Summary of Key Points
Recently, AI in healthcare has once again become a hot target for capital investment: SenseTime Medical has a valuation of over ten billion and is preparing for an IPO; ByteDance has invested 6 billion yuan in building an “AI-based hospital”; giants such as Tencent, Ant Group, and Baidu are also increasing their investments. The market size has exceeded 150 billion yuan, with a growth rate several times that of the overall healthcare industry. However, this is not the first time such enthusiasm has emerged. In 2021, four prominent companies attempted to go public through IPOs, but only one, YingTong Medical, succeeded, only to see its stock price plummet, highlighting the persistent issue of failing to generate profits and continuous losses. The difference this time is that advancements in large-scale modeling technology have transformed AI from a “tool” into a “assistant,” and policies have begun to support the inclusion of AI-assisted diagnosis in medical insurance. Nevertheless, the industry is still stuck on the most critical question: who will bear the cost of AI in healthcare?
What Can AI in Healthcare Do Today?
AI in healthcare does not aim to replace doctors; rather, it helps doctors or patients solve specific problems. Currently, there are four main areas of application:
1. Assisted diagnosis (most mature): For example, using AI for image analysis, where large amounts of labeled CT scans and retinal images are fed into algorithms to recognize patterns of diseases. These algorithms can then highlight suspicious areas in new images and accurately measure the size of lesions (such as the volume of a cerebral infarction, which is difficult for doctors to estimate manually). More than 70% of tertiary hospitals use such tools, but they are still not very effective at identifying complex lesions; essentially, it’s a form of “advanced image processing.”
2. AI in drug development (largest potential): In the past, developing new drugs took over a decade and cost billions of dollars, with a high failure rate. AI can simulate targets and molecular structures on computers, reducing the time needed for trial and error. Currently, there are over 170 AI-driven drug development projects in clinical trials, but none have been fully designed by AI. The human body is too complex, and even with precise calculations, the actual effects still need to be verified through clinical trials.
3. Disease treatment (highest barriers): For instance, surgical robots allow doctors to control the movements of robotic arms with precision (more stable than human hands), but the equipment is expensive and requires extensive training, making it unaffordable for smaller hospitals. Remote surgeries are even more challenging, as a single second of network delay can be dangerous.
4. Health management (most similar to internet services): Platforms like Ant Group’s Afu and JD Health’s AI consultation services receive millions of inquiries daily, but users are reluctant to pay. There is also the issue of overestimating the severity of minor illnesses or underestimating the seriousness of serious ones—for example, a cold might be listed as a potential serious condition, but a real tumor might go unnoticed.
Three Types of Players Competing in the Market, Each with Their Own Strategies
The participants in this market can be divided into three groups, each with clear strengths and weaknesses:
1. Large technology companies (rich in funds and traffic, but lacking medical expertise): Companies like ByteDance invest in AI hospitals, and Ant Group develops consumer health apps (with 30 million monthly active users) and provides cloud services. Their challenge is that healthcare is a slow-moving industry, and these companies, accustomed to fast-paced operations, often struggle to adapt. For example, ByteDance’s Xiahe Health was restructured. Industry insiders say that large companies’ contributions to AI in healthcare are minimal; they either focus on booking services or rely on their e-commerce traffic to generate revenue, without truly integrating into clinical practices.
2. Medical device manufacturers (with distribution channels and regulatory approvals, but limited technical capabilities): Companies like United Imaging Technology sell medical equipment with built-in AI capabilities (such as CT scanners with AI analysis) and have obtained 20 of the highest-level regulatory approvals, entering over 4,000 hospitals. Their advantage is their ability to bundle AI with hardware sales, but their large-scale modeling technology is not as advanced as that of large technology companies.
3. AI-focused companies (expert in algorithms but struggling to make profits): Companies like SenseTime Medical (pre-IPO) and Shukun Technology hold multiple regulatory approvals. They rely on their models and data for revenue, but many of their products rely on data from hospital systems, which hospitals are reluctant to share. Additionally, pure software products are not highly profitable due to competitive pricing from hospitals.
The Most Critical Question: Who Will Pay for AI in Healthcare?
Despite the large market size, not much actual money flows into the companies’ pockets. Revenue mainly comes from the following sources:
1. Medical insurance (most anticipated but yet undefined): Last year, the National Healthcare Security Administration introduced an “extension” for AI-assisted diagnosis, but the details on how it will be reimbursed are still unclear. Once finalized, medical insurance could become a stable source of revenue.
2. Hospital purchases (most direct but with significant price pressure): AI-based imaging and diagnostic software are sold to hospitals, but hospitals often negotiate prices aggressively, sometimes below cost.
3. Collaborations with pharmaceutical companies (clearest but limited to AI in drug development): AI companies help pharmaceutical companies speed up the drug development process, and these companies are willing to pay (for example, Innovent Biologics signed a deal worth 7 billion dollars). However, this is only the case for a few companies.
4. Consumer payments (many users, but low willingness to pay): Health management apps have a large user base, but most users are accustomed to free services, and few are willing to pay for additional health check-ups.
Currently, only two types of companies are making profits: those that bundle their software with medical equipment and those that collaborate with pharmaceutical companies. Pure software companies are finding it increasingly difficult to survive.
What Makes This Round Different from the Last One?
The 2021 boom was dominated by image AI, but there are two key differences this time:
1. Technological advancements: Advances in large-scale modeling have improved the generalization and multimodal capabilities of AI, turning it from a single tool into a comprehensive assistant—capable of analyzing medical records and answering patient questions.
2. Policy support: Medical insurance has begun to support AI-assisted diagnosis, and local governments are exploring new applications. The lack of clear policies in the previous round made the commercialization path more uncertain.
However, the same old problem remains: can AI products become profitable on a large scale? Many of the companies from the previous round are still losing money. If the issue of payment is not resolved this time, the same mistakes could be repeated.
Where Does AI in Healthcare Go from Here?
Industry experts believe that the winners will be companies that combine software and hardware, serve all hospital departments, and create an ecosystem that includes both in-hospital and out-of-hospital services. For example, selling equipment with integrated AI and offering commercial insurance and health management services can create a self-sustaining business model. Companies that rely on a single product (such as pure software) are likely to be phased out. Healthcare is not like the internet; it requires practical solutions to clinical problems and user acceptance, which takes time.
In summary, AI in healthcare is not a short-term trend but a long-term opportunity. While technology and policy are progressing, real profitability depends on the clarification of medical insurance regulations and the development of user habits.