“We are confident in our ability to provide competitive AI computing solutions,” a Huawei spokesperson recently told Reuters. That’s quite the understatement when you consider ByteDance and Alibaba are now lining up to place orders for Huawei’s Ascend 910C chips.
This isn’t just another tech procurement story. It’s a seismic shift in China’s AI infrastructure that could reshape the global semiconductor space for years to come.
The Nvidia Problem
Here’s what’s happening. U.S. export controls have effectively locked Chinese tech giants out of Nvidia’s latest AI accelerators—the H100 and A100 chips that power most of the world’s AI training. These restrictions, tightened in 2022 and again in 2023, left companies like ByteDance and Alibaba scrambling for alternatives.
Enter Huawei’s Ascend 910C.
The chip reportedly matches the performance of Nvidia’s A100, which might sound underwhelming until you realize the A100 launched in 2020. But when you’re cut off from the latest hardware, three-year-old performance suddenly looks pretty good. ByteDance, the company behind TikTok and its Chinese counterpart Douyin, is reportedly planning to order chips worth over $1 billion. Alibaba isn’t far behind.
Why This Matters More Than You Think
China’s AI ambitions haven’t exactly been subtle. The country aims to become the world leader in AI by 2030, and you can’t build that future on imported chips you might lose access to tomorrow. Huawei’s Ascend series represents something more significant than a workaround—it’s the foundation of technological sovereignty.
The numbers tell the story. China’s domestic chip production has surged 40% year-over-year, according to recent industry reports. Huawei’s semiconductor division, HiSilicon, has been working overtime to fill the void left by Nvidia’s absence. The Ascend 910C uses a 7-nanometer process, which is admittedly behind TSMC’s 3nm chips powering the latest iPhones, but it’s functional and, crucially, available.
ByteDance’s interest is particularly telling. The company trains massive language models for its AI products, including Doubao, its ChatGPT competitor. These workloads demand serious computational horsepower. If Huawei’s chips can handle ByteDance’s requirements, they can probably handle most commercial AI applications.
The Catch
Let’s not pretend this is a perfect solution. Huawei’s chips still lag behind Nvidia’s latest offerings by a considerable margin. The H100, which Chinese companies can’t access, offers roughly 3x the performance of the A100. That gap matters when you’re training frontier models that cost tens of millions of dollars in compute.
There’s also the software ecosystem. Nvidia’s CUDA platform has been the industry standard for over 15 years. Developers know it. Tools support it. Switching to Huawei’s CANN (Compute Architecture for Neural Networks) means rewriting code, retraining engineers, and accepting some performance penalties during the transition.
But here’s the thing about necessity: it’s a hell of a motivator.
What Comes Next
The real question isn’t whether Huawei can compete with Nvidia today—it clearly can’t, not at the highest end. The question is how quickly that gap closes. Chinese tech companies are pouring billions into domestic chip development. SMIC, China’s largest foundry, is pushing toward more advanced process nodes despite equipment restrictions. The ecosystem is maturing faster than most Western observers expected.
For Nvidia, this represents a slow-motion crisis. China accounted for roughly 20% of its data center revenue before the export controls. That’s billions in annual sales that won’t come back, even if restrictions ease. The company is trying to develop China-compliant chips, but it’s threading a needle between U.S. regulations and commercial viability.
Meanwhile, ByteDance and Alibaba are making a calculated bet. They’re accepting slightly inferior performance today in exchange for supply chain security tomorrow. It’s the same logic that drove Europe to diversify away from Russian gas, just playing out in silicon instead of pipelines.
The AI chip market is fracturing along geopolitical lines, and Huawei’s Ascend series is becoming the standard bearer for one side of that divide. Whether it can truly challenge Nvidia’s technical supremacy remains uncertain, but in a world where access trumps performance, being good enough might just be good enough.
🕒 Published: