Read more: Global capital and Chinese AI converge on Hong Kong

After a year of blockbuster listings, Hong Kong investors were treated to a rare treat in early January as an artificial intelligence (AI) model developer and a chipmaker entered the public market almost simultaneously. Hong Kong is emerging as a global testing ground for how public markets will price the next stage of China’s AI ambitions, as cash-hungry Chinese AI companies rush to take advantage of the booming Hong Kong stock exchange.

In the first two weeks of this year, graphics processing unit (GPU) manufacturers Shanghai Biren Technology and Shanghai Iluvatar CoreX Semiconductor, together with large model developers Zhipu AI and MiniMax, raised a total of HK$17.7 billion (S$2.9 billion). The wide range of products on offer, from basic computing hardware to cutting-edge AI models, highlighted investor enthusiasm across the AI ​​value chain.

International institutions and retail investors flocked to the four IPOs, extending last year’s Hong Kong listing frenzy. But the demand is raising further questions about whether the AI ​​boom is driving up valuations for companies that remain loss-making and, in some cases, lack a clear path to sustainable profits.

Still, market participants agree that more AI companies will enter the market. Mainland Chinese companies, from big-model developers to chipmakers competing to fill the void left by U.S. export controls, are lining up to head south, drawn by flash markets and stock exchanges that have moved to smooth the path to listing.

first public market

Zhipu AI and MiniMax have attracted global attention as the first large-model startups to go public, offering investors an unusually transparent window into a sector that remains largely private elsewhere. Major U.S. model developers such as OpenAI, Anthropic, and xAI remain privately held companies, disclosing only limited operational metrics.

The two Chinese companies have also taken contrasting approaches to commercializing their AI models, and each is currently under public market scrutiny.

Zhipu AI primarily relies on a business-to-business model, offering model-as-a-service products that bundle computing power, development tools, and industry-specific applications. The company posted revenue of 191 million yuan (S$35.1 million) in the first half of 2025, but had a net loss of 2.4 billion yuan as spending on computing resources and its 657-person research and development (R&D) team soared, according to its prospectus.

In contrast, MiniMax relies heavily on consumers. Over 70% of our revenue comes from subscriptions, in-app purchases, and advertising. As of September 30, 2025, the company’s AI-native products attracted 1.8 million paid users and average monthly active users of 27.6 million.

The company’s flagship AI companion apps, Talkie and Xingye, have a combined monthly active user base of more than 20 million. Launched internationally in 2023, Talkie quickly gained attention for its celebrity-style chat interactions and briefly ranked among the top five entertainment apps by downloads in the United States.

Despite this traction, MiniMax remains unprofitable. For the first three quarters of 2025, the company reported revenue of US$53.4 million and net loss of US$512 million, mainly due to research and development expenses of $180 million.

hardware rally

The surge in interest in AI software stocks has been fueled by a parallel rise in domestic chipmakers that stand to benefit from U.S. export restrictions on AI hardware, particularly Nvidia’s advanced processors.

When Nvidia’s chips were effectively cut off from China, domestic players such as Cambricon Technologies sparked a re-evaluation of the sector. The chipmaker secured ByteDance as a major customer, achieved revenue growth of more than 40 times in 2025, stock price rose more than 400 percent, market capitalization exceeded 600 billion yuan, and earned the nickname “King Cambricon”.

This rally changed investors’ perception of domestic GPU manufacturers. In December, Moore Thread Technology and MetaX Integrated Circuits (Shanghai) were both listed on Shanghai’s STAR market, raising a combined 11.5 billion yuan. Moore Threads quickly deployed new AI chips and large computing clusters, increasing competition and accelerating the funding plans of its peers. The company now expects its revenue to more than triple in 2025 due to surging demand for AI computing power.

Chinese internet companies have long relied on Nvidia chips for AI training, but are starting to embrace domestic alternatives. Cambricon and Baidu’s Kunlunxin (Beijing) Technology have emerged at the forefront of this change. ByteDance has introduced Cambricon chips for the ad recommendation system and inference workloads in its Doubao large model, one of the people close to ByteDance told Caixin.

Domestic chips were primarily a backup option in 2024, the person said. By 2025, demand had skyrocketed. The domestic chip still lags behind Nvidia’s H200 in terms of cost performance, but is already viable for certain use cases amid what sources describe as a “huge gap in domestic computing demand.”

why hong kong

For mainland businesses, Hong Kong’s appeal lies in its speed and predictability. The entire process for the four AI companies, from listing hearing to IPO, took about three weeks. One executive said the hearing happened so quickly that there wasn’t enough time to complete all the planned IPO marketing.

Zhipu AI, MiniMax and Biren Technology all relied on Chapter 18C of the Hong Kong Exchange Listing Regulations, which allows listing of “specialized technology” companies that are not yet profitable. Zhipu AI announced in December that it had switched its listing from the mainland to Hong Kong.

The number of 18C listings is expected to increase. According to PwC Hong Kong, as of early 2026, 17 of the approximately 350 public IPO applications fall under Chapter 18C, and more similar applications are expected among the more than 50 confidential applications. A dedicated hotline starting in 2025 will enable early regulatory guidance and confidential submissions. This is a feature that technology companies especially value.

In contrast, mainland IPOs remain tightly controlled, even in politically favorable sectors. The surge in the price of domestic GPU stocks after their listing has also heightened regulatory vigilance, making Hong Kong more attractive as a more stable alternative, said a fund manager at an asset management company in the city.

Its appeal extends beyond startups. Baidu has confidentially filed for an IPO of its AI chip subsidiary Kunlunxin in Hong Kong, underscoring the exchange’s growing role in funding China’s AI hardware drive.

IPO fever and its limits

Hong Kong’s IPO market has regained its vigor. The ‘big four’ accountancy firms predict that this could reach up to HK$350 billion this year, up from HK$285 billion in 2025.

The blockbuster debuts of Biren Technology and MiniMax both attracted more than 400,000 retail investors, and the international tranche of MiniMax was more than 36 times oversubscribed.

Through Hong Kong’s IPO market, domestic and foreign capital is participating in China’s AI investment boom. Li Zhenguo, vice chairman of UBS Global Investment Banking, pointed out that a significant amount of foreign capital has already flowed into the Greater China market in 2025, and there is room for further allocation.

But investment managers at Fosun Asset Management said scarcity could complicate valuation. Most of the big model’s assets remain embedded in tech giants such as Alibaba Group Holding Ltd. and Tencent Holdings Ltd., leaving few purely accessible options. This rarity partly explains the enthusiasm for Zhipu AI and MiniMax.

But it also increases the risk of mispricing. On the eve of Zhipu AI’s listing, one institutional investor said the IPO’s valuation had come under increased scrutiny as the stock only rose about 6% on the gray market. Using OpenAI and Anthropic for comparison, Citic Securities forecasts MiniMax’s 2026 price-to-sales ratio of 30x, implying a market capitalization of HK$53 billion, less than half its current market capitalization.

“There is still no effective evaluation framework for large model enterprises,” said Chu Yanming, chairman of KGI Investment Advisory, warning that the rapid iteration of technology could quickly make pure large model enterprises obsolete. Xu Yan, global partner at Tiger Brokers, added that high valuations, low survival rates and cross-border regulatory risks increase the likelihood of a future correction.

Wang Fengyu, founder of Oakwise Capital Management, said China’s dominance in AI applications does not guarantee a clear winner. He said capital must be patient. “When investing in Chinese AI companies, be wary of companies that burn endless cash, lack sustainable revenue, and lack core technology.” Caixin Global