China is giving investors a weird but very clear setup right now. The economy looks soft, shoppers are not spending with much force, and April retail sales grew at the slowest pace since the post-COVID reopening.
Yet the stock trade is not really about malls, restaurants, or hospital names. It is about AI, semiconductors, hard tech, software, cloud capacity, and the companies sitting close to Beijing’s self-sufficiency push.
Investors keep buying China’s AI supply chain while the wider economy stays uneven
Modern alpha manager WisdomTree’s Liqian Ren thinks the technology growth story will continue as well. While she said that many companies in the AI ecosystem continue earning good profits, she clearly warned that such companies do not have sufficient scale to turn the entire economy around. “It’s very, very uneven,” she stressed.
While many hardware manufacturers trade A-shares on the exchanges of mainland China, not Hong Kong, this is important considering that mainland stocks have outperformed this year.
The Chinese CSI 300 index, which includes large firms trading in Shanghai and Shenzhen, has gained almost 5% this year while the Hang Seng index in Hong Kong is nearly unchanged.
Large private firms are not accessible to stock investors. Private firms like ByteDance and Huawei are not publicly listed. However, many Chinese chip producers, artificial intelligence model developers, and high-tech components makers have gone public recently.
Leonid Mironov’s fund holds Tencent Holdings (0700.HK, TCEHY) and Alibaba Group (BABA, 9988.HK) as its largest positions. He also owns hardware names such as Anji Microelectronics (688019.SS) in Shanghai.
Leonid said investors still miss how much policy support has helped smaller and mid-sized firms make money. “I think people don’t really see and appreciate how fundamentally beneficial the policy has been to the bottom line of these smaller and mid-cap names,” he said.
He is not buying every AI model story, though. Leonid said he is still waiting on Zhipu and MiniMax because he wants clearer proof that customers will stay and that the business model can hold up. Morgan Stanley (MS) is taking the other side. The bank is overweight on Zhipu, MiniMax, and Alibaba. It also has an overweight rating on Cambricon Technologies (688256.SS) with a 2,000 yuan price target, or about $294.
DeepSeek cuts V4 Pro pricing and puts China’s AI cost trade against OpenAI and Anthropic
Finally, an aspect that plays a significant role in telling the China AI story is pricing. DeepSeek, a Hangzhou startup, has maintained the 75% discount on its V4 Pro for one month after launching its V4 series. The V4 series comprises both V4 Pro and lightweight V4 Flash.
By doing so, DeepSeek places itself in the middle of global competition over cost. According to Artificial Analysis, which is a third-party benchmark firm, V4 Pro ranks top globally considering intelligence per dollar cost.
In other words, this ranking depends not only on intelligence but also on the amount of output that buyers receive from the model. The latter factor is especially relevant since powerful computation is constrained, while running large AI models is costly.
The official API price of DeepSeek’s V4 Pro model ranges from as low as $0.0036 per 1 million cached input tokens and $0.87 per 1 million output tokens.
According to Artificial Analysis, the cost of running the Intelligence Index benchmark on this model amounts to about $268. Meanwhile, the cost to do the same thing on OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.7 models would be 12 and 19 times higher, respectively.
This is relevant for all software developers, exchange houses, trading houses, and AI tool developers. The output cost may be a small issue considering the cost that will be added up from tokens. Third-party tests are significant because not all AI businesses use the same pricing or scores for their AI models.
DeepSeek is not the only Chinese name on the list for costs per unit of intelligence. The M2.7 model of MiniMax and the MiMo V2.5 Pro of Xiaomi make the list.