China’s iFlytek says local chips delay AI development by 3 months

Liu Qingfeng, chairman of the Chinese voice recognition firm iFlytek, said that reliance on domestically produced chips may extend the development of AI models by three months.

This delay is due to the additional computing resources required compared to using Nvidia chips, which have a more established software ecosystem.

In a statement on WeChat, Liu indicated that iFlytek is using local chips, including Huawei’s Ascend 910B, to reduce risks associated with imported semiconductors.

The company has been unable to acquire advanced US chips since being added to the US Entity List in 2019.

🔗 Source: South China Morning Post

🧠 Food for thought

1️⃣ The widening performance gap illustrates China’s semiconductor self-sufficiency challenges

Liu’s admission of a three-month delay in AI model development when using domestic chips highlights the persistent technological gap despite massive investments.

China currently produces only 16% of semiconductors used domestically, far short of its goals of 40% by 2020 and 70% by 2025 under national initiatives 1.

The Chinese government has committed approximately $118 billion to semiconductor development over five years, yet technical gaps remain substantial in advanced chip production 1.

This market reality is reflected in 2024 sales figures, where Nvidia sold over 1 million H20 chips in China while Huawei sold only 200,000 Ascend chips, demonstrating the continued dominance of US technology despite restrictions 2.

iFlytek’s experience with Huawei’s Ascend 910B chip improving from 25% to 73% efficiency compared to Nvidia’s A800 shows progress, but confirms Chinese firms still face significant challenges in catching up within a critical technology sector.

2️⃣ Software ecosystem limitations compound hardware challenges for Chinese chips

Liu’s explanation that extended development time stems partly from immature software ecosystems reveals a critical second front in China’s semiconductor independence battle.

While hardware improvements are being made, Nvidia’s comprehensive software stack remains highly preferred by developers, creating adoption barriers for domestic alternatives even when hardware improves 2.

This software gap helps explain why Chinese companies continue seeking Nvidia chips through various channels despite government pressure to adopt domestic alternatives and despite the improving performance of chips like the Ascend 910B 3.