AI Enters Industrial-Scale Cycle
The Taiwan Semiconductor Industry Association held its annual meeting on the 29th, where TSMC Vice President of Business Development Yuan Liben was invited to speak at the forum on Taiwan's competitive advantages and challenges in AI development trends. He noted that the launch of ChatGPT in Q4 2022 marked the first widespread use of large language models by humans, and that before 2025 the AI industry was in an innovation cycle focused on large language model breakthroughs, algorithm exploration, and early agentic AI proof-of-concept, a stage of hybrid traditional and AI operation. Starting this year, it has entered an industrial-scale cycle, moving beyond the lab stage into large-scale deployment, including infrastructure buildout and broad applications such as enterprise-grade AI agents, dedicated data center construction, and embodied AI robots.
Hundred-Fold Computing Power Tests Supply Chain Execution
Yuan Liben said capital expenditure and industry adoption have seen explosive growth, deeply felt across Taiwan's industry chain, and the speed of the boom has brought unprecedented challenges. The supply chain must have high-multiplier execution capability to complete system validation and rapid deployment simultaneously. To support a hundred-fold surge in computing power, substrates, printed circuit boards, power supply, and cabinet-related high-speed optical fiber connections must be deeply integrated; if any one link hits a bottleneck, the entire cabinet stalls, so the industry should adopt all-around co-innovation from chip to cabinet.
One-Stop Manufacturing Is Taiwan's Moat
He stressed that Taiwan has the most complete and concentrated industrial cluster, serving not only as the wafer foundry leader but also as a global supplier of AI system integration solutions. From wafer manufacturing, packaging, memory, substrates, thermal interface materials to factory shipment, Taiwan has one-stop manufacturing capability proven in high-volume production, which no other region in the world can replicate in the short term, and this is precisely Taiwan's moat.
Four Major Challenges: Talent, Power, Regulations, Geopolitics
On challenges, Yuan Liben said Taiwan's industry must provide real-time solutions to system and performance bottlenecks, and consider how to capture higher service value from AI data centers and applications beyond hardware manufacturing, as well as how to scale AI and physical AI applications. It must also address cultivating domestic talent and attracting overseas talent, supplying the power resources needed for AI data center buildout and chip capacity expansion, supporting rapid capacity expansion in land acquisition and regulatory adaptation, and responding to the impact of geopolitics on overseas deployment and supply chain regionalization.
Calls for Early Co-Design and National Talent Pool
To this end, Yuan Liben proposed assistance in creating new proposals, advocating a shift from a passive foundry model to early-stage technology co-design, and promoting a joint experimental platform to break through technological bottlenecks together. He also called for promoting open-source tools, setting standard formats and governance rules, and building a secure cross-industry data platform. On the policy front, he suggested upgrading talent cultivation, expanding AI professional training, building a national-level AI talent pool, improving regulations and standards, accelerating AI product certification and intellectual property protection mechanisms, providing funding and tax incentives, and deepening local self-developed supply chain resilience, with the hope of propelling Taiwan to become a global AI neural hub.