As China sidelines Nvidia, Huawei and Alibaba step up

• Despite ongoing dema​​nd from major local tech firms, Nvidia’s business opportunities in China appear to have been scuppered by the country’s authorities
• At the same time, Huawei has unveiled its own AI infrastructure systems based on its Ascend NPUs (neural processing units) 
• And China Unicom is deploying domestic AI processors in its Qinghai datacentre

Timing is everything, so the saying goes. Just as the Financial Times (FT) reported that Chinese companies have been barred from buying Nvidia processors, local media has reported that major telco China Unicom is deploying high-performance domestic chips at its datacentre in the Qinghai province and Huawei has unveiled what it claims are the world’s most powerful AI processing ‘superpods’ built using its Ascend NPUs (neural processing units). 

According to the FT, the Cyberspace Administration of China (CAC) told Chinese companies, including ByteDance and Alibaba, to end their testing and orders of Nvidia’s RTX Pro 6000D processor, which was designed specifically for the Chinese market and to conform with US export rules. That is something of a blow for Nvidia, but CEO Jensen Huang told the media in London (where he has been visiting to highlight the tech giant’s investments and partnerships in the UK) that, while he was disappointed by the developments, he acknowledges the ongoing US/Sino trade tension is leading to such decisions. 

Separately, China’s news agency Xinhua reports that Nvidia is now under investigation by China’s State Administration for Market Regulation (SAMR) for anti-competitive practices.

The news came as Reuters, the South China Morning Post and other local news outlets reported that China Unicom, one of the country’s big three telcos, has deployed Chinese processors at its giant datacentre in Xining, the capital of the Qinghai province. According to Reuters, Unicom has deployed thousands of chips developed by T-Head, Alibaba’s semiconductor arm, and has been testing a T-Head high-performance accelerator called the PPU that boasts performance on a par with Nvidia’s H20 GPU (previously the company’s biggest selling product in China) and is more advanced than Huawei’s Ascend 910B processors. 

Adding to the growing intrigue about China’s AI chip development and production capabilities is a Financial Times report, also from this week, which suggests that the country’s largest chip foundry, Semiconductor Manufacturing International Corporation (SMIC), is testing advanced chipmaking equipment, based on a deep-ultraviolet lithography machine made by Shanghai-based startup Yuliangsheng, designed to make AI processors. 

Huawei’s pitch

And there’s more, because Huawei has been holding its Connect 2025 event in Shanghai this week and used the occasion to unveil what it claims is the world’s most powerful superpods and superclusters for AI processing. 

Huawei’s latest superpods (comprising multiple processing machines that operate as a single entity) are the Atlas 950 SuperPoD (with 8,192 Ascend NPUs) and the Atlas 960 SuperPoD (with 15,488 Ascend NPUs). (The vendor did not specify which NPUs, but earlier this year unveiled its 910D processor, though its 910C chip has been more widely deployed.)

Huawei stated: “These two SuperPoDs will deliver an industry-leading performance across multiple key metrics, including the number of NPUs, total computing power, memory capacity and interconnect bandwidth. Based on publicly announced product roadmaps from peers in the industry, these SuperPoDs are currently the most powerful SuperPoDs in the world, and will remain so for years to come.” 

The vendor also announced its Atlas 950 SuperCluster (with more than 500,000 Ascend NPUs) and Atlas 960 SuperCluster (with more than 1 million Ascend NPUs), “which are large-scale computing clusters comprised of multiple Huawei SuperPoDs. These too are poised to outperform all other computing clusters on the market,” it claimed. 

For the networking capabilities required to interconnect these processors and systems, Huawei has developed UnifiedBus, “a groundbreaking interconnect protocol for SuperPoDs,” it added. 

Eric Xu, Huawei’s deputy chairman who is currently heading up the vendor’s executive team, noted that Huawei’s aim is to “meet long-term computing demand by building SuperPoDs and SuperClusters with the semiconductor manufacturing process nodes that are practically available to the Chinese mainland,” noted the vendor. 

“SuperPoDs and SuperClusters powered by UnifiedBus are our answer to surging demand for computing, both today and tomorrow,” stated Xu. 

As it stands, China looks to be on the path to AI infrastructure self-sufficiency, even if the country’s tech giants and telcos would also like to have access to products from Nvidia, AMD and others. 

And while Nvidia’s share price took a hit on the news of the latest setback in China, seasoned tech sector analyst Richard Windsor noted in his latest Radio Free Mobile blog that China’s strategy will ultimately play into Nvidia’s hands and that Jensen Huang’s company will ultimately land more business in other markets than it loses in China – you can check out his informed and entertaining analysis here.

- Ray Le Maistre, Editorial Director, TelecomTV