Why do Microsoft, Huawei, and Ant do not want to miss out on AI healthcare?
Why do Microsoft, Huawei, and Ant do not want to miss out on AI healthcare?
New Berry Daybreak
follow with interest
2025-03-27
0 comments
756 views
2 collections
12 minutes
B-end products need to rely more on sales teams and channel cooperation to promote their products, while C-end products need to make more use of online marketing and word-of-mouth communication to promote their products
From Microsoft, Huawei to Ant, these tech giants are investing heavily in trying to get a piece of the pie in this hopeful blue ocean. But the question is, what magic does AI healthcare have that can make these giants so obsessed?
The power of big models no longer needs to be popularized, now it's time to talk about applications. Thousands of industries actively seek change, among which the medical industry ranks first in terms of application speed, breadth, and depth.
According to an industry statistics report, by the end of 2023, the top three industries in China in terms of the distribution of large models are healthcare, finance, and scientific research. Robots have not yet appeared on the Spring Festival Gala stage, but have already been on the operating table.
At the beginning of 2025, DeepSeek emerged, further catalyzing the commercial landing of AI healthcare. The two core reasons are affordable prices and the ability to deploy open-source models privately, which better meet the security needs of sensitive medical data. According to incomplete statistics, as of now, over a hundred tertiary hospitals in China have officially announced the completion of DeepSeek localization deployment.
Most importantly, after the last wave of Internet medical education, C-end users seem to be more accepting of AI assistant diagnosis and treatment.
So, foreign giants such as Nvidia and Microsoft are investing heavily, while domestic giants such as Huawei and Ant continue to increase their investment. AI healthcare has become the most desirable application scenario for big models.
01 AI Medical Evolution Direction
When large model manufacturers enter the medical field, their problem-solving approaches are not limited. Companies with a certain foundation in medical services may even be more radical.
For example, Huawei's 21st Army, which was formed in early March, is the Medical and Health Army. Focus on building an AI assisted diagnostic solution system in real-time, and promote the application of medical models in clinical scenarios.
Later, they frequently launched pathological models and acute chest pain models in conjunction with different hospitals, and actively promoted all-in-one machine solutions with Internet companies. This includes ants with the same deep layout of AI healthcare.
After adopting AI healthcare as a deterministic strategy, Ant Group is almost simultaneously focusing on three aspects - institutions, healthcare providers, and users. Few domestic players have such a comprehensive and in-depth layout, which indirectly demonstrates Ant's ambition and determination to target medical AI.
Alipay is the foundation for ants to do medical treatment. In the stage of Internet health care, Alipay has built awareness in hospital institutions and clients in registration, consultation and payment. Accumulated in the past 11 years, it has radiated to 3600 hospitals across the country, serving over 800 million users. It is the largest medical insurance payment service platform in China and also a one-stop medical and health service platform.
In the AI stage, the meaning of Ant Medical is actually expanding, from infrastructure hardware, industry models to ecological partners, and then extending to application scenarios, which can be called panoramic penetration.
A few days ago, we officially released an upgraded AI product system for medical institutions, doctors, and users
The most highly anticipated solution is the "Ant Medical Big Model All in One Machine" full stack solution jointly launched by Huawei and Alibaba Cloud, which enables the hospital system to obtain a private deployment of domestic computing power, medical big models, and AI training and promotion. The first batch of institutions to be connected includes 7 institutions such as Hangzhou Medical Insurance Bureau and Beijing Traditional Chinese Medicine Hospital.
It is worth mentioning that in November 2024, the National Healthcare Security Administration included artificial intelligence assisted diagnosis in the project guidelines, and AI assisted diagnosis was included in medical insurance for the first time. This is a positive signal and a moment to see the truth for AI healthcare solution providers.
At the same time, the 290000 registered doctors on Good Doctor Online can improve efficiency in medical, teaching, and research scenarios through Ant's "AI Medical Record Assistant", "AI Science Popularization Assistant", and the latest "AI Research Assistant". In the future, there will be more diverse matrix tools to form AI super assistants.
AI services on the user side have already received data feedback. The "AI Health Manager" launched in September last year served nearly 40 million users in six months, helping ordinary people find doctors, read reports, accompany consultations, and more.
At the same time, Ant's ambition for medical AI lies in building a closed-loop system from diagnosis and treatment, service to health management.
This cannot be achieved in a short period of time. Ant has been investing in this track for nearly 11 years, moving from digital technology inside to AI technology inside, undoubtedly continuing to do deep industry and crossing longer life cycles.
02 The possibility of a single point breakthrough
Unlike Ant's more comprehensive reinvestment, the current entrants in the market are still in the stage of matching single point technology with single point scenarios.
