The US Commerce Department has announced new guidelines that classify the use of Huawei Technologies’ Ascend chips “anywhere in the world” as a violation of US export controls.

This marks the first time Huawei’s AI chips have been explicitly identified in a US government document.

The updated regulations replace the Biden administration’s “AI diffusion” framework. Under the new rules, global tech companies must now choose between using Chinese or US hardware.

Analysts warn this change could further deepen the tech divide between the two nations. Huawei, a key player in China’s AI sector, has not publicly commented on the new restrictions.

China has not officially responded to this policy change. Public discussions about Huawei’s Ascend chips are currently limited within the country.

🔗 Source: South China Morning Post

🧠 Food for thought

1️⃣ Export controls have accelerated China’s technological self-sufficiency

US sanctions against Huawei have inadvertently fast-tracked China’s efforts to reduce reliance on foreign technology, particularly in semiconductors.

When the US first blacklisted Huawei in May 2019, it triggered the company’s mobilization of engineers to enhance production capabilities of local chip manufacturers, described as a “wartime-like” response to ensure survival 1.

This extends beyond Huawei to China’s broader “Made in China 2025” initiative, which aims to boost domestic production and innovation in advanced technology sectors 2.

The response has produced tangible results. Despite restrictions, Huawei has developed the Ascend AI chip series that can reportedly deliver 60-70% of Nvidia’s H100 performance at a significantly lower cost 3.

This cycle of restrictions prompting greater investment in domestic alternatives may undermine the long-term effectiveness of export controls.

2️⃣ Huawei’s competitive strategy focuses on system architecture over single-chip performance

Rather than competing on a chip-to-chip basis with Nvidia, Huawei has developed a distinctive approach focusing on scale and system-level architecture.

Huawei’s CloudMatrix 384 system connects 384 Ascend 910C chips in a full mesh network, reportedly delivering 300 PFLOPs of dense BF16 compute – nearly double the performance of comparable Nvidia systems 4.