Hsu Tso-Sheng's Perspective: When AGI Meets the Chip Law – The Second Half of the Global Semiconductor Race

The global chip war is evolving from a single dimension of 'geometric transistor miniaturization' into a comprehensive confrontation of 'system integration.' Huawei recently proposed 'Tao's Law,' which emphasizes system architecture optimization—not to replace the traditional Moore's Law, but to reveal a new direction for industrial transformation. As chip development hits dual bottlenecks of physical limits and geopolitical constraints, the key to future success is no longer merely the 'nanometer process number,' but rather how to accelerate R&D through artificial intelligence technologies such as 'Scientific Intelligence (AI4S)' and 'Autonomous Scientific Discovery Systems (ASDS),' even heralding the cross-disciplinary system integration toward the era of 'Artificial General Intelligence (AGI).'

The core battleground in the semiconductor industry in recent years has been Huawei's proposed 'Tao's Law.' This law argues that chip competition must shift from 'relentlessly pursuing smaller transistors' to 'optimizing the entire system architecture.' This argument directly challenges 'Moore's Law,' which has dominated the semiconductor industry for half a century. These two laws are not zero-sum opposites, but represent a paradigm shift in semiconductor competition: extreme miniaturization remains foundational, but innovation in system architecture is now the key determinant of who will hold the话语权.

For the past fifty years, global technological progress has largely relied on Moore's Law. Its core logic is: make transistors smaller, pack more components onto a single chip, and the chip becomes faster, more power-efficient, and cheaper. However, as processes approach the physical limits of atomic scales, quantum effects cause leakage currents and heat dissipation issues. Even as transistors shrink, delays and power consumption from moving data between different chips and memory continue to hinder overall performance. This is commonly known as the 'data traffic jam' bottleneck.

'Tao's Law' is an engineering solution designed to overcome this bottleneck. Its core objective is to reduce the system's total latency. By reconfiguring data pathways, adopting advanced packaging, and using 3D stacking technologies, chips can run faster and more efficiently without relying on extreme nanometer processes.

In this competition between the two laws, the recently emerged 'Scientific Intelligence (AI4S)' and 'Autonomous Scientific Discovery Systems (ASDS)' have become the most critical game-changers. These technologies transform the traditionally human-dependent, time-consuming trial-and-error R&D process into a high-speed, automated algorithmic loop, and are also seen as foundational building blocks for the underlying computing architecture needed to achieve 'Artificial General Intelligence (AGI).'

On the dimension of Moore's Law, discovering new materials is a major challenge. In the past, scientists had to conduct countless experiments and simulations in labs to find novel 2D materials that could outperform silicon. Now, through AI algorithms, systems can screen tens of thousands of stable crystal structures in a short time, enabling a leap forward in the discovery and application of new materials, thereby extending the lifespan of Moore's Law.

On the system integration dimension of Tao's Law, 3D chip stacking involves complex heat dissipation and electromagnetic interference, whose complexity has surpassed the intuitive limits of human engineers. By introducing AI-assisted design systems, algorithms can quickly compute optimal layouts and heat dissipation paths, achieving global optimization of system performance. This makes the two laws no longer mutually exclusive, but instead integrated by AI behind the scenes—providing high-density physical components on one side (Moore's Law) and globally configuring system architecture on the other (Tao's Law)—forming a double-helix evolution.

This transformation is also reshuffling the global geopolitical landscape and industrial value chain. Past competition focused on who could control the most advanced manufacturing capacity; today's battlefield has shifted upstream, centering on who can dominate R&D tools, AI platforms, and system architecture standards. The United States still controls upstream design tools and computing platforms, while China, under trade restrictions, is aggressively pursuing breakthroughs in advanced packaging and system-level innovation.

Faced with this shift, Taiwan's long-standing advantages in foundry services and advanced processes face new challenges. Taiwan has long been accustomed to playing a manufacturing role under architectures defined by Western giants. But as chip competition enters an era driven by AI-powered R&D and battles over system definition rights, merely excelling in manufacturing risks weakening strategic autonomy under the dual pressure of the U.S. and China.

Taiwan must transcend the singular obsession with 'nanometer numbers,' transforming decades of accumulated high-precision semiconductor manufacturing data into fuel for autonomous R&D. We should actively build autonomous semiconductor R&D and design platforms, elevating manufacturing advantages into the power to define system architectures and establishing an industrial ecosystem with 'foundational autonomy.' Only then can Taiwan firmly grasp irreplaceable strategic leadership in the next thirty years of technological waves.

*Author is a retired professor of engineering management, emerging industry commentator, and columnist for TAIA.