity to transform from an AI tool provider to a medical infrastructure R&D and operator.
AI Needs to Integrate into the Workflows of Hospitals, Doctors and Patients
Zijing Zhikang's technical route does not allow AI to enter real hospitals from the very beginning. Instead, it first built a "virtual hospital" in the computer.
In 2024, Professor Liu Yang's team from Tsinghua University released relevant research on Agent Hospital, trying to build a virtual medical world composed of different agents such as AI doctors and AI patients. Here, AI doctors can repeatedly receive simulated patients, experience different symptoms, disease courses and treatment processes, and continuously improve their medical capabilities.
There is a practical problem behind this: medical data is particularly precious.
Real patient data involves privacy and compliance, and the cost of acquisition and use is very high. Relying only on medical textbooks, papers and static medical records cannot cover the complex and changeable situations in real medical practice.
The SEAL technical route adopted by Zijing Zhikang tries to solve this problem through a dual-world evolution mechanism of "high-throughput training in the virtual world + companion evolution in the real world". To put it simply, it is like training a medical student: first learn knowledge at school and complete simulated exams, then enter the hospital to accumulate experience in actual work.
The same goes for AI. In the virtual world, medical agents learn through a large number of simulated diagnosis and treatment; after entering real hospitals, they start to work collaboratively with real human doctors, and the adoption, modification and correction of suggestions by doctors become the feedback for their continuous improvement.
Why add this step in the real world? Qiao Yuchen gave an example: for a patient with ordinary cough, AI that purely pursues excluding all medical risks may tend to suggest a large number of examinations. However, experienced clinicians will also comprehensively consider the probability of the condition, the necessity of examination, the burden on patients and medical costs.
A good medical decision is not just a stack of medical knowledge, but more importantly, an understanding of real scenarios.
Based on this idea, Zijing Zhikang further developed the "Doctor Avatar Agent".
Different from general AI doctors, the doctor avatar is associated with a specific real doctor, trying to learn his consultation style, diagnosis and treatment logic and risk preference, and continue to evolve under the supervision of the doctor. Its value lies in that part of the professional capabilities of excellent doctors are no longer completely limited by working hours and geographical location, so that the technology gradually moves to reality.
In 2025, the version 1 of Agent Hospital was released, and the system began to enter physical hospitals for testing. In August of the same year, the system launched internal tests at Tsinghua University Hospital. In April 2026, the online virtual consultation room function was open for testing, where doctors and medical students could carry out diagnosis and treatment training by interacting with AI patients and build their own doctor avatars. By October, Zijing Zhikang took another step forward: from a single doctor avatar to the collaboration of multiple medical agents.
In the newly released MHAC Multidisciplinary Human-AI Collaboration Framework, the avatar agents of doctors from different specialties can jointly participate in the same medical problem. For example, a hyperthyroidism patient also suffers from asthma. When the endocrinologist formulates the treatment plan, the respiratory doctor avatar can put forward medication risk reminders from another professional perspective, and the attending doctor will make a comprehensive final judgment.
In the past, a formal Multi-Disciplinary Team (MDT) consultation often required coordinating the schedules of many experts. Agent collaboration has the opportunity to advance part of the interdisciplinary support to the general outpatient service, or even the pre-consultation stage.
This means that the multidisciplinary support that was previously only available to a small number of patients may cover more medical scenarios in the future.
Of course, the newly released MHAC is still in the early stage of application promotion. What can better reflect Zijing Zhikang's actual implementation capability is the previous use of doctor avatars in real hospitals.
According to the data provided by the company, as of July 2026, the Agent Hospital system has been deployed and tested in 9 physical hospitals in Beijing, Guangxi, Wuxi and other regions. Among them, Beijing Tsinghua Changgung Hospital has carried out outpatient tests in 18 clinical departments, and more than 40 doctors use it on a regular basis.
According to the statistics of in-hospital data, the overall diagnosis and treatment efficiency of doctors has increased by about 30%; after the doctor avatar agent collaboratively processes 200 cases, the full or partial adoption rate of diagnosis and treatment suggestions exceeds 90%.
A neurosurgeon interviewed said that the system has obvious help in medical record sorting, preoperative discussion records and information aggregation, and the workload of related records can be reduced by nearly half according to his personal experience.
From simulated training to real diagnosis and treatment, and then to multi-agent collaboration, Zijing Zhikang has formed a technical path that gradually penetrates into medical scenarios.
When AI begins to truly integrate into doctors' workflows, the value of a system is no longer just the capability of the model itself.
Who Pays for Medical AI?
If the medical AI industry in the past two years mainly focused on proving what the technology can do, the problem worthy of more attention in the next stage is who is willing to pay for these capabilities.
The business model of traditional medical software is relatively simple: selling systems to hospitals and charging for procurement, deployment and maintenance. However, what Zijing Zhikang targets is a medical service ecosystem involving multiple participants such as hospitals, doctors, patients and pharmaceutical enterprises.
