At the 2025 Shanghai Stock Exchange International Investors Conference held on November 12-13, Dai Weimin, Chairman and CEO of Verisilicon Microelectronics (Shanghai) Co., Ltd., highlighted the growing demand for AI-customized chips (AI ASICs). Addressing the increasing global need for AI ASICs, the semiconductor industry veteran outlined Verisilicon’s strategic roadmap and future opportunities.
**GPU and AI ASIC: Complementary, Not Competing** Dai Weimin clarified a common industry misconception that GPUs and AI ASICs operate in separate markets. He emphasized that while GPUs excel in flexible, general-purpose deployment, AI ASICs deliver unparalleled cost-performance efficiency. As AI models continue to evolve, the two technologies develop synergistically. Verisilicon offers a comprehensive and adaptable portfolio, spanning AI accelerators like GPUs, GPGPUs, and ASICs, enabling tailored solutions for diverse applications. Dai also underscored the vast potential of AI ASICs in edge computing.
**From "Reading" to "Experiencing": AI Needs a "World Model"** Dai used vivid analogies to describe AI’s next leap. He likened large language model training to "reading thousands of books"—absorbing patterns from vast text data. However, this alone is insufficient. "Traveling thousands of miles is less valuable than interacting with countless people," he added, stressing that AI must grasp human emotions and complex scenarios to build a "world model." Future AI must process not just text and images but spatial, physical, and contextual information to achieve true "thinking," demanding diverse computing power.
**The Rise of Edge Inference: "Digital Leaves" Hold Untapped Potential** Dai divided computing needs into "cloud" and "edge," comparing cloud-based large-scale training to a sturdy "tree trunk," while the real value lies in the "lush branches and leaves" growing from it. "Edge devices primarily handle inference and fine-tuning," he explained. Deploying AI models on smartphones, cars, smart glasses, and IoT devices—optimized for verticals like healthcare, finance, and education—is key to AI’s commercialization.
**Empowering the Edge: From Smart Glasses to AI Toys** Dai showcased edge AI’s potential through real-world examples. Smartphones equipped with AI-customized chips could vastly improve photography, image quality, and power efficiency. He singled out smart glasses as a promising growth market, envisioning compact, efficient AI models enabling real-time, offline voice translation and scene interaction without cloud dependency. Education, particularly AI-driven toys, is another disruptive frontier. AI toys with embedded models could generate stories based on a child’s daily experiences, narrated in a parent’s voice, addressing the lack of parental time. Dai stressed that edge applications must operate offline. For instance, AI toys shouldn’t be confined to Wi-Fi-enabled spaces, and privacy and security are paramount.
**Focusing on Core IP and AI ASICs for the Edge Revolution** Dai’s insights signaled a shift in AI computing from the cloud to the edge. Verisilicon’s strategy leverages its semiconductor IP portfolio and chip design expertise to deliver AI ASIC solutions for this revolution. He concluded, "Edge intelligence will unlock a larger market. This is Verisilicon’s vision and a strategic call to the industry: empowering end devices across sectors is the next trillion-dollar opportunity in AI democratization."
Comments