TSMC: Semiconductors to reach $1.5 trillion by 2030; AI will replace phones as the biggest driver

Zhang Xiaoqiang, senior vice president and deputy chief operating officer at TSMC, said at TSMC’s annual technology forum on May 14 that the global semiconductor market is expected to reach $1.5 trillion by 2030, higher than the $1 trillion figure that the market had broadly expected earlier. He noted that over the past 10 years, the semiconductor industry was mainly driven by smartphones, but in the next 10 years the core growth momentum will fully shift to AI and high-performance computing (HPC).

2030 Semiconductor Market Structure Forecast: Sub-item Data

At the forum, Zhang Xiaoqiang disclosed TSMC’s forecast for the global semiconductor market structure in 2030:

AI and HPC: About 55% of global semiconductor market value

Smartphones: About 20% (a sharp decline from today)

Automobiles: About 10%

Internet of Things (IoT): About 10%

TSMC also expects that if global semiconductor market value reaches $1.5 trillion in 2030, the related electronic equipment market size will further expand to about $4 trillion, and the overall information industry market value could rise to as high as $15 trillion.

COUPE Technology: Definition, Goals, and Technical Background

Zhang Xiaoqiang said that the development of AI chips is gradually approaching the physical limits of traditional electronic signal transmission. In the future, AI systems will rely heavily on optical communications and photonics technologies to solve data center bandwidth and power consumption problems.

The core goal of COUPE is to enhance high-speed data transmission capabilities among AI systems through miniaturization and generalization of a photonic engine, while also reducing energy consumption. TSMC’s proposed “AI three-layer cake” architecture this time includes a logic computing layer, heterogeneous integration, and a 3D IC layer, as well as a high-speed transmission layer centered on photonics and optical interconnects.

Supply Chain Impact: Demand Spreading in the Inference Stage and the AI Infrastructure Race

Zhang Xiaoqiang said that currently, almost all AI accelerators worldwide are built on a division-of-labor model between IC design companies and foundries, which accelerates innovation in AI chip architectures. TSMC has observed that the focus of AI development is shifting from model training to the inference stage. This means AI chip demand will spread from a small number of large technology companies to the enterprise segment, end-user devices, and edge computing markets. Market analysts pointed out that the continued expansion of AI infrastructure investment by major technology companies such as NVIDIA, AMD, Google, and Amazon is driving demand growth for related supply-chain areas including advanced packaging, HBM, high-speed interconnects, and AI ASICs.

FAQ

By how much does TSMC’s $1.5 trillion forecast for 2030 exceed market consensus?

Zhang Xiaoqiang said TSMC expects global semiconductor market value to reach $1.5 trillion by 2030, while the market had previously generally expected about $1 trillion. TSMC’s forecast is higher by roughly 50%, with the gap mainly coming from a higher estimate of the scale of AI and HPC demand.

What are the differences in positioning between COUPE and CoWoS?

CoWoS is TSMC’s advanced packaging technology currently used for high-bandwidth memory (HBM) and AI chip packaging, widely used in AI accelerators such as NVIDIA. Zhang Xiaoqiang positions COUPE as an important new direction after CoWoS, focusing on breakthroughs in photonic interconnect technology that push past the limits of electronic signal transmission. TSMC has not yet disclosed COUPE’s complete technical specifications or a commercial timeline.

What structural impact does the shift of AI from training to inference have on TSMC’s chip demand?

The demand distribution for chips in the inference stage is broader and no longer limited to only a small number of large infrastructure systems for training; instead, it spreads to enterprise servers, end-user devices, and edge computing. Zhang Xiaoqiang said this means the potential customer base for AI chips expands significantly, further supporting long-term growth momentum in the semiconductor industry.