Huawei Challenges Paradigms with "Her's Law" for AI Chips

Huawei has unveiled a new chip design principle, dubbed the "Tau Scaling Law" or "Her's Law," named after executive He Tingbo. This initiative comes at a crucial time for Chinese technology companies, which are striving to reduce their reliance on Nvidia's AI accelerators and bolster their autonomy in the sector. The backdrop is one of continuous technological restrictions imposed by the United States, pushing Beijing to invest heavily in developing domestic capabilities.

"Her's Law" is proposed as an alternative to Moore's Law, the long-standing industry benchmark based on transistor miniaturization. While Moore's Law faces increasingly stringent physical limits, Huawei's approach shifts the focus from the size of individual transistors to overall system performance. This includes optimizing the connection and integration between groups of chips within AI clusters and data centers, a fundamental aspect for large-scale deployments.

System-Level Innovation and Energy Efficiency

At the core of the "Tau Scaling Law" is the objective to shorten the time constant in chip operations, with the aim of significantly improving both performance and energy efficiency across multiple layers of chip design. Among the key technologies developed under this approach, Huawei cites LogicFolding, an element that helps optimize data flow. The company has stated that this principle will allow it to achieve transistor density equivalent to 1.4 nanometers by 2031, an ambitious target when compared to Taiwan Semiconductor Manufacturing Co (TSMC)'s projections, which expects to begin 1.4nm chip production in 2028.

The adoption of techniques linked to "Her's Law" is anticipated for phone chips later this year, with full integration into its Ascend series AI accelerators expected by 2030. The primary goal of this methodology is to minimize the time and energy required for data movement both within and between chips. For CTOs and infrastructure architects evaluating self-hosted solutions, the promise of greater energy efficiency and throughput at the system level represents a critical factor for TCO and the sustainability of on-premise deployments.

The Competitive Chinese Landscape and Open Challenges

Huawei's announcement is set against the backdrop of a rapidly expanding and increasingly competitive Chinese AI accelerator market. Companies like Alibaba, with its T-Head chip design unit, are intensifying their efforts. Alibaba recently unveiled the Zhenwu M890 AI chip, outlining a multi-year roadmap for future performance gains. Unlike Huawei, Alibaba is not subject to the same US sanctions, which offers it a different scope of action.

IDC data, reviewed by Reuters, highlights the dynamics of this market. In 2025, approximately 4 million AI accelerator cards were shipped in China. Nvidia maintained its leadership with about 2.2 million units, accounting for a 55% market share. However, Chinese vendors collectively shipped 1.65 million AI chips, capturing about 41% of the market. Huawei led domestic suppliers with 812,000 units, followed by Alibaba T-Head with approximately 265,000, and then Baidu Kunlunxin and Cambricon, each with about 116,000 units. Despite these advancements, Huawei has not yet provided a full explanation of how its approach will address complex engineering issues such as heat dissipation, nor how transistor density projections will translate into real-world AI chip performance.

Implications for Technological Sovereignty and On-Premise Deployments

The push by Huawei and other Chinese companies towards self-sufficiency in the semiconductor sector is a direct response to technological restriction policies. This scenario underscores the importance of data sovereignty and control over the entire technology pipeline, crucial aspects for organizations operating in air-gapped environments or with stringent compliance requirements. The ability to develop and produce AI hardware domestically reduces reliance on external supply chains, mitigating geopolitical risks and ensuring greater resilience.

For decision-makers evaluating on-premise LLM deployments, the emergence of domestic hardware alternatives, although still under development and facing challenges, represents a factor to consider. The availability of a local chip and solution ecosystem can significantly influence TCO, architectural flexibility, and customization capabilities. AI-RADAR offers analytical frameworks to evaluate the trade-offs between self-hosted and cloud solutions, considering factors such as concrete hardware specifications, VRAM requirements, throughput, and latency—elements that "Her's Law" aims to optimize at the system level.