This means that future revenue sources may not be limited to a single software procurement order.
The first type of payer is hospitals. According to the introduction of Zijing Zhikang team, in the early stage of commercialization, the company will explore a sustainable business cooperation model by carrying out scientific research cooperation, AI scenario construction and technical services with the government, hospitals and medical institutions.
For hospitals, the direct value brought by AI is efficiency and service capability. If doctors no longer need to spend a lot of time sorting out medical history and entering medical records, they will have the opportunity to devote more energy to patient communication and medical judgment. Efficiency improvement can also be converted into incremental service capacity. With the same consultation time, doctors have the opportunity to serve more patients; hospitals can improve the supply of medical services while controlling human resource input.
For large tertiary hospitals, this means further optimizing existing resources; for primary hospitals, it may mean obtaining professional support that was difficult to access in the past.
The second type of payer is doctors. Traditional medical services are highly dependent on doctors' personal time. Even if a well-known expert works more than ten hours a day, the number of patients he can receive is still limited. The doctor avatar provides another possibility. In the future, primary doctors can call the avatars of high-level experts through authorization to assist in diagnosis and treatment analysis; after patients are discharged from hospital, they can also carry out continuous follow-up visits through the system. The professional capabilities of excellent doctors have the opportunity to extend from face-to-face diagnosis and treatment to more times and locations.
As a result, new charging models may be formed around scenarios such as doctor avatar call, medical institution service procurement, and remote collaboration.
The third type of payer is patients. From pre-consultation and health records to post-discharge follow-up, AI can enable patients to establish more continuous connections with medical services.
In the past, patients often received relatively concentrated attention from doctors only when they walked into the consultation room. Medical AI has the opportunity to extend part of the services to before consultation and after discharge.
This continuous service capability also provides space for personalized health management and value-added services in the future.
Outside hospitals, commercial value can be further extended.
For example, pharmaceutical enterprises. Clinical trials have long faced problems such as difficulty in patient recruitment and complex multi-center collaboration. Especially for some patients distributed in primary medical institutions, traditional recruitment methods can hardly cover them efficiently. When Agent Hospital connects more medical institutions in the future, under the premise of meeting data compliance and research ethics requirements, it will have the opportunity to help pharmaceutical enterprises improve the efficiency of patient screening, clinical trial recruitment and real-world research. Its commercial value is no longer just providing an algorithm, but participating in an important service link of pharmaceutical R&D.
Another direction is medical education. Virtual consultation rooms and simulated patients can be used for doctor training, diagnosis and treatment capability assessment and medical teaching, so that the underlying technology can be further extended to non-clinical service scenarios. There is a common logic behind these businesses: Zijing Zhikang tries to transform medical AI from a single function into a basic capability that can be called by different industrial participants.
For the capital market, this platform-based business logic also provides a new valuation perspective.
The growth of a traditional software company usually depends on the number of customers, software procurement and renewal revenue. Once the medical AI infrastructure forms a large-scale application network, potential revenue can extend to more participants and service scenarios. The more hospitals it connects, the wider the coverage of doctors and patients, the richer the medical tasks and ecological cooperation opportunities it can undertake.
This is also an important reason why industrial background investors pay attention to medical AI.
In this round of financing, both Shanshi Capital and Inno Angel, as old shareholders, chose to continue to increase their investment.
Among them, Liu Xiaolei, partner of Shanshi Capital, has long focused on the changes brought by the development of AI technology to the medical industry. He believes that Zijing Zhikang, with the technical path of "virtual hospital + evolvable agent" and backed by Tsinghua's disciplinary and clinical resources, is expected to reshape medical supply and alleviate the shortage of primary doctors, with huge social value and commercial opportunities. Investors from Inno Angel are also optimistic about the AI-native medical global ecosystem created by Agent Hospital technology. The newly released doctor avatar agent of the company helps human doctors break through the restrictions of time, space and language to achieve large-scale services, realizing the sinking and equalization of high-quality medical resources, making it possible for patients in remote and underdeveloped areas to enjoy the highest quality medical services.
From this perspective, the capital's judgment criteria for medical AI are also changing. In addition to the model capability itself, whether the technology can continue to evolve, whether it can be stably transformed into products, and whether the team has the ability to expand scenarios in the long term are becoming more important considerations.
Looking at the international market, medical AI is also becoming a direction of capital attention. Take the US company OpenEvidence as an example, it cuts into doctors' workflows through medical evidence retrieval and auxiliary decision-making, and explores a business model connected with the pharmaceutical industry.
Although different enterprises choose different paths, they are all trying to answer the same question: how can medical AI move from demonstrating technical capabilities to truly participating in medical services and creating commercial value.
When more and more medical agents enter real consultation rooms, the future competition of medical AI may no longer only focus on who has the smartest AI doctor. More importantly, who can connect these agents and make them truly part of the medical system.
And the commercial value of a medical AI company may also be redefined accordingly.