Medical AI, which had largely been forgotten by major TMT (Technology, Media, Telecom) giants in 2024, seems to have experienced a dark quiet before the dawn. At WAIC 2025, these giants showcased their medical AI products in full force. In this hyper-competitive environment, companies did everything they could to demonstrate how AI would soon be deployed across all kinds of medical scenarios. The “AI hospital” is fast becoming reality—it’s no longer just an ambitious promise on a PowerPoint slide.
What Does an “AI Hospital” Look Like?
If the WAIC venue this year were a hospital of the future, then:
• The outpatient hall would be operated by Ant Group’s “AI Health Assistant AQ,” which is simultaneously registering, interpreting scans, and accompanying doctor visits for 100 million users—while also reminding grandpas, “Don’t stop taking your blood pressure meds.”
• The emergency room is contracted to Lansong Digital Science’s “LAN-AI Agents,” which generate a “digital twin ICU” at the click of a button, down to real-time monitoring of the ambulance tire pressure.
• The imaging department has the shortest queue—because United Imaging Intelligence’s “Chest One-Scan Multi-Check” agent detects 73 types of abnormalities in one go, with an average AUC (Area Under Curve) of 94%.
• The Traditional Chinese Medicine department is the liveliest: Zhihuiyan’s “Bianshi AI Four Diagnostics Instrument” compresses tongue, facial, and pulse diagnostics into just 3 minutes, and the constitution report even includes a seasonal tip like “Drink pear soup in autumn.”
• In the hospital director’s office, New H3C’s “LinSeer ICT Agent” monitors the hospital’s Wi-Fi, utilities, and CT machines 24/7.
From the 2025 WAIC’s medical AI product exhibitions, we can observe several major development trends:
1. From “Small Models for Single Diseases” to “Large Models for Specialties”—AI as a Senior Medical Expert
In the past, medical AI resembled a game of whack-a-mole: one model for lung cancer, another for fractures. Doctors had 20 executable programs installed on their computers, and their desktops were messier than the receding hairline of a 1990s-born programmer.
This year’s WAIC keyword is: “Specialty Agent Clusters.”
• United Imaging Intelligence’s “Chest One-Scan Multi-Check” integrates 73 types of chest abnormalities, including lung nodules, emphysema, fractures, and aortic widening, into one model. Doctors at Zhongshan Hospital tested it and found that report writing time decreased from 8 minutes to 6 minutes—a 25% increase in efficiency.
• Ruijin Hospital × Huawei’s RuiPath pathology large model covers 90% of China’s most common cancers. It’s open-source, free for commercial use, and elevates the pathology department from the “hell of slide reading” to the “heaven of question-answering.”
• New H3C × Desheng Bio’s iMedImage™ large model supports “no-code” fine-tuning with a single click. Even primary care hospitals can train their own “lung nodule mini-expert” in 10 minutes. It’s essentially an AI-powered “equity movement.”
2. From “Single Modality” to “Multimodal”—Achieving a Closed Loop in Diagnostics
Medical AI’s new level of competition: not only must it interpret CT scans, it must also listen to coughs, examine tongues, and recite healthcare policy with perfect fluency.
• Ant’s AQ supports mixed inputs of voice, text, images, and lab reports. Users can upload a photo of their blood test results—AI first performs OCR, then inference, and finally explains in plain terms: “High white blood cell count means your body is blowing a whistle to call security.”
• NeurOS’s “Demon Mirror” performs a 30-second facial scan to read subcutaneous blood flow and predict 18 health indicators. A woman trying the mirror on-site was told she had “mild dark circles—recommended to avoid staying up late.”
• Shanghai Conservatory of Music has built a “Music Healing Capsule” that combines EEG, facial emotion, and WFS spatial audio. It delivers an all-in-one music prescription for depression, insomnia, and hypertension. A 5-minute demo brought “sunlight to both body and soul.”
3. From “Hospital IT Projects” to “Digital Health Experience Pods”—Patients Receive Care While Lying Flat
This year, every company is working on the same goal: deconstruct the hospital, and “deliver” its services straight to the patient’s home.
• JD Health has launched over 500 expert AI avatars. Even at 2 a.m., patients can ask the “Peking Union Medical College Hospital Thyroid AI Expert” questions—and get responses faster than food delivery.
• New H3C’s “Digital Health Experience Pod” is open 24/7, featuring 350+ AI doctor avatars and 27 languages. It supports calorie recognition via photo, instant comprehension of medical reports, and online prescriptions. It’s like an “unmanned convenience store” for healthcare.
• The First Affiliated Hospital of Guangxi Medical University’s “Zhenshan AI Digital Doctor” supports multiple languages—Chinese, Vietnamese, Lao, Thai—bringing remote consultations to Southeast Asia. Patients can video call with Nanning experts using just their phones, saving even on translation fees.
4. From “Burning Money and Telling Stories” to “Hardware + Data + Real Scenarios” Trinity—Only Giants Can Survive
Take three laps around WAIC, and you’ll notice: the startups that raised funding with “lung nodule algorithms” in the last wave have mostly disappeared. Taking center stage now are big players—Alibaba, Huawei, New H3C, JD, Ant—all of whom have “hardware hooks” or “traffic entry points.” Why?
• Computing Cost: Moore Threads showcased an interconnected GPU cabinet with 1,024 cards for training trillion-parameter models. The electricity bill alone equals a day’s lighting costs in a mid-sized city.
• Data Barrier: United Imaging Medical relies on imaging data from 4,000 hospitals to build a model with 94% AUC across 73 abnormalities. Small companies can’t even gather data for 7 abnormalities.
• Real-World Deployment: New H3C uses its iMed MaaS platform to turn model training, deployment, and maintenance into a one-click package. Primary care hospitals can get it running in 10 minutes—saving the manpower of three engineers.
Conclusion
The exhibits at WAIC are rapidly transforming into real-world deployments. In the next 12 months, we can expect to see:
• Q4 2025 — RuiPath pathology large model from Ruijin Hospital launches its plugin marketplace in the open-source community.
• H1 2026 — United Imaging’s Chest One-Scan Multi-Check rolls out to 2,000 grassroots hospitals.
• H2 2026 — Ant’s AQ integrates with personal health insurance payment systems.
[Disclaimer]: The above content reflects analysis of publicly available information, expert insights, and BCC research. It does not constitute investment advice. BCC is not responsible for any losses resulting from reliance on the views expressed herein. Investors should exercise caution.