The semiconductor industry has come close to a physical barrier beyond which traditional methods of increasing AI chip performance will cease to work. This statement was made by Huawei's chief scientist for semiconductors, Liao Heng, during an extensive four-hour interview dedicated to the future of computing.

In my analysis, this is one of the most candid admissions of a systemic crisis in the industry in recent years. Liao directly points out that the reliance on increasing die area, transistor density, and HBM (High Bandwidth Memory) capacity has exhausted itself. HBM, which is critically important for AI accelerators, provides high-speed data exchange, but it too is hitting physical limits.

"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," the scientist warns.

An Alternative Path: Tau Scaling Law

In contrast to the traditional approach, Huawei is promoting its own concept — the Tau Scaling Law. This scaling law shifts the focus from transistor miniaturization to reducing signal transmission time between chip components and the computing system as a whole. In May, at the IEEE ISCAS conference in Shanghai, Huawei's head of semiconductor business, He Tingbo, stated that this technology would allow the company to achieve transistor density equivalent to 1.4nm by 2031.

A key element of this strategy is LogicFolding. It involves distributing critical logic circuits across multiple layers, which radically reduces latency. According to my data, Huawei is already preparing to unveil its first smartphone chip created using this technology, which will be a crucial real-world test of the concept.

Sanction Pressure and Efficiency

The forced search for alternative paths is a direct response to the U.S. restrictions in effect since 2019. Chinese manufacturers are denied access to ASML's advanced lithography equipment, making it impossible to produce chips using extremely thin process nodes. Liao emphasizes that under these conditions, Chinese companies must compensate for the shortage of computing power through more complex architectural design.

"We need to invest more effort in design to trade higher complexity for lower consumption of computing resources," he explained.

In this context, the scientist highly praised the work of DeepSeek founder Liang Wenfeng, whose success, in his opinion, is directly linked to the need to find innovative architectural solutions under conditions of limited access to computing.

The Formation of Two Ecosystems

Liao also touched on an inevitable geopolitical scenario — the complete separation of technology supply chains. In his view, each side will be forced to create its own self-sufficient production infrastructure, even if no open conflicts occur.

"To survive, each side must build its own complete manufacturing and supply capabilities," he stated.

It is worth recalling that in June, Huawei already promised to double the performance of its Ascend AI chips annually. This statement, combined with the new technological philosophy, indicates the company's confidence in its ability not just to catch up, but to redefine the rules of the game. Earlier, Google DeepMind CEO Demis Hassabis admitted that China has nearly caught up with the U.S. and the West in the field of AI.

My comment: Liao Heng's warning is not just a technical discussion, but a signal of a paradigm shift. If the Tau Scaling Law truly works, we will witness not catch-up development, but the formation of a fundamentally different technological trajectory that could make the traditional transistor race secondary. Investors and analysts should closely monitor the first commercial products based on this architecture — they will define the contours of the post-silicon era.