Who could gain from DeepSeek’s V4 with China chips poised for stronger demand?

DeepSeek’s launch of its latest artificial intelligence model could likely trigger a broad reassessment of stocks across the industry chain from chipmakers to large language model developers, analysts say, with the breakthrough poised to drive demand for computing power and more commercial adoption. The unveiling of the V4 series marked another milestone for the Hangzhou-based start-up, which described the model as the most powerful open-source platform capable of challenging US rivals such as OpenAI and Anthropic. It delivered top-tier performance in coding benchmarks and showed significant advantages in reasoning and agentic tasks, according to DeepSeek. AI chipmakers such as Cambricon Technologies and Moore Threads Technology stand to benefit, with the new progress spurring demand for more high-performance chips produced in mainland China. The launch may also lower the costs of AI integration into daily life and broaden commercial usage, potentially benefiting AI firms such as MiniMax and Knowledge Atlas Technology, better known as Zhipu. “DeepSeek’s V4 has lowered the threshold for using high-performance AI models and will offer more affordable AI capabilities to small and medium-sized enterprises or even individuals,” said Su Lingyao, an analyst at BOC International. “DeepSeek’s V4 is also highly compatible with domestically made chips, and that will accelerate the commercialisation of AI computing power in China.” Potential beneficiaries include chipmakers from Hygon Information Technology and MetaX Integrated Circuits to fab operators such as Semiconductor Manufacturing International Corp (SMIC) and Hua Hong Semiconductor, according to brokerages. China’s AI chip market could rise to 1.34 trillion yuan (US$196.2 billion) in 2029 from 142.5 billion yuan in 2024, translating into an annual compound growth rate of 54 per cent, according to Guotai Haitong Securities. The Star Market 50 Index of the biggest technology stocks in Shanghai – where Cambricon and SMIC are listed – jumped 25 per cent in April to approach a five-year high. Cambricon rose to a record high on Thursday. DeepSeek’s launch coincided with a renewed rally in US tech stocks, with the Nasdaq 100 index recovering all losses from the oil shock to hit a record high last month. Investors bet that earnings at mega tech firms would be insulated from crude supply disruptions and that continued investments in AI infrastructure by hyperscalers would bolster earnings at Asia’s chipmakers. The V4 was optimised for domestic chips, potentially reinforcing its role within China’s technology stack under sanctions, according to analysts. DeepSeek was reportedly in talks with Tencent Holdings and Alibaba Group Holding for its first external funding round. Alibaba owns the South China Morning Post. Morgan Stanley described Chinese models as efficient, delivering performances similar to US peers at only 15 to 20 per cent of inference costs. The US bank said Chinese AI model companies such as MiniMax were undervalued, given their growth prospects. Major players raised product prices from the second quarter last year to the first quarter this year, driven by performance improvements that indicated a shift away from pure commoditisation to performance-based monetisation, it said in a report in April. These stocks could also reshape Hong Kong’s stock market by drawing more inflows, Morgan Stanley said. MiniMax and Zhipu would very probably join the Hang Seng Tech Index in June with combined weightings between 5 and 7 per cent, which would bring inflows of as much as US$1.75 billion from passively managed funds, it said. The two stocks would lure additional buying from mainland China investors after joining the exchange link programme by August, with purchases equal to as much as a fifth of their free-float market capitalisations within six months of inclusion, Morgan Stanley said. “DeepSeek’s collaboration with domestic chipmakers is of great significance in driving the coordinated development of hardware and software for large language models and speeding up self-controlled computing power domestically,” said Li Kefu, an analyst at Sinolink Securities.