Tech giants are spearheading an AI revolution in the medical field. Who has a better chance of winning?

In the past two years, in the field of large models, domestic and foreign technology giants have continuously accelerated their layout in the medical track. Wang Xiaochuan even publicly claimed that healthcare is the "pearl on the crown of large models."

The demand for AI in medical scenarios is indeed high: tens of thousands of drugs, the cumbersome and repetitive work of reading medical images, and even thick stacks of professional books and clinical guidelines... Using the capabilities of "silicon - based beings" to solve problems such as "high - cost and difficult access to medical services" and "uneven distribution of medical resources" among "carbon - based beings" undoubtedly represents the future of technology.

For this reason, since the earliest "Internet +" era, from medical imaging to pathological diagnosis, from surgical robots to medical large models, the investment boom and market attention triggered by AI in the medical field have never diminished.

Objectively, the actual implementation of medical AI depends on the combined efforts of multiple dimensions such as technology, products, patient and doctor education, and policy supervision. This is also the reason why many companies under the concept of medical AI have failed to break through the commercialization bottleneck in the past 10 years.

Now, as large - model technology breaks through boundaries again, many problems that have troubled medical AI companies in the past are expected to find a new solution. The increased layout of major companies may just be the beginning.

In this medical AI "revolution" triggered by large models, who is leading the way? Who can succeed?

I. What medical problems can generative AI potentially change?

From grassroots health centers to top - tier tertiary hospitals, hospitals are scrambling to deploy their own large - model applications, which can be regarded as the first "spectacle" in the medical industry after the Spring Festival in 2025.

Frankly speaking, although there are some doctors who are willing to embrace new technologies, the "conservatives" still make up the majority. Compared with asking AI for help, they trust more in the "true knowledge" obtained from hard study and clinical practice. So, how was this perception broken?

Han Wei is now the director of the information department of Beijing Hospital of Traditional Chinese Medicine Affiliated to Capital Medical University. Recently, they have accessed Ant's medical large model, which has the ability of medical thinking reasoning and multi - modal interaction. "It can not only provide solutions but also present the thinking process." "This is of special significance to doctors because what we want is not just a result, but also hope to refer to its thinking process. In the future, when encountering similar problems, we can think according to this process."

He gave an example of a patient who had lost consciousness and could not be diagnosed with a heart attack. Usually, when the hospital received such cases, it could only convene multi - disciplinary consultations, which had problems such as long time - consuming and low efficiency. But this time, based on the patient's previous medical records and history, the large model gradually completed the condition analysis and diagnosis suggestions and also prompted potential risks.

"The whole thinking process and results are basically the same as the diagnosis and treatment suggestions of the two real - life experts we consulted. It shows that the large model can indeed save us some consultation steps and allow doctors to focus more diagnosis and treatment time on patients." Han Wei explained.

Behind this experience of "AI being more useful", it mainly benefits from the technological upgrade of large models. A practitioner in the medical large - model field introduced that if comparing two generations of AI technologies, the medical AI in the previous 2.0 era was more like "discriminative AI", which was better at making medical diagnoses based on medical images and deduced data. In the 3.0 era based on generative large models, it is creating a "sequence prediction model", that is, it can predict future health status and disease development based on an individual's historical health data sequence, and then form an individual's health development trajectory. And this trajectory is "extremely valuable for personalized treatment and precise health management."

In other words, in the past, although people could feel the upgrade of medical services brought by AI (such as AI pre - consultation), the experience always seemed to be lacking something because the supply of medical services had not really changed, and the core workforce was still doctors. Now, the "productivity revolution" brought by large - model technology allows AI to start to penetrate into the front - end of medical care, and expand the entire supply of resources through higher - quality auxiliary capabilities.

"Some changes seem small, such as a 10 - minute increase in the efficiency of remote consultations, but for patients, it means saving more than an hour. Through in - depth cooperation with medical service practitioners in the past, we have become more certain of the value of AI in healthcare." Ant explained.

After more than 10 years of exploration, technology companies have finally found the "balance point" between AI technology and human doctors, rather than "replacing doctors" or "radically reforming healthcare." That is, by promoting the expansion of medical resources and the popularization of services, such as quickly narrowing the diagnosis and treatment gap between hospitals at all levels and improving the efficiency of doctors in reading new literature, researching cases, and managing patients, everyone can have their own high - quality "AI personal doctor."

Especially this year, the popularity of DeepSeek has carried out an "AI usage education" among professional medical institutions and the general public in a surging manner, making it possible for AI to penetrate deeper into medical scenarios.

This potential has injected confidence into major technology giants and Internet companies to "bet" on medical AI again.

II. The key to victory: the battle for medical resources

After large - model technology has brought enough confidence to practitioners, the next question arises: how to make their products stand out among many competitors?

In the many explorations and narratives in the field of medical AI, one constant is the competition for medical resources centered around hospitals. Training models requires high - quality medical data such as diagnosis and treatment records and imaging reports. Cultivating the "doctor" thinking of models requires targeted training in specific departments... In other words, the more high - quality hospitals a company covers, the higher its chances of winning.

