Distinguished leaders, industry pioneers, ladies and gentlemen,
Good morning!
I am Zhao Hejuan, Founder & CEO of TMTPost Group. It is my great honor to join the AI Computing Power Development Forum in MWC.
As a longtime researcher, analyst, and entrepreneur in AI applications, I would like to share some of my observations on how the edge AI model or on-device AI model is reshaping industrial intelligence, which will have three parts: the rise of edge AI, the key challenges for edge AI and China's unique advantages in edge AI.
Firstly, about the rise of edge AI
We are at a pivotal moment in the Fourth Industrial Revolution. According to the latest data, the global edge AI device market size has exceeded $60 billion, with a compound annual growth rate (CAGR) of 22%, far surpassing the growth rate of cloud-based AI services.
China accounts for more than 35%, and it is expected to exceed 150 billion US dollars by 2030.
This signals a fundamental shift—AI is moving from centralized cloud computing to real-time edge processing. Gartner projects that by 2025, 75% of enterprise data will be processed at the edge, marking a historic transition from "centralized intelligence" to "distributed intelligence."
Secondly, what will be the key challenges for edge AI?
To fully realize this transformation, three major challenges must be addressed:
1. Model Optimization for Edge Deployment
AI models are growing exponentially—Stanford's AI Index Report states that model parameters increase by 230% annually. Yet, edge AI requires lightweight solutions.
For example:
• Carnegie Mellon University developed a blind navigation ring that compresses environmental recognition models to just 52KB.
• Dutch startup Epitel created an epilepsy warning system in 0.5MB, providing 90-second early alerts while reducing false alarms by 40%.
These breakthroughs prove that smaller AI models can be just as powerful in real-world applications.
2. Continuous Learning and Evolution
AI must continuously improve based on real-world data.
Google's DeepMind lab has unveiled a new AI diagnostic system, "Med-PaLM Oncology," which can identify early signs of 13 types of cancer within 3 seconds. The system has achieved a clinical validation accuracy rate of 96.7%, surpassing that of human doctors.
This aligns with IDC's Edge Intelligence Evolution Theory—when edge devices gain continuous learning capabilities, their efficiency improves exponentially.
3. Breaking Industry Barriers
Edge AI is revolutionizing industrial sectors.
• In Tesla's Shanghai factory, an edge AI vision system has reduced the false alarm rate to 0.5%, increased the detection accuracy rate to 99.98%, and improved the efficiency by five times.
• In Shouguang, eastern China's Shandong province, an edge AI-powered agricultural drone improved pest detection accuracy by 40% and reduced pesticide consumption by 35%.
Gartner predicts that by 2025, the efficiency of local links in the manufacturing industry will increase by 20%-50%.
However, to maximize edge AI’s potential, we must build three essential pillars:
1. A “Data Flywheel” Ecosystem
IDC predicts every day, the world generates 14.849 billion TB of edge data, but less than 15% is utilized.
• In the latest AI smartphone improved local data processing 6x, reducing latency to 8 milliseconds.
• Smart excavators cut energy consumption by 22% using edge decision-making.
2. AI-5G-IoT Integration
According to Boston Consulting Group, integrating AI with 5G and IoT is unlocking new efficiencies:
• At Qingdao Port, a 5G + Edge AI system improved container scheduling efficiency by 40%.
• At Ant Group, Blockchain + Edge AI reduced cross-border payment processing time from hours to seconds.
3. An Open and Collaborative Industry Community
Today, over 200 global open-source edge AI projects exist, with Chinese enterprises contributing 22%.
The Linux Foundation’s 2024 Edge Computing White Paper states that open collaboration can reduce edge AI deployment costs by 60%.
A great example is the Huawei Ascend + SenseTime partnership, which developed a lightweight AI model toolchain, tripling development efficiency.
In the last part, I would like to talk about China’s unique advantages in edge AI.
China is in a strong position in the global Edge AI revolution:
• 37% of global edge AI patents originate from China.
• The deployment rate of edge AI devices on the smart city side exceeds 60%.
• 45% of edge AI applications in industrial quality inspection scenarios.
• By 2025, China's edge computing market is expected to reach 200 billion yuan.
Looking ahead, the future of edge AI isbased on comprehensive forecasts from multiple institutions:
• By 2026, 50% of enterprise edge AI systems will adopt dynamic task allocation strategies.
• By 2027, 90% of edge AI devices will support multimodal interaction.
• By 2030, 30% of industrial edge devices will be equipped with self-learning capabilities.
• Edge AI will boost global GDP by 0.3–0.8 percentage points annually.
This is not just about technological advancement—it is a critical step in transitioning towards an intelligent society.
To conclude, let me share a real-world case from TMTPost’s research—the AI-powered transformation of an automotive factory.
After edge AI was integrated into 287 production steps:
• Per capita output increased by 4.6 times.
• Defect rates dropped to just 3 PPM (parts per million).
This confirms today's core message—when AI computing power reaches the industrial frontline, we unlock not just an efficiency revolution but a fundamental upgrade in human productivity.
Let's work together to drive this silent yet transformative revolution forward.
Thank you!
更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App
