On March 27, XUNFEIHEALTH (02506) released its 2025 annual performance report. The company achieved revenue of 915 million yuan, a 25% year-on-year increase, with a contract value of 1.45 billion yuan, up 55% year-on-year. Net losses narrowed by 52%. These figures represent a strong performance against the trend in a year when the overall healthcare informatization industry faced pressure. A breakdown shows revenue from government-end primary-level solutions grew by 20%, hospital solutions for the business sector increased by 31%, and patient management services for the consumer end rose by 29%. The simultaneous positive growth across all three segments is quite unique among medical AI companies.
Beyond the financial data, the annual report is noteworthy for reflecting industry changes: long-standing issues in China's healthcare system, previously constrained by technology and cost, are now becoming solvable at scale due to AI, especially the rapid evolution of large language models. The corresponding industrial opportunity is likely far larger than XUNFEIHEALTH's current revenue scale.
The primary healthcare sector in China is not lacking in institutional design or financial investment. The per capita funding standard for national basic public health services has gradually increased from 15 yuan in 2010 to 99 yuan in 2025, with the total national budget reaching 140 to 150 billion yuan. The family doctor contract system has been promoted for nearly a decade, with clear policy frameworks for tiered diagnosis and treatment, chronic disease management, and health record construction. However, the implementation effectiveness of these systems has long been constrained by a simple reality: human resources are insufficient. A single primary care doctor is responsible for hundreds of contracted families; completing the full suite of services—including record establishment, follow-ups, health education, and chronic disease management—using traditional methods creates a workload far beyond individual capacity. The phenomenon of "signed but not served" family doctor contracts is not due to negligence but stems from inherent human resource limitations.
Hospitals face similar bottlenecks. Of the approximately 300 million annual hospital discharges, over 60% involve patients with long and complex recovery cycles. Post-discharge medication guidance, rehabilitation management, and risk warnings are essential needs, but it is impractical for hospitals to conduct continuous follow-ups manually for every discharged patient due to cost and staffing constraints. Chronic disease management accounts for about 80% of medical insurance expenditures, yet the daily management of most chronic disease patients outside the hospital remains largely unaddressed.
The demand has always existed, and investment has been increasing. The real variable is technology. The rapid development of large language models over the past two years has enabled services that were previously highly dependent on professional manpower—such as health profile generation, intelligent follow-ups, personalized rehabilitation guidance, and round-the-clock health consultations—to become feasible for low-cost, large-scale delivery. As technology becomes more reliable and user-friendly, initiatives that were "always desired but unaffordable" can finally be launched. This is also the context for accelerated policy promotion. The "15th Five-Year Plan" outline for the first time includes provisions to "promote high-level resident health assistants and expand the application of intelligent assisted diagnosis in primary medical institutions." The National Health Commission requires basic intelligent assistance for primary-level diagnoses to achieve near-universal coverage by 2030, with AI-assisted diagnosis becoming common in secondary hospitals and above. Industry analysis indicates that the AI transformation market for hospitals alone is expected to grow nearly tenfold within five years. The combined opportunities from primary-level intelligence, hospital AI transformation, and consumer-end health management conservatively estimate an industrial space worth hundreds of billions. This market has not emerged out of thin air; rather, long-accumulated demand has finally met technologically capable and cost-feasible solutions.
The market is opening, and the number of participants is rapidly increasing. Over the past year, the AI health assistant sector has seen concentrated growth, with large model companies, internet platforms, and startups entering the field. The habit of "consulting AI first for minor and major illnesses" is becoming common among some users. However, medical AI fundamentally differs from other AI application scenarios: it requires both "sufficiently affordable" and "sufficiently reliable." Technological advancements can drive down costs, addressing the "affordability" issue, but the "reliability" challenge is far from universally solved.
XUNFEIHEALTH's Chairman Liu Qingfeng noted in an interview that some large model products "merely stitch together health information from the internet, appearing coherent but lacking medical logic, diagnostic standards, and evidence-based foundations." His judgment is: "The logic of fast-moving consumer goods cannot be applied to healthcare; we oppose a traffic-oriented approach that neglects the bottom line of doctors and medicine." This points to an industry-wide issue. Creating an "AI that can discuss health issues" is not difficult, but developing a "medical AI with true specialist-level capabilities that can assist clinical decision-making and be responsible for user health" involves a significant gap. Real medical scenarios are extremely complex; patients do not present illnesses as described in textbooks, often having multiple coexisting conditions and intersecting medical histories. AI must possess evidence-based reasoning capabilities, not just information retrieval and stitching.
