Dr. Robert Castellano's Semiconductor Deep Dive Newsletter

Dr. Robert Castellano's Semiconductor Deep Dive Newsletter

Share this post

Dr. Robert Castellano's Semiconductor Deep Dive Newsletter
Dr. Robert Castellano's Semiconductor Deep Dive Newsletter
Why U.S. Sanctions Fail to Deter China’s AI Chip Progress, as Huawei Fights Back with Ascend 910C

Why U.S. Sanctions Fail to Deter China’s AI Chip Progress, as Huawei Fights Back with Ascend 910C

Dr. Robert Castellano's avatar
Dr. Robert Castellano
Sep 30, 2024
∙ Paid
1

Share this post

Dr. Robert Castellano's Semiconductor Deep Dive Newsletter
Dr. Robert Castellano's Semiconductor Deep Dive Newsletter
Why U.S. Sanctions Fail to Deter China’s AI Chip Progress, as Huawei Fights Back with Ascend 910C
Share
Summary

  • Huawei's AI Push: Huawei defies U.S. sanctions with its Ascend 910C AI chip, rivaling Nvidia and showcasing China's growing tech independence.

  • Nvidia vs. Huawei: Nvidia dominates globally, but Huawei’s AI chip shipments are set to grow domestically, driven by demand from ByteDance and Baidu.

  • China Tech ETF Boost: ETFs like KWEB benefit as Chinese companies increasingly adopt homegrown AI chips like Huawei’s 910C.

  • Nvidia’s Strength: Nvidia remains a top investment, with stock up 178.6% over the past year, proving its resilience despite China’s rising competition.

U.S. Regulations and a Growing Self-Sufficiency of China’s Chip Industry

When the U.S. imposed strict regulations on China's rapidly growing AI market, Nvidia was forced to offer scaled-down versions of its AI accelerators. This created an opening for local companies, who began turning to domestic solutions like Huawei's Ascend 910B chips.

According to a report from the South China Morning Post (SCMP), Huawei has now introduced samples of its new Ascend 910C AI chips to major players in China's AI sector, many of whom are significant Nvidia customers. While full details of the Ascend 910C remain unknown, early indications suggest it provides significant upgrades over its predecessor. Some reports have said it is  comparable to NVIDIA's H100 AI GPU in terms of the raw performance.

The Wall Street Journal (WSJ) has also noted that companies like ByteDance and Baidu are in talks to purchase Huawei’s 910C AI chips. It's rumored that initial orders for 70,000 chips, collectively valued at $2 billion, are being placed. This suggests each chip is priced at around $28,000, a competitive figure for Huawei, given the potential of the 910C to attract significant attention in the market.

The ongoing technological rivalry between the United States and China has escalated dramatically, particularly in the semiconductor sector. The U.S. has aimed to restrict China’s access to advanced chip-making technologies, notably targeting key companies such as Huawei and SMIC (Semiconductor Manufacturing International Corporation). In fact, I first alerted readers and subscribers to the move by SMIC to 7nm without EUV lithography back on May 18, 2022 in an article entitled Applied Materials: SMIC Move To 7nm Node Capability Another Headwind, which I wrote on another publication.

Central to this effort has been limiting Nvidia's AI chips, which are essential for AI development in data centers and autonomous systems worldwide.

However, the strategy may be less effective than anticipated. Despite these restrictions, Huawei has managed to develop its own Ascend 910C AI chip, which competes with Nvidia’s H100. Moreover, China has launched its own countermeasures, sanctioning Nvidia to weaken its position in the Chinese market. This article explores the technological and business developments that have occurred despite these sanctions, focusing on Huawei’s ability to manufacture cutting-edge AI chips and Nvidia’s dominant role in the market.

Huawei’s Ascend 910C – Breaking Through U.S. Barriers

In the face of U.S. sanctions, Huawei has defied expectations by releasing its Ascend 910C AI chip, an upgrade to the previous 910B. This chip is not just a marginal improvement; it marks a major advancement in performance, power efficiency, and AI processing capabilities. The 910C is designed to compete with Nvidia’s H100, demonstrating that Huawei can still innovate despite a lack of access to Western technologies.

The Ascend 910C boasts improved performance in AI tasks, offering significant gains in AI model training and inference. Its power consumption has also been optimized, delivering better energy efficiency compared to the older 910B. These advancements suggest that Huawei is not only capable of matching Nvidia’s AI performance domestically but is also positioning itself to be a major player in China’s AI ecosystem.

Manufacturing and Nodes – SMIC’s Role in Huawei’s Success

Huawei’s chips, including the Ascend 910C, are manufactured by SMIC, which is China’s largest chipmaker. Despite U.S. restrictions that block access to critical semiconductor equipment like Extreme Ultraviolet (EUV) lithography machines, SMIC has managed to produce 7nm chips using Deep Ultraviolet (DUV) lithography. The Ascend 910C is fabricated using SMIC’s N+1 process, which is equivalent to a 7nm node, demonstrating China’s growing independence from U.S. semiconductor technology. This development highlights how China's domestic supply chain is advancing to keep pace with global leaders in chip manufacturing.

While SMIC’s ability to produce these chips at 7nm represents a significant technical milestone, it is important to note that the lack of EUV lithography limits SMIC’s ability to scale down to smaller nodes such as 5nm and beyond. However, this capability at 7nm is still sufficient to produce powerful AI chips like the Ascend 910C, which are essential for China's AI industry.

Nvidia’s Dominance in the Global AI Market

While Huawei is making strides in AI chip development, Nvidia remains the global leader, holding a 95% market share in the AI chip space as of 2023. Nvidia’s H100 AI GPU, which is built on the Hopper architecture, is considered the gold standard for AI tasks, offering unmatched performance in deep learning and large-scale AI models.

The Nvidia H100 is a powerhouse, delivering 700 TFLOPS of FP8 performance with 80GB of high-bandwidth memory. Its ability to handle large datasets and transformer-based models makes it the AI processor of choice for cloud providers, data centers, and autonomous systems worldwide. By contrast, while Huawei's Ascend 910C is competitive, it still lags in terms of memory bandwidth and power consumption, which may limit its broader adoption outside China.

Shipment Forecasts – Nvidia and Huawei AI Chips (2022-2027)

Despite U.S. sanctions, Huawei is rapidly scaling its AI chip production, particularly for the domestic market. However, Nvidia remains the dominant player globally, with shipments expected to far outpace Huawei’s in the near term. Table 3 below outlines projected shipments for both Nvidia and Huawei AI chips from 2022 through 2027.

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 Dr. Robert Castellano
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture

Share