Will DeepSeek Change China’s AI Game?
News Desk
Islamabad: The emergence of DeepSeek’s artificial intelligence (AI) models is seen as a potential game-changer for Chinese chipmakers like Huawei, offering them a stronger foothold in the domestic market against more powerful US processors.
For years, Huawei and other Chinese semiconductor firms have struggled to rival Nvidia’s dominance in high-end chips used for training AI models. Training involves feeding data to algorithms, enabling them to learn and make accurate predictions.
However, DeepSeek’s AI models prioritize computational efficiency over sheer processing power, focusing on the inference stage—where AI models generate conclusions rather than being trained.
Analysts believe this approach could help narrow the gap between China’s AI processors and their more advanced U.S. counterparts. In recent weeks, Huawei and other Chinese AI chipmakers, including Hygon, Tencent-backed EnFlame, Tsingmicro, and Moore Threads, have announced support for DeepSeek models, though specific details remain scarce.
While Huawei declined to comment, Moore Threads, Hygon, EnFlame, and Tsingmicro did not respond to queries from Reuters seeking further clarification.
Industry experts predict that DeepSeek’s open-source nature and lower costs will encourage AI adoption and the development of real-world applications. This, in turn, could help Chinese firms mitigate the impact of US export restrictions on high-end AI chips.
Even before DeepSeek gained prominence, Huawei’s Ascend 910B chip was already regarded by companies such as ByteDance as more suitable for inference tasks rather than computationally intensive AI training. Many Chinese companies, spanning sectors from automotive to telecommunications, have announced plans to integrate DeepSeek’s models into their operations.
“This development aligns with the strengths of Chinese AI chipset manufacturers,” said Lian Jye Su, chief analyst at tech research firm Omdia. “While Chinese AI chipmakers struggle to compete with Nvidia’s GPUs in AI training, inference workloads require more localized and industry-specific optimization, making them a viable alternative.”
Nvidia Maintains Global Dominance
Despite these advancements, Nvidia continues to dominate the AI chip sector. According to Bernstein analyst Lin Qingyuan, while Chinese AI chips are cost-effective for inference tasks, their appeal remains largely confined to the domestic market, as Nvidia’s chips still outperform even in this domain.
Although US export controls prevent Nvidia from selling its most advanced AI training chips to China, the company is still allowed to supply less powerful variants that Chinese firms can use for inference applications. In a recent blog post, Nvidia emphasized that inference workloads are becoming increasingly demanding and argued that its chips remain essential for maximizing the effectiveness of AI models like DeepSeek.
A key factor in Nvidia’s continued dominance is CUDA, its proprietary parallel computing platform that enables software developers to optimize GPU performance beyond AI and graphics applications. Many Chinese AI chip companies have historically avoided direct competition with Nvidia by ensuring compatibility with CUDA.
Huawei, however, has taken a more aggressive approach by introducing Compute Architecture for Neural Networks (CANN), a CUDA alternative. However, industry experts note that persuading developers to switch remains a challenge.
“Chinese AI chip firms still lag in software performance,” Omdia’s Su explained. “CUDA offers an extensive library and a broad range of software capabilities, which have been built through long-term investment.”
As the AI race between China and the US intensifies, DeepSeek’s models may provide Chinese chipmakers with a competitive edge in inference computing. However, overcoming Nvidia’s entrenched software ecosystem and global market dominance remains a formidable challenge.
Additional input from Reuters.
