During TSMC's recent shareholder meeting, Chairman C.C. Wei highlighted a major paradigm shift in AI demand. The industry is rapidly moving from generative AI and search query models toward Agentic AI and action-oriented command execution. This evolution is exponentially driving up token consumption required for Large Language Models (LLMs) to process text. Consequently, the demand for computing power continues to grow rapidly, providing robust foundational support for advanced semiconductor technologies. Wei noted that TSMC's customers and their respective clients continue to share highly positive outlooks for the AI sector.
Wei emphasized that TSMC remains highly confident in the long-term AI megatrend. Backed by strong technological differentiation and a broad customer base, TSMC projects its full-year revenue growth to exceed 30% in USD terms. Additionally, the company reiterated its guidance for the current quarter, with consolidated revenue expected to range between US$39 billion and US$40.2 billion, and a gross margin between 65.5% and 67.5%.
Looking ahead, while demand for advanced nodes (such as 3nm and 2nm) remains exceptionally strong, TSMC will closely monitor the impact of rising component prices, particularly on consumer products and price-sensitive end markets. Geopolitical uncertainties, such as recent tensions in the Middle East, also add macroeconomic volatility. TSMC plans to navigate these challenges prudently, focusing on operational fundamentals to further strengthen its competitive leadership.
[AgentUpdate Depth Analysis] C.C. Wei's emphasis on "Agentic AI" highlights the shifting bottleneck of AI infrastructure. Traditional search-and-response AI models scale token consumption linearly. In contrast, Agentic AI—characterized by autonomous planning, tool use, iterative self-reflection, and multi-agent workflows—operates on complex reasoning loops that generate exponential token volumes for a single user query. This paradigm shift relocates the center of gravity of compute demand from heavy offline training to continuous, high-volume production inference. As the exclusive manufacturer of the world's most advanced silicon, TSMC's optimistic growth guidance validates that the AI Agent ecosystem is driving real, massive physical infrastructure expansion. The rise of Agentic AI is no longer just a software trend, but a material force reshaping the semiconductor supply chain.