If targeting hospitals and other institutions, AI will demonstrate its effectiveness in several scenarios:
The most typical example is imaging diagnosis, After learning hundreds of thousands of chest CT images marked by professional physicians, AI can quickly read the images and provide results. For some lesions smaller than 1cm, it is time-consuming and laborious for doctors to search with the naked eye. AI can provide results in one second, even marking the size, location, and density of nodules, and preliminarily distinguish between benign and malignant.
The lung CT imaging AI assisted diagnosis system introduced by Beijing Haidian Hospital has assisted in analyzing about 220000 cases so far.
AI is also assisting in diagnosis and treatment, such as surgical planning and surgical robots. Some medical model practitioners compare the former to car map navigation and the latter to autonomous driving.
At the same time, AI products targeting healthcare are gradually being implemented.
For example, Abridge, an AI recording assistant, helps medical staff complete clinical document recording. This product automatically recognizes the diagnosis and treatment process through voice recognition, and AI generates documents that meet the requirements.
The data provided by the company shows that Abridge can complete over 91% of the document recording workload for doctors, while deeply integrating with Epic, the largest electronic healthcare system in the United States. This not only saves doctors time but also eliminates the need to change their existing work habits.
On March 4th, Microsoft also launched its voice AI assistant Dragon Copilot for healthcare workers. As long as the Microsoft solution covers hospitals, the system will automatically capture and record conversations between doctors and patients, and AI will perform contextual analysis to automatically create clinical records.
In addition to hospital institutions and professional groups of medical staff, ordinary C-end users are also important service targets for AI healthcare. true
MedMatch is an AI driven healthcare solution used in sensitive areas such as mental health, men's health, and women's skin management. The core is to combine historical clinical data, treatment types, and other user training models, and finally provide recommended clinical decisions and treatment plans by AI.
There are also some AI assistants in China used for autonomous disease diagnosis, identification of common drugs, and establishment of personal health spaces.
But in the field of medical services, there is an unavoidable reality that hospitals, medical staff, and users cannot exist in isolation. It is difficult for medical AI to truly land and generate value by simply integrating large models in any scenario.
In addition, the problem of the illusion of large model capabilities is more challenging in the medical field due to the large amount of medical data with low quality, insufficient structure and standardization. It is particularly important to form a closed loop between technological iteration and multi scenario applications.
On the one hand, healthcare is naturally suitable for large-scale model applications, and on the other hand, the threshold in the healthcare field is relatively high.
All of this determines that entering medical AI cannot be done without touching the ground, and the reason why Ant Play is worth paying attention to is that 11 years of practice have helped it clarify this point faster.
How far is it from adding icing on the cake to becoming a must-have?
What we see is already a product of the development of AI healthcare to a certain stage.
Early medical AI was mostly limited to optimizing a single task, with obvious information fragmentation, such as the inability to determine tumor pathological classification based solely on images, or the need to combine laboratory results;
In addition, early AI was unable to explain diagnostic criteria in natural language like doctors do. Even now, after the evolution of large models, AI still has eight suspicions, unable to accurately obtain effective information and being criticized for lacking human touch.
Furthermore, the most important issue is the problem of data silos, where different hospitals use incompatible image formats and medical record systems.
These problems have not been eliminated at this stage, but the popularity of large models, DeepSeek's high performance, low cost, and open source have brought the ability of large models to a new level, and AI healthcare is also entering a new stage.
The most fundamental issues are data sensitivity and security, and open source models facilitate local deployment. Ant also advocates the lightweight design slogan of "integrated training and promotion, ready to use out of the box", with data available but invisible, and the entire diagnosis and treatment process traceable.
For example, reducing costs is beneficial for medical equality and universal access.
The research on the clinical application of artificial intelligence in medical imaging in China conducted by the team led by Liu Shiyuan, Director of the Radiology Department of Shanghai Changzheng Hospital, in the first half of 2022 showed that 73.9% of tertiary hospitals were equipped with AI assisted diagnosis software for imaging, while in primary medical institutions, this proportion was only 10.1%. After cost reduction, it will promote the shift of AI from pilot projects in top hospitals to grassroots inclusive applications.
These money and security issues may have solutions. The most important issue currently facing AI healthcare may be how to transform it from an icing on the cake to a must-have, making people willing to use it or even pay for it.
Last July, You Mao, Deputy Director of the Health Development Research Center of the National Health Commission, stated that 95% of China's research or output is focused on medical imaging; However, research in other fields such as medical robots, knowledge bases, and natural language processing is relatively insufficient; The research on "decision rules" is almost blank.
So for AI imaging diagnosis, institutions have a relatively high willingness and rate of payment. If adjuvant therapy is free, doctors are willing to actively try it, but if it costs hundreds of thousands of yuan to purchase officially, it may be treated with caution.
