TSMC's Zhang Xiaoqiang: Semiconductor growth momentum shifts from smartphones to AI, "COUPE" will become a key term

TSMC estimates that by 2030, the output value will reach 1.5 trillion USD, driven by AI. The core shifts toward high-performance computing, with COUPE technology breaking through transmission bottlenecks, leading supply chain upgrades.

TSMC revises upward its semiconductor market forecast, with AI as the core growth engine

At its annual technology forum, TSMC released a new growth forecast for the global semiconductor market. Senior Vice President and Co-Chief Operating Officer Dr. Wei Zhang stated that AI is rapidly transforming the entire semiconductor industry structure, with the global semiconductor output value expected to reach 1.5 trillion USD by 2030, higher than the previously common market estimate of 1 trillion USD.

Dr. Wei Zhang pointed out that over the past 10 years, the semiconductor industry’s growth was mainly driven by smartphones and mobile devices, but the core momentum for the next 10 years has fully shifted to AI and high-performance computing (HPC).

According to TSMC’s estimates, by 2030, AI and HPC will account for about 55% of the global semiconductor output value, while smartphones will decrease to about 20%, and automobiles and IoT each account for about 10%.

He described that the impact of AI on semiconductors may be the most significant structural shift in the tech industry in recent years. From generative AI and AI agents to future Physical AI, global computing demand is rapidly expanding, also prompting upgrades in chip design, advanced packaging, and data center architecture.

AI chip demand surges, photonic technology becomes the next battleground

In response to the continuous expansion of AI system scale, Dr. Wei Zhang also proposed TSMC’s version of the “AI Three-Layer Cake” architecture, including logic computation, heterogeneous integration, and 3D ICs, as well as future high-speed transmission core photonic and optical interconnect technologies.

He stated that current AI chip development is gradually approaching the limits of traditional electronic signal transmission, so future AI systems will heavily rely on optical communication and photonic technology to address the massive bandwidth and energy consumption issues in data centers.

TSMC also specifically introduced the “COUPE (Compact Universal Photonic Engine)” concept, regarded as an important new technological direction following CoWoS. The core goal of COUPE is to enhance high-speed data transfer between AI systems through miniaturization and universal optical engines, while reducing energy consumption.

As AI models grow larger, data center bottlenecks are gradually shifting from mere computational capacity to data movement efficiency and power consumption issues. The market generally believes that optical interconnect technology will become a key competitive focus for AI infrastructure in the future.

AI super-cycle takes shape, reshuffling the global tech supply chain

Market analysts believe that AI has ushered the semiconductor industry into a new long-term growth cycle. Past semiconductor cycles were mostly influenced by fluctuations in PC and mobile device demand, but the demand from large data centers and infrastructure driven by AI is changing the overall industry growth curve.

Especially this year, the focus of AI development has shifted from model training to inference. This indicates that in the future, besides large tech companies, enterprises, end devices, and edge computing markets will also begin to heavily adopt AI computing needs.

Dr. Wei Zhang also mentioned that almost all AI accelerators today are built on a division of labor model between IC design companies and foundries. This model allows chip companies to focus on architectural innovation, while the highly complex manufacturing processes are handled by foundries, further accelerating AI chip development.

As major tech companies like NVIDIA, AMD, Google, and Amazon continue to expand their AI infrastructure investments, the market is also beginning to anticipate benefits for advanced packaging, HBM, high-speed interconnects, and AI ASICs.

In addition to semiconductors themselves, TSMC estimates that if the global semiconductor output value reaches 1.5 trillion USD by 2030, related electronic device markets could further expand to 4 trillion USD, and the overall information industry output could reach 15 trillion USD. The global tech industry is gradually forming an “AI infrastructure race,” with governments and large corporations simultaneously increasing investments in data centers, energy, networks, and advanced chips.