Key points
- Kunlunxin, a Baidu-backed semiconductor firm, won over ¥1 billion in AI chip orders from China Mobile.
- The chips are compatible with Nvidia’s CUDA platform, easing the transition for developers.
- Orders will supply firms like H3C Technologies and ZTE. This win highlights China’s push for domestic alternatives in AI chips.
- Baidu’s Kunlun chips boast better CUDA compatibility than Huawei’s Ascend chips.
Kunlunxin, a Chinese semiconductor design company with significant backing from tech giant Baidu (holding a 59% stake), announced a substantial contract win. The company secured over ¥1 billion (approximately USD 139 million) in orders to supply artificial intelligence (AI) chips for China Mobile’s ambitious AI projects.
This development underscores China’s strategic initiative to reduce reliance on foreign technology, particularly in the crucial AI sector currently dominated by Nvidia.
The chips provided by Kunlunxin are designed with compatibility in mind. Specifically, they are compatible with Nvidia’s CUDA software platform, a crucial aspect for developers already invested in Nvidia’s ecosystem.
This compatibility minimizes the costs and complexities associated with transitioning to domestically produced alternatives, making the switch more appealing to businesses. The orders will be fulfilled through first-tier suppliers to China Mobile, including prominent names like H3C Technologies and ZTE.
Interestingly, tender results from China Mobile also revealed that a Huawei-affiliated company secured a portion of the orders. However, these chips require compatibility with Huawei’s CANN (Compute Architecture for Neural Networks), an equivalent to Nvidia’s CUDA.
This highlights the existing diversity in the Chinese AI chip market and the ongoing competition to establish a robust domestic alternative to Nvidia’s industry-leading GPUs.
Baidu’s commitment to Kunlunxin is substantial. In April, Baidu announced the successful operation of a cluster containing 30,000 of its third-generation P800 Kunlun chips, capable of supporting the training of large AI models.
Baidu’s Kunlun chips, along with other Chinese-made alternatives, are actively competing to gain market share from Nvidia, leveraging their superior CUDA compatibility to attract developers. This strategic advantage reduces the hurdles to adoption compared to options like Huawei’s Ascend chips.
