Rival AI Systems: Huawei’s CloudMatrix 384 vs Nvidia’s GB200

Huawei has introduced its CloudMatrix 384 AI computing system at Shanghai’s World Artificial Intelligence Conference, being the company’s boldest challenge yet to Nvidia’s dominance in AI hardware.

The system drew crowds at Huawei’s booth during the three-day conference, where companies across the AI sector gathered to showcase their latest developments in machine learning (ML) and neural network technologies. 

For Huawei, the public debut embodied months of anticipation since the company first announced the CloudMatrix in April.

Industry watchers have been closely following the development, viewing it as a direct shot across the bow at Nvidia’s GB200 NVL72 system. 

Simultaneously, Huawei faces ongoing US export restrictions that have cut off access to advanced American semiconductor technologies since 2019.

Now, the CloudMatrix 384 packs 384 of Huawei’s latest 910C chips into a single system, dwarfing Nvidia’s GB200 NVL72 which uses 72 B200 chips. 

But the comparison isn’t simply about chip count – it is fundamentally different approaches to AI computing architecture.

How Huawei’s supernode design compensates for chip limitations

Where Nvidia relies on chip manufacturing and individual processor performance, Huawei has taken a different path. 

The CloudMatrix uses what the company calls “supernode” architecture, enabling ultra-high-speed connections between chips that work together as a cohesive unit.

This design philosophy allows Huawei to compensate for what might be weaker individual chip performance by deploying more processors and leveraging system-level innovations. 

It’s an engineering approach born of necessity, given that US export controls prevent Huawei from accessing the most advanced chip manufacturing processes available to Nvidia.

According to SemiAnalysis, this strategy is paying off. 

The research group reports that Huawei’s system actually outperforms Nvidia’s GB200 NVL72 on certain key metrics, despite relying on older manufacturing technologies. 

The achievement shows how architectural innovation can sometimes trump raw silicon advancement.

Even Nvidia’s leadership has taken notice. 

Jensen Huang, the chip giant’s CEO, acknowledges Huawei’s rapid progress during a Bloomberg interview in May, specifically citing the CloudMatrix as evidence that the Chinese firm had been “moving quite fast.”

The recognition from Nvidia’s CEO carries particular weight, given that his company has dominated the AI chip market and continues to see soaring demand for its products across the globe. 

The bigger picture behind CloudMatrix 384

The CloudMatrix isn’t just a research project or conference showpiece.

Zhang Pingan, CEO of Huawei Cloud, confirmed in June that the system was already operational on the company’s cloud computing platform, making it available to customers who need substantial AI processing power.

This commercial deployment marks a crucial milestone for China’s push toward technological self-reliance. 

Chinese companies developing everything from large language models (LLMs) to computer vision applications require enormous computational resources – and the CloudMatrix offers a domestic alternative to American-made systems.

The timing is important for China’s AI ambitions. 

US export restrictions have effectively blocked Chinese companies from purchasing Nvidia’s most advanced AI chips, creating a significant gap in the market that domestic suppliers like Huawei are rushing to fill.

At the same time, Huawei’s approach shows broader Chinese government priorities around developing indigenous semiconductor capabilities. 

State-backed investment funds have poured billions into AI and chip development projects, viewing these technologies as essential for both economic competitiveness and national security.

As a result, the company has invested heavily in chip design through its HiSilicon division, though it still relies on foreign manufacturers for actual chip production. 

Meanwhile, Taiwan Semiconductor Manufacturing Company and other foundries face their own restrictions on producing advanced chips for Huawei, adding another layer of complexity to the company’s supply chain challenges.

Despite these constraints, Huawei appears to have found a viable path forward through system-level innovation. 

Rather than trying to match Nvidia chip-for-chip, the company has focused on architectural advantages that maximise the collective performance of its available components.

Industry observers see this as potentially transformative for the global AI hardware market.

If Chinese companies can achieve competitive performance through alternative design approaches, it could change how the entire industry thinks about AI system architecture.

The CloudMatrix targets applications spanning natural language processing, computer vision and autonomous vehicle development – all areas where Chinese technology companies are pushing hard to compete globally. 

Success in these markets could validate China’s broader technology strategy.

Zhang’s confirmation that the CloudMatrix 384 was operational on Huawei’s cloud platform indicates the system’s commercial readiness: “The CloudMatrix 384 system was operational on Huawei’s cloud platform,” he says.