The birth of DeepSeek, the 'cost-effective AI monster,' is expected to accelerate the development of AI semiconductors in China. Despite technical limitations, DeepSeek's cost-effectiveness strategy, which achieves maximum results through resource intensiveness and efficient allocation, can be similarly applied across China's AI semiconductor development.
DeepSeek built a cost-effective AI infrastructure by utilizing low-spec GPUs (graphics processing units) such as Nvidia H800 chips in parallel. Unlike traditional AI model development, which heavily relied on high-performance semiconductors, DeepSeek has attracted attention by implementing ChatGPT-like performance at minimal cost through software parallel computation optimization technology.
This strategy aligns with the approach Chinese semiconductor companies have taken to avoid falling behind in the global AI competition despite U.S. advanced semiconductor export restrictions.
Chinese Big Tech Accelerates AI Chip Development to Break Free from Nvidia Dependence
China's AI semiconductor industry is experiencing rapid growth due to a combination of U.S. sanctions, concentrated government investment, big tech companies' research and development (R&D), and price competitiveness.
The Chinese government has prioritized AI chip development to increase semiconductor self-sufficiency and has strengthened support for the domestic semiconductor industry. Chinese big tech companies such as Huawei, Alibaba, and Baidu have concentrated their entire corporate capabilities on developing their own AI chips in response. As a result, a Chinese AI semiconductor ecosystem has emerged that, while offering similar or slightly lower performance, is significant in terms of price and volume.
Huawei has achieved the most notable results in the AI semiconductor field. Its self-developed Ascend series offers performance comparable to Nvidia's H100 while maintaining high price competitiveness, challenging Nvidia's monopoly in the Chinese market. Additionally, Huawei has developed AI chips based on a 7nm process and is mass-producing them in cooperation with Chinese foundry company SMIC. It is also accelerating the development of chips optimized for AI inference tasks.
Alibaba is also actively developing AI semiconductors. The 'Hanguang 800' chip, unveiled in 2019, demonstrated performance 15 times higher than Nvidia's T4 GPU and 46 times higher than the P4 GPU. Notably, after integrating it into its visual search platform 'Pailitao,' the time required to process one billion product images per day was reduced from one hour to five minutes.
Baidu is expanding the application scope of its AI chip Kunlun series across various fields such as cloud computing, autonomous driving, and smart city construction. In particular, after mass-producing the first-generation Kunlun chip based on a 14nm process in collaboration with Samsung Electronics, Kunlun 2 significantly improved computational performance, enhancing AI training and inference capabilities.
Founded in 2020, Chinese AI GPU startup Moore Threads is developing AI-dedicated GPUs and is known as the 'Chinese Nvidia.' The company is pursuing technological differentiation by developing GPUs compatible with Nvidia's software development tool CUDA and is expanding its market presence in China as an alternative to Nvidia's low-spec GPU products. Another Chinese AI semiconductor startup, Biren Technology, has also introduced the Biren BR100 GPU, competing with Nvidia's A100 AI chip.
Strengthening DRAM and Foundry Technologies... Expanding Influence in the Global Market
China's leading DRAM manufacturer CXMT (Changxin Memory Technologies) is also showing rapid growth. Its DRAM wafer production capacity, which was 40,000 wafers per month in 2020, has expanded fourfold to 160,000 wafers per month in 2024, accounting for about 10% of global DRAM production. This year, it plans to further increase production capacity to 300,000 wafers per month and is actively advancing DDR5 DRAM mass production and HBM (high bandwidth memory) development.
SMIC is China's largest foundry (semiconductor contract manufacturing) company and plays a crucial role in AI semiconductor production. It handles AI chip production for major companies such as Huawei and holds strengths in processes above 28nm. Recently, it succeeded in developing 7nm process technology, expanding its position in the global semiconductor competition. SMIC's clients include Huawei, Qualcomm, Broadcom, and Texas Instruments, with about 74% of its total revenue in 2023 coming from the Chinese domestic market.
A representative from the domestic semiconductor industry said, "The technological capabilities of Chinese semiconductor companies like CXMT and SMIC have reached a level that cannot be ignored. In the past, Chinese companies circumvented Southeast Asia to secure Korean semiconductors, but now they have established a system to directly secure semiconductors through their own domestic semiconductor supply chain."
Thus, China's AI semiconductor industry is rapidly growing based on an independent supply chain and price competitiveness, and competition in the global semiconductor market is expected to intensify further.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
