Huawei AI Servers Test Effectiveness of US Curbs
Sign up for ARPU: Stay ahead of the curve on tech business trends.
Despite stringent US export restrictions aimed at slowing its technological progress, China’s Huawei is showcasing advancements in its artificial intelligence infrastructure. Last month, Huawei unveiled a new AI server cluster in China’s Anhui province powered by its in-house Ascend chips, not the dominant GPUs from NVIDIA. This development, alongside reports of performance gains and a growing domestic ecosystem, raises questions about whether US curbs are effectively hindering China’s AI ambitions or merely accelerating its drive for self-sufficiency.
What were the US export controls designed to achieve?
Beginning in October 2022, the US government implemented broad export controls targeting China’s access to advanced computing semiconductors and semiconductor manufacturing equipment. The stated goal was to prevent China from acquiring and manufacturing chips that could be used to modernize its military, develop advanced AI capabilities, and gain a technological edge over the United States. These restrictions initially focused on high-performance AI chips (like certain Nvidia GPUs) and the most advanced manufacturing tools (like ASML’s EUV machines). They have since been expanded to include even slightly less capable chips designed specifically for the China market. The strategy aimed to create a “chokehold” on critical technologies required for leading-edge AI development and production.
What progress is China making despite these restrictions?
The recent Huawei unveiling in Anhui province is one example of China’s continued push. The new server cluster utilizes Huawei’s self-developed Ascend chips, specifically the Ascend 910 series. While estimates vary, comparisons by firms like SemiAnalysis suggest that by leveraging a large number of Ascend chips in a cluster (reportedly five times as many as in a comparable Nvidia system), Huawei’s system can outperform rival offerings like Nvidia’s GB200 cluster on some metrics, despite individual Ascend chips potentially having less raw performance. Huawei is also building a software ecosystem, MindSpore, and training a large developer base (over 40,000 engineers trained) to support its hardware.
Adding to this picture, Chinese companies are making strides in chip design. Recently, smartphone giant Xiaomi announced it has begun mass production of its self-designed 3-nanometer system-on-a-chip (SoC), the XRING 01, positioning itself among a handful of global tech leaders capable of designing such advanced mobile processors. The chip reportedly packs billions of transistors and aims to compete with top-tier mobile chips from Apple and Qualcomm. Crucially, this 3nm chip is likely manufactured outside of mainland China, as domestic foundries are currently unable to mass produce chips at this advanced node due to US equipment restrictions. This highlights China’s continued strength in chip design and its ability to leverage foreign manufacturing where controls permit, particularly for consumer-focused products.
Chinese foundries like SMIC are also improving their capabilities on less advanced nodes (like 7nm), which can be used for a range of applications and even for relatively powerful chips, albeit less efficiently than leading-edge processes.
How is China achieving this advancement under restrictions?
China’s strategy involves a multi-pronged approach:
- In-house Design: Companies like Huawei and Xiaomi are investing heavily in designing their own chips, building domestic intellectual property and design expertise across various segments (AI accelerators, mobile SoCs).
- Leveraging Accessible Manufacturing: While cut off from the most advanced foreign foundries for its top AI designs, China is improving its domestic manufacturing capabilities on nodes still accessible with older equipment (e.g., SMIC’s 7nm production), or leveraging foreign foundries for chips not explicitly restricted (like Xiaomi likely using TSMC for its 3nm consumer chip).
- System-Level Innovation: Compensating for potentially lower individual chip performance by focusing on integrating larger numbers of chips effectively in server clusters and optimizing interconnects and software at the system level (Huawei’s approach).
- Ecosystem Building: Developing alternative software frameworks (MindSpore vs. CUDA) and nurturing a domestic developer community.
- Potential Circumvention: Reports have also alleged attempts to acquire restricted technology or chips through indirect means, although companies emphasize their in-house capabilities.
What does this mean for the effectiveness of US export controls?
Huawei’s progress suggests that the US restrictions, while undoubtedly creating significant challenges and delays for China, are not a complete roadblock. They appear to be forcing China to double down on domestic R&D and build parallel supply chains. This could potentially lead to a bifurcated global tech ecosystem, where China develops its own standards and capabilities separate from the US and its allies.
Nvidia CEO Jensen Huang has publicly questioned the curbs, stating they have caused Nvidia’s market share in China to drop significantly (from 95% to 50%), arguing that restricting US companies allows rivals like Huawei to gain ground domestically. This view is shared by some within the US administration who argue for a more nuanced approach that balances security with maintaining US market presence to influence global tech development.
Others maintain that slowing China’s access, even if temporarily, is necessary for national security, arguing that the US technological lead, particularly in manufacturing tools from companies like ASML and cutting-edge foundry processes from TSMC, remains substantial and is the ultimate chokepoint.
What are the implications for the US-China tech rivalry?
The advancements by Huawei underscore that the tech rivalry is an escalating competition. China is demonstrating resilience and a commitment to developing domestic alternatives where access to foreign technology is restricted. The debate over the effectiveness of the controls is likely to continue influencing US policy as both superpowers vie for leadership in critical future technologies like AI.
Reference Shelf:
Staff Writer
ARPU
ARPU