The artificial intelligence industry is on the brink of a fundamental crisis: the largest chip manufacturers, including Nvidia, are approaching the physical limit of scaling. This is not just a hypothesis, but a conclusion I consider extremely important for understanding the coming years of the market. Huawei's chief scientist for semiconductors, Liao Heng, in an extensive four-hour interview with a Chinese blogger, delivered a harsh diagnosis: the industry has relied for too long on extensive methods—increasing die size, boosting transistor density, and expanding HBM memory volumes.
"There must be a limit to how they scale by using ever-larger compute dies and more HBM capacity. The industry keeps pushing forward, but once it crosses this physical limit, an avalanche will occur," Liao stated. I fully agree with this assessment: the race for "bigger and faster" is already hitting the laws of physics, not engineering ingenuity.
Huawei's Alternative Paradigm
As a solution, Liao promotes the company's own approach—the Tau scaling law. Instead of endlessly shrinking transistors, Huawei proposes reducing the signal transmission time between components of the chip and the computing system. This method was presented in May at an IEEE conference in Shanghai, and according to management, it should help achieve a transistor density equivalent to 1.4 nm by 2031. The key element is LogicFolding, a technology for distributing logic circuits across multiple layers to reduce latency.
Sanction Pressure and Efficiency
The forced isolation due to U.S. restrictions since 2019 and the ban on advanced ASML lithography equipment have pushed Chinese manufacturers to seek alternative paths. Liao emphasizes that the shortage of computing resources is compensated by more sophisticated design: "We need to invest more effort in design to trade higher complexity for lower computing resource consumption." In this context, he highly praised DeepSeek founder Liang Wenfeng, whose success was made possible precisely through architectural solutions under conditions of limited access to computing.
The geopolitical divide, in the scientist's view, inevitably leads to the formation of two isolated technological ecosystems. "To survive, each side must create its own complete production and supply capabilities," he summarizes. This is no longer just a forecast, but a statement of the reality we live in.
My conclusion: Liao's statements are not just corporate rhetoric, but a signal of a paradigm shift. While Nvidia and other leaders fight for every nanometer, Huawei is betting on intelligent architecture. If the Tau Scaling Law proves viable, we will witness not just catch-up development, but the birth of a fundamentally different school of semiconductor design. The market should closely watch this clash of approaches.