Among the old and new players, for established medical informatization enterprises and equipment manufacturers representing the old players, the channel advantages accumulated in the past naturally become their biggest trump card. Many technology companies, as new players, choose to use their advantages in computing power and algorithms to cooperate with top - tier hospitals to develop large models to make up for this shortcoming.

However, in - depth cooperation with hospitals takes time. For either party involved, the action of covering hospitals is actually slow and limited. Is there a more flexible way?

In many cases, Ant's "strengths" may be slightly different. Ant's earliest experience in Internet healthcare can be traced back to 2014 when it launched Alipay for registration and payment at Guangzhou Women and Children's Medical Center. As of now, through the integration of the payment link, Ant has cooperated with 3,600 hospitals and served more than 800 million users. To this day, it is still difficult to say that there is a second Internet healthcare company that can replicate this.

One end of the scale faces the hospitals: Usually, hospitals only open a small port for cooperation with software companies, such as for a specific department. But in the "overall link" of payment, the entire hospital system process must be opened to Alipay, which means a deeper cooperation foundation with hospitals. This also lays a certain foundation for Ant to introduce products such as large models that can intervene more deeply in medical services into hospitals.

The other end of the scale faces the users. The Internet is one of the most important channels for patients to obtain medical information. However, although there are many conventional search platforms and content platforms, few can provide a high - quality user experience. Until today, there has not yet been a fully - fledged and trustworthy Internet healthcare platform in China.

Relying on the unique implementation channel provided by Alipay's "super - platform", Ant has actually completed the market education for users to use Alipay to solve health problems one step ahead.

Now, many patients are already used to not bringing their medical insurance cards and can use Alipay to check in, wait for consultation, pay for medical services, and get medicine in hospitals. In the "Healthcare" module on the Alipay homepage, more than 90% of tertiary hospitals across the country have been aggregated, covering more than a hundred services such as buying medicine and having physical examinations, providing users with a more convenient health service experience.

Ultimately, in the "battle" for core medical resources, all participants are actually using the channels they have previously established to do what they are better at. Among many players, Ant, with its long - term in - depth exploration in the field of Internet healthcare, may become a model of "comprehensive layout and in - depth participation."

III. Is linking hospitals, doctors, and users to build an "ecological barrier" a new way to solve the problem?

After mastering key medical resources, what's next?

In the past, the problem with many medical AI companies when developing products was that, limited by experience and resources, they often started from a single point, such as developing AI for reading radiology images and providing auxiliary diagnosis in the radiology department. However, the entire in - hospital medical system is large and complex, involving many chains. This "single - point entry" method can only solve limited problems.

For this reason, we can see that some medical AI concept companies and large - model startups have indeed managed to break through in the market through technological or business - model innovation in the short term, but it is difficult to build long - term barriers. In the highly competitive environment of medical AI, a rich product matrix and an ecological "moat" may be the more difficult - to - replicate key to victory.

In this regard, Ant also told 36Kr that the large model is the product, and in many scenarios, functional points can be solved in one dialog box. However, in the medical industry, it is difficult for one product to "dominate the market." AI can do different things in different scenarios, and the logic of meeting needs is also different. Therefore, "when we started researching medical large models in 2023, we decided to cooperate deeply with medical institutions and intervene in the complete traditional medical scenarios."

Ant announced the integration of existing resources and the implementation of a "three - terminal integration" strategic layout based on the medical large model to upgrade the three major product systems for hospitals, doctors, and users, precisely for this reason.

The so - called "three - terminal integration" means launching a full - stack solution of a "large - model all - in - one machine" that can be directly deployed by medical institutions; providing an AI doctor assistant tool for 280,000 registered doctors on its Haodf platform, offering services such as literature retrieval and research assistance; and a "AI health butler" for users.

This layout is not blindly expanding the scope. Each item targets a pain point in a current medical link. Taking the "AI health butler" as an example, this service is mainly aimed at meeting the daily medical service needs of ordinary users, such as finding doctors, reading medical reports, and accompanying patients during consultations. Since its launch in September last year, the product has served 40 million users in more than half a year.

It is worth mentioning that throughout the process, Ant does not work alone but invites industry partners for in - depth co - construction. For example, in terms of hardware deployment, Ant has joined hands with manufacturers such as Huawei and Alibaba Cloud to launch a lightweight design of "integrated training and inference, ready - to - use out of the box." Beyond the large - model technical capabilities, it collaborates with medical institutions to achieve innovation at the application level. For example, the official AI health application "Anzhen'er" jointly launched with the Zhejiang Health Commission now covers more than 1,000 public hospitals and serves more than 30 million person - times.

In the past, it was difficult to succeed alone. Now, can platform - type enterprises relying on the unique path of "scenario + technology + open ecosystem" layout create a future for medical AI?