A healthcare practitioner's viewpoint is that the current development rate of AI medical products on the market may be less than 5%. You can interpret this as the difficulty of this track being extraordinary, or as the enormous market potential.
This article is written by [Newberry Daybreak], the product manager of Renren, and the WeChat official account is [Newberry Daybreak]. The original/authorized release to Renren is the product manager, and it is forbidden to reprint without permission.
The title image is from Unsplash, based on the CC0 protocol.
为什么微软、华为和蚂蚁，都不愿错过AI医疗？
从微软、华为到蚂蚁，这些科技巨头纷纷重金押注，试图在这片充满希望的蓝海中分得一杯羹。但问题是，AI 医疗到底有什么魔力，能让这些巨头们如此痴迷？
大模型的威力无需再普及，现在是讲应用的时候了。千行百业主动求变，其中应用速度、广度和深度排在之最的，当属医疗。
一份行业统计报告显示，2023 年末，国内行业大模型的分布排在前三甲的就是医疗医药、金融和科研。机器人还没有亮相春晚舞台，已经上过手术台。
2025 年年初，DeepSeek 横空出世，AI 医疗商业落地被进一步催化，最核心的两个原因：价格普惠，以及开源模型能进行私有化部署，更契合医疗数据敏感的安全需求。据不完全统计，截至目前，国内已有超百家三级医院官宣完成DeepSeek本地化部署。
最重要的是，C 端用户在经过上一波互联网医疗的教育普及之后，对AI 助手诊疗似乎更为接受。
所以，国外如英伟达、微软等巨头在重金投入，国内如华为、蚂蚁也在持续加码。AI 医疗都成为大模型最不愿错过的应用场景。
01 AI 医疗进化方向
大模型厂商进入医疗领域，他们的解题思路不受局限，如果有一定医疗服务基础的公司，甚至可能更为激进。
比如华为，3 月初组建的第 21 军团，正是医疗卫生军团。重点即时构建AI 辅助诊断解决方案体系，推动医疗大模型在临床场景的应用。
之后频繁落子，联合不同医院分别推出病理大模型、急性胸痛大模型等，同步也跟互联网公司积极推进一体机解决方案。其中包括同样深度布局AI 医疗的蚂蚁。
蚂蚁集团把AI 医疗作为确定性战略之后，几乎是在三端——机构、医护和用户同时发力。国内玩家少有如此全面深度布局，侧面也说明蚂蚁锁定医疗AI的野心和决心。
蚂蚁做医疗的基础在于支付宝。互联网医疗阶段，支付宝就在医院机构、用户端在挂号、问诊和支付等环节建立感知。近 11年积累，辐射全国 3600 家医院，累计服务用户超过 8 亿，是国内最大的医保支付服务平台，也是一站式医疗健康服务平台。
AI 阶段，蚂蚁医疗含义其实在拓宽，从基础设施硬件、行业大模型到生态伙伴，然后延伸到应用场景，已经能称得上全景式渗透。
前几天，也正式对外发布升级了面向医疗机构、医生和用户三端的AI 产品体系：
最受关注的还是联合华为、阿里云推出的「蚂蚁医疗大模型一体机」全栈解决方案，医院系统因此获得国产算力、医疗大模型、AI 训推一体的私有化部署。首批接入的有包括杭州市医保局、北京中医医院等 7 家机构。
值得一提的是，2024 年11月，国家医保局已经将人工智能辅助诊断列入立项指南，AI 辅助诊断首次被纳入医保。这对于AI 医疗解决方案提供商而言，是积极信号，也是见真章的时刻。
同时好大夫在线的 29 万注册医生，在医、教、研场景，可以通过蚂蚁开发的「AI 病历助手」、「AI科普助手」以及最新的「AI 科研助手」提升效率。未来会有更多丰富的矩阵工具，形成AI 超级助理。
AI 在用户端的服务已经有数据反馈。去年 9 月推出的「AI健康管家」，半年时间服务近 4000 万用户，帮助普通人找医生、读报告、陪看诊等等。
三端同时发力，蚂蚁对医疗AI 的雄心在于构建从诊疗、服务到健康管理的闭环。
