AI in China

BOE built an in-house deep-learning defect classification system for display panel lines, trained on AOI images in its data center and run on Intel Xeon edge servers with OpenVINO. Deployed across its factories, it reached 96% inspection accuracy and a 70% labor substitution rate.

KingMed Diagnostics and the First Affiliated Hospital of Guangzhou Medical University jointly released DeepGEM 2.0, a multimodal AI model that predicts lung cancer driver gene mutations from routine digital pathology slides in under one minute, covering all gene targets corresponding to domestically approved targeted therapies. The model is being lightweight-deployed across 500+ Chinese medical institutions via KingMed's KMDP digital pathology platform, with development timelines of 12-16 weeks for new gene additions when new drugs receive regulatory approval.

West China Hospital of Sichuan University deployed an AI voice robot that phones discharged patients for follow-up. By October 2024 it covered 99% of discharged patients, dialed 40 calls at once, removed 106,000 manual calls in one quarter and shortened each follow-up from 10-15 minutes to about 2 minutes.

Zhejiang Cancer Hospital and Alibaba DAMO Academy developed DAMO GRAPE, the first non-contrast CT AI for gastric cancer screening, achieving 85.1% sensitivity and 96.8% specificity, with retrospective studies showing earlier detection of 11 cases by 2-10 months. The model is being deployed in Zhejia

Yadea, the world's top-selling electric two-wheeler maker with 40,000 stores, rebuilt its service center with Alibaba's Lingyang in a two-phase project launched September 2025. Hotline bot resolution rose from 20% to 42%, online bot resolution hit 78%, hotline connect rates rose from 88% to 95%, and

First Affiliated Hospital of Zhengzhou University deployed a Kubernetes-based distributed medical imaging AI platform integrating 8 imaging AI applications across all campuses — managing 350,000+ imaging studies, lifting exam volume 20%, and cutting research cycles from 10-20 months to under 6.

Shenzhen Hello Tech Energy, maker of Jackery portable power stations, worked with AWS on a plug-in for its overseas customer service system that uses Claude 3 Sonnet on Amazon Bedrock to identify intent and draft replies from a Bedrock knowledge base. About five weeks after launch, agents sent generated replies directly for more than 30% of enquiries, at 1/20 of the cost of a third-party AI tool.

Beijing Friendship Hospital uses a DeepSeek-based system for imaging report generation on more than 330 workstations, with about 10,000 uses a month, and a knowledge Q&A service. A distilled model fine-tuned on hospital reports drafts imaging reports through voice input linked to PACS, cutting report steps from five to three, while the full DeepSeek model answers staff questions from a RAG knowledge base.