From a technical perspective, the current medical AI landscape is diverging into two types of participants: one group enters from the consumer end, rapidly building health consultation products based on general large models; the other group enters from within the medical system, refining vertical model capabilities in real diagnostic scenarios before gradually expanding outward. The two paths face different challenges. The former's challenge lies in transitioning from "usable" to "reliable," while the latter's challenge is scaling efficiency from "reliable" to "large-scale."
XUNFEIHEALTH has taken the latter path. Since its establishment in 2016, it has engaged directly with clinical scenarios. It passed the National Medical Licensing Examination in 2017, piloted its Intelligent Medical Assistant in primary care settings in 2018, and has since expanded to higher-level hospitals and the consumer end. Nearly a decade of accumulation along the clinical route forms the foundation for its narrative in the billion-dollar market, but it also dictates that its growth rhythm will not be an internet-style explosion; rather, it is a process requiring patience.
Among XUNFEIHEALTH's three business segments, the hospital solutions for the business end showed the highest growth rate and strongest certainty in 2025, with revenue increasing by 31% year-on-year. By the end of the reporting period, it served over 600 level-based hospitals, including more than 50 among China's top 100 hospitals and 7 among the national top ten. Over 50 new top-tier tertiary hospitals were added during the year. This growth is set against the industry trend of hospital informatization transitioning towards intelligence. XUNFEIHEALTH's President Tao Xiaodong mentioned that after the DeepSeek hype in the first half of 2025, many hospitals installed large models but were often unsure of practical applications afterwards. Companies with a deep understanding of hospital workflows and core diagnostic scenarios gained more opportunities after the initial excitement subsided.
XUNFEIHEALTH's hospital products operate on three levels. At the smart service level, AI follow-ups achieve 100% coverage of discharged patients, improving patient satisfaction. At the smart medical level, automated medical record generation saves doctors an average of 2.8 hours daily on documentation; connotation quality control improves medical record quality scores from 7.87 to 9.08; and the AI diagnostic assistant increases treatment rationality rates from 73% to 91% through interdisciplinary recommendations. At the smart management level, a big data intelligent platform helps hospital decision-makers improve efficiency by over 50%.
Building on these foundational capabilities, XUNFEIHEALTH is developing an operational business in hospitals: post-discharge patient management. This addresses a specific gap: rehabilitation management for patients after discharge has long been a vacuum, with doctors willing but unable to provide continuous support, and patients lacking professional guidance. XUNFEIHEALTH's AI automatically extracts patient data generated during hospitalization to create personalized rehabilitation plans (confirmed by doctors), using intelligent outbound calls and app notifications to replace most manual follow-ups. This helps patients receive continuous guidance on medication, diet, exercise, and rehabilitation at home. Patients pay for service packages, allowing doctors to extend services beyond the hospital without increasing their workload.
Qilu Hospital serves as a benchmark case. Starting with just 6 specialized disease service packages in Q1 2025, it expanded to over 70 packages by Q3, covering major departments like cardiology, orthopedics, oncology, and perinatal care. With an annual discharge volume of approximately 400,000 patients, 50%-60% have management needs, potentially generating 8-10 million yuan in annual revenue per hospital. These specialized service packages are reusable across hospitals, with marginal costs decreasing as scale increases. Liu Qingfeng defines this type of business as a "base area" model. After the initial phase, it becomes a long-term collaboration, allowing for sustainable expansion into more disease areas and creating opportunities for research cooperation. For hospitals facing increasing operational pressures, this represents a cooperative model that enhances patient satisfaction while generating incremental revenue. Across the industry, tertiary hospitals handle about 150 million discharges annually, yet current penetration rates remain low.
The Intelligent Medical Assistant is XUNFEIHEALTH's most deeply entrenched product. It covers 806 districts and counties across 31 provinces, serving over 77,000 primary medical institutions. It processes 1.7 million auxiliary diagnoses daily, with a cumulative total exceeding 1.1 billion. Public information suggests no other provider has achieved routine deployment on a similar scale in primary care intelligent assistance. After reaching 806 districts and counties, XUNFEIHEALTH is extending its reach along the government-to-consumer path. Building on the diagnostic assistance capabilities of the Intelligent Medical Assistant, it adds AI health profiles and the Xunfei Xiaoyi app, transforming family doctor contracts from mere "signatures" into "continuous services." On one end, large models integrate residents' health records to generate intuitive, readable health profiles; on the other end, they connect to family doctors' mobile workstations, assisting with contract management and personalized follow-ups. The Xunfei Xiaoyi app reaches residents directly, providing health profile access and 24/7 consultation.