这不是短时间内可以实现的。蚂蚁在这条赛道投入近11年，从数字科技inside走向AI技术inside，无疑是在继续做深产业，穿越更长生命周期。
02 单点突破的可能性
与蚂蚁更全面的重投入不同，目前市面上的入局者，尚处在单点技术匹配单点场景的阶段。
如果面向医院等机构，AI 在几类场景显现作用：
最典型如影像诊断， AI 在学习了数十万张专业医师标记的胸部CT 阅片信息之后，可以快速阅片并给出结果。有些小于 1cm 的病灶，医生肉眼寻找费时费力，AI可以一秒给出结果，甚至标出结节大小、位置、密度，初步分辨良恶性。
北京海淀医院引入的肺部CT影像AI辅助诊断系统至今已协助分析了约22万病例。
AI也在进行辅助诊疗，比如手术规划和手术机器人。有医疗大模型从业者将前者比喻为汽车地图导航，后者则是无人驾驶。
与此同时，面向医护的AI 产品也在逐步落地。
比如Abridge，一款AI 记录助手，帮助医护人员完成临床文档记录。这款产品通过自动语音识别诊疗过程，AI会生成符合要求的文档。
公司给出的数据是，Abridge 能完成医生 91% 以上的文档记录工作量，同时与美国最大的一家电子医疗系统Epic 深度整合，不仅节省医生时间，也不需要改变医生现有工作习惯。
3 月 4 日，微软面向医护人员也推出语音AI 助手Dragon Copilot。只要微软解决方案覆盖的医院，系统会自动捕捉记录医生和病人之间的对话，AI 进行语境分析，自动创建临床记录。
除医院机构和医护人员专业群体之外，C 端普通用户也是AI 医疗重要服务对象。商业落地产品主要集中在AI 个人健康管理和助手。
MedMatch 就是一款AI 驱动的医疗保健解决方案，用于心理健康、男性健康、女性皮肤管理等敏感领域。核心是结合历史临床数据、治疗类型等用户训练模型，最后AI 给出推荐的临床决策和治疗方案。
而国内也有一些AI 助手，用于自主疾病诊断，识别常见药品，建立个人健康空间。
但在医疗服务领域，有一个回避不了的现实是，医院、医护人员和用户，三者其实无法割裂而存在，在任何一个场景中浅接入大模型很难让医疗AI真正落地、产生价值。
此外，医疗数据量大同时质量不高、结构化和标准化不足，大模型能力的幻觉问题在医疗领域更具挑战，技术迭代和多场景应用之间形成闭环尤为重要。
一方面，医疗天然适合大模型应用，另一方面，医疗领域门槛较高。
这都决定了进入医疗AI没办法「脚不沾地」，而蚂蚁打法值得关注的原因就在于，11年实践帮助它更快明确了这一点。
03 从锦上添花到刚需必备有多远？
我们所看到的，已经是AI 医疗发展到一定阶段的产物。
早期医疗AI 多局限于单一任务优化，存在明显的信息割裂，比如仅凭影像无法判断肿瘤病理分型，还是需要结合实验室结果；
此外，早期AI无法像医生一样用自然语言解释诊断依据，即便现在如大模型进化之后，AI 也有八股嫌疑，无法精准获取有效信息，还被人嫌弃没有人情味儿。
再者最重要的就是数据孤岛问题，不同医院使用的影像格式、病历系统互不兼容。
这些问题现阶段没有消除，但是大模型的普及，DeepSeek的高性能、低成本和开源又将大模型能力带到一个新台阶，AI 医疗也在进入新阶段。
最基础的就是数据敏感和安全问题，开源模型方便本地部署。蚂蚁也打出「训推一体，开箱即用」的轻量化设计口号，数据可用不可见，诊疗过程全程可溯源。
再比如降低成本，利于医疗平权和普惠。
上海长征医院放射诊断科主任刘士远团队2022年上半年做的中国医学影像人工智能临床应用情况调研，73.9%的三级医院配备了影像的AI辅诊软件，而在基层医疗机构，这一比例仅有10.1%。成本降低之后，会推动AI从头部医院试点转向基层普惠应用。
这些金钱和安全问题，或许都有解决办法。目前AI 医疗可能面临最重要的问题是，如何从锦上添花变成刚需必备，让人有意愿使用，甚至付费。
去年7月，国家卫生健康委卫生发展研究中心副主任游茂曾表示，中国95%的研究或产出都集中在医学影像类；而在其他领域如医疗机器人、知识库、自然语言处理的研究相对不足；在「决策规则」的研究几近空白。
所以对于AI 影像诊断，机构的付费意愿和付费率比较高，辅助治疗如果是免费，医生也愿意积极尝试，但是如果要花大几十万元正式采购，可能就会被慎重对待。
一位医疗从业者的观点是，目前市面上已有的AI 医疗产品开发率可能不足 5%。你可以将此理解是这条赛道的难度并不一般，也可以解读为巨大的市场潜力。
本文由人人都是产品经理作者【新莓daybreak】，微信公众号：【新莓daybreak】，原创/授权 发布于人人都是产品经理，未经许可，禁止转载。
题图来自Unsplash，基于 CC0 协议。