Taipei: The next phase of artificial intelligence growth will be driven by the large-scale deployment of integrated computing systems as the industry enters the "true industrialization of AI," a Taiwan Semiconductor Manufacturing Co. (TSMC) executive said Monday.
According to Focus Taiwan, April Li, TSMC's AI and high-performance computing business development director, mentioned that AI agents are moving into enterprise workflows, while investment in specialized data centers and physical AI is accelerating. She emphasized that market leaders in this new era will be those who build the most integrated systems rather than those who simply make better models. Li delivered these insights during a keynote speech at the SEMICON Taiwan 2026 IC forum.
Future AI infrastructure, Li explained, will increasingly require the integration of computing, memory, interconnects, storage, and power management across chips, server racks, and entire data centers. This shift is partly driven by the soaring demand for AI inference-the process of using trained models to respond to requests and perform tasks.
Li reported that global inference token volume has increased nearly 500-fold from its 2022 level, with more complex reasoning and agentic AI systems consuming far more tokens than conventional one-shot queries. "Inference is no longer a low-overhead task," Li said, noting that it is becoming the dominant driver of system-level expansion as AI agents operate continuously in the background.
The growing workloads are exerting pressure on four key areas: logic scaling, interconnect efficiency, memory performance, and power delivery and cooling. Data movement can account for up to 60 percent of system activity in typical workloads, leaving expensive accelerators operating at below 40 percent utilization.
Li also highlighted that AI packages could contain more than 1 trillion transistors by 2030, necessitating multi-die architecture and heterogeneous integration. To tackle these constraints, TSMC is developing technologies that cover advanced logic, packaging, and optical interconnects.
In particular, Li pointed out the chipmaker's 3DFabric platform, which includes technologies such as SoIC 3D stacking and CoWoS advanced packaging, along with its Compact Universal Photonic Engine (COUPE) platform for high-speed optical data transmission. TSMC is also working on improving high-bandwidth memory performance using advanced logic technology in high-bandwidth memory base dies.
During a question-and-answer session, Li stated that TSMC must develop a comprehensive understanding of the supply chain "from silicon to the data center to the token" and collaborate with ecosystem partners to validate its technologies as early as possible across that chain. This approach reflects how TSMC's role has expanded beyond manufacturing individual chips. "Gone are the days when we can just ship wafers across the fence," Li concluded.