The chief scientist of Huawei Technologies’ semiconductor division has offered a rare and defiant glimpse into the company’s strategy for overcoming crushing United States export controls, asserting that Western chipmakers are increasingly trapped by their reliance on legacy software ecosystems.
In a viral interview hosted by a mainland business podcast, Liao Heng, the chief architect behind Huawei’s Ascend artificial intelligence processors, detailed how the company transitioned from initial skepticism to developing a self-sustaining computing infrastructure. According to the South China Morning Post, Liao argued that evaluating China’s semiconductor industry solely through the lens of nanometer fabrication metrics misses a fundamental shift toward system-level architectural innovation.
Liao’s background lends significant weight to his technical assessments. A former child prodigy who entered Beijing’s prestigious Tsinghua University at age 14 in 1987, he subsequently pursued postdoctoral research at Princeton University and spent over a decade working in the US tech sector. He noted that his time abroad bred a “blind arrogance” that only Western firms were capable of true innovation.
However, upon joining Huawei’s HiSilicon chip-design unit in 2016, his perception changed rapidly. “Reality taught me a quick lesson,” Liao stated, observing that HiSilicon routinely finalized designs for dozens of new chips annually without a single failure to reach mass production. “Such a high hit rate shows that every team [at the company] is highly qualified.”
(Related: Huawei’s Tau Scaling Law: A New Path Amid US Sanctions)
The Dong Cunrui Mindset Under Sanctions
The true test of Huawei’s engineering capabilities arrived in 2019 and 2020, when Washington severed the company’s access to advanced chip-design software and global foundry giants like Taiwan Semiconductor Manufacturing Company (TSMC). Forced to rapidly redefine its product roadmap, Huawei transitioned its pipeline from the pre-sanction Ascend 910 to the post-sanction Ascend 950 processors.
Describing the intense psychological pressure inside the company during this period, Liao compared Huawei’s attitude to Dong Cunrui, a legendary Chinese war hero who sacrificed himself to destroy an enemy bunker. “Imagine if you were Dong Cunrui holding up the explosive pack,” Liao said. “You don’t think about your own feelings. You face a difficulty, and you just want to solve it.”
Cut off from the world’s most advanced extreme ultraviolet (EUV) lithography equipment, Huawei was forced to abandon the traditional path of shrinking transistors and instead focus on macro-level engineering. The company relied heavily on system architecture, stacked dies, optoelectronic interconnects, and its open-source Compute Architecture for Neural Networks (CANN) software toolkit to link thousands of processors into unified supernodes. These architectural innovations culminated in the Tau Scaling Law, a design framework unveiled by Huawei in May, which claims the company can achieve transistor densities equivalent to a 1.4-nanometer process by 2031, effectively bypassing US hardware bottlenecks.
Bypassing the CUDA Trap
While Liao acknowledged the existing technical gaps between the US and China in semiconductor manufacturing, he argued that the real bottleneck is software, specifically Nvidia’s proprietary CUDA toolkit. However, he issued a stark warning to domestic developers: attempting to clone CUDA is a trap that guarantees a perpetual second-place finish. Instead, Huawei is focusing on close hardware-software co-design to create entirely new computational paradigms. “The protective wall of the CUDA ecosystem is rapidly disappearing, as it is no longer the primary interface,” Liao asserted, suggesting that the industry is moving toward higher-level abstractions where underlying hardware dependencies are less relevant.
Liao specifically praised domestic AI innovators like DeepSeek for their ability to leverage sparse activation and algorithmic complexity to drastically reduce compute requirements. By optimizing algorithms for existing, less advanced hardware, Chinese developers are achieving world-class results despite severe constraints. Liao also pointed to China’s structural advantage in energy supply as a critical factor in the long-term semiconductor race. “The main gap between Chinese semiconductors and TSMC’s is energy cost,” he explained. “While our power consumption per chip is slightly higher, China has an abundance of energy,” allowing domestic foundries to offset efficiency losses with raw power.
(Related: China’s Domestic Chipmakers Seize 41% of the Local AI Market as Nvidia’s Grip Loosens)
The Debate Over System-Level Innovation
Despite Huawei’s confident posturing, the global semiconductor industry remains divided on the long-term viability of the Tau Scaling Law and the company’s system-level approach. Analysts point out that while architectural workarounds can compensate for a lack of EUV lithography in the short term, they introduce massive new engineering challenges. Linking thousands of less efficient chips into supernodes creates severe hurdles in heat management, manufacturing yields, and advanced 3D packaging. Global rivals equipped with the world’s most advanced fabrication tools may ultimately retain an insurmountable edge in pure performance and energy efficiency.
Nevertheless, Liao’s interview provides a clear articulation of China’s semiconductor strategy in the face of enduring US containment. Rather than attempting to beat Silicon Valley at its own game of nanometer shrinkage, Huawei is attempting to change the rules of the contest entirely. By prioritizing system architecture, algorithmic efficiency, and software independence, the company is building a parallel technological ecosystem that is fundamentally resilient to external shocks. Whether this “war-hero” mindset can sustain Huawei’s momentum over the next decade will depend on its ability to continually innovate around the physical limitations of its restricted hardware supply.