The advancements of large AI chips from Nvidia and other manufacturers are nearing their physical scaling limits, according to Liao Heng, Huawei's chief semiconductor scientist, who shared his insights in a four-hour interview with a Chinese blogger, as reported by Bloomberg.

Liao emphasized that the industry has relied too heavily on increasing chip sizes, transistor density, and high-bandwidth memory (HBM) capacity, which is crucial for fast data exchange between processors and computing systems.

"There must be a limit to how much they can scale through larger computational chips and more HBM. The industry keeps pushing forward, but once it crosses that physical boundary, it will trigger an avalanche," he stated.

Huawei's Alternative Approach

In response to these challenges, Liao advocated for Huawei's Tau Scaling Law, which shifts the focus from continually reducing transistor sizes to minimizing signal transmission times between chip components and computing systems.

Huawei introduced this approach at a conference by the Institute of Electrical and Electronics Engineers in Shanghai in May. At that time, He Tingbo, head of Huawei's semiconductor business, stated that this technology could enable the company to achieve transistor densities equivalent to 1.4 nm by 2031.

One component of this approach is LogicFolding, which Huawei describes as a method for distributing critical logic circuits across multiple layers to reduce signal transmission delays. According to Bloomberg, Huawei is preparing to unveil its first smartphone chip developed using this technology.

Sanctions and the Drive for Efficiency

Due to U.S. restrictions, Huawei is compelled to explore alternative solutions. Since 2019, the company has lost access to certain Western technologies, and Chinese manufacturers face challenges in acquiring advanced lithography equipment from ASML for producing cutting-edge chips.

Liao linked the new approach not only to chip manufacturing but also to the architecture of AI models. He noted that Chinese firms must compensate for a lack of computational resources with more sophisticated designs.

"We need to invest more effort into design, trading off higher complexity for lower computational resource consumption," he remarked.

In this context, Liao praised Lyan Wenfan, founder of DeepSeek, asserting that the company's success stems from its necessity to seek architectural solutions for model training amid limited computational access.

Two Technological Ecosystems

Liao further asserted that, amid geopolitical divisions, each side will need to establish its own manufacturing and technological supply chains.

"To survive, each side must create its own complete manufacturing and supply capabilities, even without serious confrontations between the two sides," he stated.

According to Bloomberg, Liao's remarks signal Huawei's confidence following the introduction of the Tau Scaling Law in May. The company aims to demonstrate that U.S. restrictions have not only pushed it to seek alternatives but could also steer the Chinese semiconductor industry onto a distinct path of development.

In June, Huawei promised to release a new generation of Ascend AI chips annually, doubling their performance each time.

Previously, Google DeepMind CEO Demis Hassabis suggested that China is nearly catching up to the U.S. and the West in AI technology.