Huawei begins mass shipment of Ascend 910C AI chip to fill Nvidia void in China

Back in August 2024, The Wall Street Journal reported that Huawei was preparing to launch a new AI chip called the Ascend 910C, aimed at challenging Nvidia’s dominance in China.

Fast forward eight months, and Huawei is now preparing to mass-ship the Ascend 910C to Chinese customers. According to a Reuters report published Monday, Huawei plans to begin large-scale deliveries as early as next month—some shipments have already gone out.

Huawei’s 910C AI Chip Set for China Shipment Amid Nvidia Dip

The 910C had already been undergoing testing with several major tech firms at the time, with early signals suggesting it could match—or even outperform—Nvidia’s H100 in both performance and efficiency. Huawei had been quietly briefing potential clients, confident the chip would become a key turning point in China’s AI race.

“Huawei Technologies plans to begin mass shipments of its advanced 910C artificial intelligence chip to Chinese customers as early as next month,” Reuters reported, citing two people familiar with the matter. “Some shipments have already been made, they added.”

The timing couldn’t be more strategic. The U.S. has tightened export restrictions on Nvidia’s H20 chip, cutting off a crucial supply for China’s top tech companies—and opening the door for a serious domestic alternative.

A Challenger to Nvidia’s AI Reign

The Ascend 910C is Huawei’s latest GPU built for AI training and inference tasks. It fuses two of the older 910B processors into a single unit, using some clever packaging to bump up performance. Industry watchers say it delivers around 800 TFLOP/s at FP16 and pushes 3.2 TB/s in memory bandwidth. That puts it close to 80% of Nvidia’s H100 performance—more than enough for Chinese companies looking for alternatives.

The chip is built on a 7nm process, mostly sourced from Taiwan’s TSMC, with China’s SMIC stepping in as well. Huawei had stocked up on TSMC orders before sanctions hit, and SMIC is reportedly scaling up its own production. Together, they’re expected to support Huawei’s plan to ship millions of chips.

CloudMatrix 384: Huawei’s AI Cluster Play

Huawei isn’t just selling chips—it’s bundling them into a full-scale system. The CloudMatrix 384 AI cluster includes 384 Ascend 910C units and delivers a reported 300 petaflops of compute. That outpaces Nvidia’s GB200 NVL72 system, which hits around 180 petaflops. The trade-off? Higher energy consumption. But in China, where electricity is cheap and high-end chips are harder to get, that’s a compromise most customers are willing to make.

U.S. Curbs, China’s Opening

Just early this month, the U.S. government rolled out new restrictions that force Nvidia to get licenses for exporting the H20 chip to China. The H20 had been Nvidia’s go-to AI chip for the Chinese market. Now it’s stalled. That’s left companies like Baidu, Tencent, and ByteDance scrambling—and Huawei is already filling the gap.

Huawei began sampling the 910C to major tech firms back in late 2024, and it’s already fulfilling early orders. Analysts say the shift is significant. Paul Triolo from Albright Stonebridge Group said, “Huawei’s Ascend 910C GPU will now become the hardware of choice for Chinese AI model developers.” Nori Chiou from White Oak Capital echoed that, noting the export rules are “effectively pushing Nvidia’s Chinese customers toward Huawei’s AI chips.”

Nvidia Feels the Pressure

Nvidia’s dominance in AI isn’t going away anytime soon, but the Chinese market is taking a hit. After the latest restrictions, Nvidia shares dipped 3.1% in premarket trading. The company took a $5.5 billion charge tied to H20 exports. CEO Jensen Huang admitted that revenue from China has fallen to half of what it was before the controls kicked in.

Huawei, meanwhile, is making real progress. Huang himself acknowledged Huawei’s rising presence, calling it a serious rival in a recent Financial Times interview.

Chipping Away at Dependence

The 910C is central to China’s bigger plan to reduce reliance on foreign AI chips. As Nvidia’s most advanced processors get harder to buy, Chinese companies are racing to develop homegrown alternatives. Huawei has already improved yields on its Ascend line by 40% since last year and aims to produce 100,000 910C units and 300,000 910Bs in 2025.

On the software side, Huawei’s platform still trails Nvidia’s CUDA in terms of maturity, but it’s catching up. Partnerships with AI firms like DeepSeek are helping, and some believe this collaboration could pose a long-term challenge to Nvidia’s dominance globally.

A Few Bumps Ahead

Still, not everyone is sold. Critics say Huawei’s chip strategy depends too much on brute force—stacking more processors instead of optimizing performance. That leads to higher power usage and bigger systems. Others are cautious, calling the 910C “PowerPoint silicon” until it proves itself with widespread deployment and real-world benchmarks.

There’s also the issue of supply. Huawei still relies partly on foreign chipmakers like TSMC. If the U.S. tightens restrictions further, that could put pressure on the whole operation.

What’s Next?

Huawei is walking through an open door left by Nvidia’s export issues—and it’s doing it with confidence. The 910C may not beat Nvidia’s chips spec-for-spec, but it’s good enough for now. More importantly, it’s available.

If Huawei can build out its supply chain, improve its software stack, and deliver real-world performance at scale, it might be more than just a backup option. It could be the beginning of a real shift in the AI hardware race, especially in markets cut off from U.S. tech.