The Private BankerSaturday, November 21, 2026

Huawei Scientist Flags Physical Chip Limits Facing Nvidia

Liao Heng, in a four-hour interview streamed in China, highlights constraints on geometric scaling that Western firms including Nvidia will encounter, while praising Huawei’s Tau Scaling Law.

chipset computer cyber hardware
Photo by Jimmy Chan on Pexels.com

Huawei Technologies Co.’s top semiconductor scientist Liao Heng warned of looming physical limits that Western chipmakers including Nvidia Corp. face in developing more powerful processors. In a rare public appearance consisting of a four-hour interview aired in late July and streamed in China, Liao discussed paths forward in chip design, the hardships Huawei encountered after US sanctions severed access to global suppliers about five years ago, and the company’s Tau Scaling Law approach. The comments come as the industry confronts the slowing of traditional geometric scaling, where transistors are repeatedly shrunk to pack more onto silicon wafers.

Liao’s remarks reference the Tau Scaling Law, previously detailed by Huawei semiconductor chief He Tingbo at the IEEE International Symposium on Circuits and Systems in Shanghai in May. That framework shifts focus from shrinking physical transistor sizes to reducing signal latency and time constants through techniques such as LogicFolding, which stacks circuit layers vertically. Huawei has stated the method could deliver transistor densities equivalent to a 1.4-nanometer process by 2031 without relying on extreme ultraviolet lithography equipment restricted by export controls. The company has already applied related principles to hundreds of chips across product lines including Kirin smartphone processors and Ascend AI accelerators.

US export controls since 2019 have limited China’s access to advanced manufacturing tools from suppliers such as ASML Holding NV, leaving domestic production largely at 7-nanometer-class nodes. Huawei’s Ascend series aims to serve as domestic alternatives to Nvidia products for artificial intelligence workloads, with system-level clustering used to offset single-chip performance differences. Nvidia has maintained leadership in high-end AI accelerators available outside restricted markets, while Huawei has expanded output targets for Ascend chips amid continued domestic demand.

Prevailing coverage often frames Huawei’s innovations solely as a response to sanctions that leave the company permanently behind in process technology. Liao’s comments and prior Huawei statements position the physical boundaries of geometric scaling as an industry-wide issue that will confront all major players, including Nvidia, within the next decade regardless of current equipment access. Independent analyses have noted that advanced packaging and 3D stacking techniques similar to those Huawei emphasizes have been explored by firms such as Taiwan Semiconductor Manufacturing Co. for years, suggesting the underlying engineering concepts are not exclusive to the Chinese company.

Huawei has publicly referenced designing and mass-producing more than 300 chips based on Tau-related principles over six years, spanning general-purpose and AI applications. The first commercial LogicFolding implementation, the Kirin 2026 system-on-chip, is scheduled for release later in 2026 and is described as increasing transistor density by more than 50 percent on the same process node used for its predecessor while improving energy efficiency. He Tingbo has previously stated that geometric scaling plateaued for Huawei six years earlier than for unrestricted peers due to sanctions, accelerating the shift to alternative methods.

Enjoyed this article? Access ready-to-use chart frameworks combined with training content and market research to apply it on any market by investing in a membership.