Pilot data preliminarily validates this approach. After deployment across 110 community health centers in Shenzhen's Longhua District, primary care doctor coverage reached 90%, with a 38% monthly active rate among registered residents. Contract signing rates increased by 5%, renewal rates by 9%, and fulfillment rates by 21%. Following a successful health profile pilot in Shanghai's Jing'an District, the initiative expanded to Xuhui and Changning districts, deployed across over 1,000 outpatient doctor workstations, generating health profiles for 1.54 million residents, with doctors accessing them over 12 million times. These pilot results demonstrate that when AI enhances service capability to a usable level and reduces marginal costs sufficiently, existing primary care funding and staffing frameworks can support far richer services than before, thereby improving resident perception.
It is important to recognize that these are still pilots in limited areas. Scaling from a few model districts to over 2,800 counties nationwide involves variables like local policy pace, financial capacity, and primary-level execution differences. While the government-to-consumer direction has received preliminary validation in pilots, large-scale replication from points to areas is still underway.
The Xunfei Xiaoyi app is a core product in XUNFEIHEALTH's consumer-end strategy, with over 30 million cumulative downloads and more than 180 million AI health consultations completed. In evaluations conducted by the Shanghai Medical Large Model Application Testing and Verification Center, the Spark Medical Large Model received straight A's in all four assessment categories. Unlike many health AI products, Xunfei Xiaoyi's user base primarily flows from two natural entry points: residents within the government-end family doctor contract system and rehabilitation-phase users from the business-end hospital patient management programs for tertiary hospitals.
Tao Xiaodong cited a case where a cancer rehabilitation patient interacted with Xunfei Xiaoyi over a thousand times within a year, noting it provided "not only professional medical assistance but also non-medical comfort." When usage scenarios involve real treatment plans and rehabilitation management, user stickiness is on a different level compared to general health information products. However, consumer-end commercialization remains in its early stages. As the resident engagement tool within the Government-Business-Consumer ecosystem, Xunfei Xiaoyi's independent, scalable revenue model is still under exploration.
Although the youngest segment, the consumer-end business, together with the government and business ends, forms a complete data flywheel for XUNFEIHEALTH. Over 1.7 million real-world diagnostic data points flow back daily from primary care and hospitals, continuously training the Spark Medical Large Model. In 2025, the total volume of high-quality training data grew by 30%, with the medical knowledge base approaching 1 billion entries. The large model achieved four key capability upgrades during the reporting period, becoming the industry's first to reach the level of a department head physician in tertiary hospitals within specialized AI fields, and the only fully self-developed, controllable medical large model trained entirely on domestic computing power.
Liu Qingfeng emphasized the significance of this flywheel: "The generation of 1.7 million new auxiliary diagnosis data points daily creates a powerful data flywheel, which is unique to our organization." In an era of rapidly iterating agent models and lowering application development barriers, the capability of the underlying base model itself becomes a scarcer resource. XUNFEIHEALTH's positioning targets precisely this layer. In Liu Qingfeng's words, "XUNFEIHEALTH is not a medical software application company; we are a company building the foundational medical large model." This structural shift is reflected in the 2025 annual report: the company is transitioning from a project-based entity "installing systems for primary care" to a platform company centered on a medical large model, integrating Government, Business, and Consumer segments.
A company with annual revenue just over 900 million yuan faces an industrial opportunity worth hundreds of billions. This gap represents both potential and an execution challenge. The 55% growth in contract value outpacing the 25% revenue growth, with ample orders in hand but revenue recognition lagging, is typical during a transition from project-based to operational business models. While losses narrowed by 52%, indicating convergence, the path to breakeven remains. The business-end hospital solutions are beginning to demonstrate replicable models. The government-end primary care segment has deep roots and wide coverage, but business-to-consumer and government-to-consumer initiatives are still in the pilot phase. The consumer-end has a clear differentiated path, but independent, scaled commercialization is still exploratory.
Healthcare is an industry requiring extreme patience. XUNFEIHEALTH has chosen a path grown from clinical scenarios. Its decade-long primary care network, data积累 from 1.1 billion consultations, and continuously iterated base model are assets not easily replicated short-term, but converting them into large-scale revenue is not an overnight task. The hundred-billion-yuan market will not open suddenly. It awaits continuous technological improvement, ongoing policy support, and participants with sufficient depth in real-world scenarios to activate it step by step. Nevertheless, it is evident that XUNFEIHEALTH is currently one of the companies furthest along this path.