China's AI Rivals Chart Divergent Paths in Hong Kong IPO Race

Two of China's leading artificial intelligence startups have debuted on the Hong Kong Stock Exchange within a day of each other, embodying fundamentally different approaches to building AI businesses in China. Zhipu AI opened at HK$120 per share on January 8, achieving a market capitalization of HK$52.8 billion, while MiniMax is set to list on January 9 with an offering price range of HK$151 to HK$165 per share.

The back-to-back listings underscore a critical bifurcation in China's AI startup ecosystem. Zhipu AI raised approximately HK$4.35 billion ($558 million) through its IPO, with its public offering receiving 1,164 times oversubscription. MiniMax's public offering attracted even stronger demand, garnering 1,209 times oversubscription for a planned fundraising of HK$3.83 billion to HK$4.19 billion ($491 million to $537 million).

The contrasting trajectories of these companies reveal distinct strategies for navigating China's complex AI landscape. Zhipu AI has anchored itself in state-backed enterprise clients and government contracts, leveraging academic credentials and domestic technology. MiniMax has pursued global consumer markets through entertainment applications, backed by international capital and optimized for viral growth.

Academic Roots Versus Commercial Drive

The divergent paths began with the founders' origins. Zhipu AI emerged from Tsinghua University's Knowledge Engineering Group laboratory in 2019, led by Zhang Peng and Tang Jie. The company evolved from AMiner, an academic platform launched in 2006. Zhongke Chuangxing, an investment firm affiliated with the Chinese Academy of Sciences, became Zhipu AI's first external investor, establishing a foundation rooted in China's premier research institutions.

MiniMax founder Yan Junjie followed a different trajectory. After completing his doctorate at the Chinese Academy of Sciences' Institute of Automation in 2015, he joined SenseTime, where he rose to vice president within three years, managing a 700-person core research team. Yan departed in late 2021 as SenseTime prepared for its IPO, convinced that task-specific AI had reached commercial limits and that artificial general intelligence represented the future. He founded MiniMax in Shanghai in January 2022 with backing from Yunqi Capital, a dollar-denominated venture firm.

These founding circumstances imprinted distinct organizational DNA. Zhipu AI inherited academic rigor and government relationships through its Tsinghua lineage and state-affiliated early investors. MiniMax absorbed product development velocity and international market orientation from its founder's corporate experience and dollar-capital backing.

Capital Structures Reflect Strategic Priorities

Funding sources further amplified the companies' divergence. Between 2019 and 2025, Zhipu AI completed eight financing rounds, raising a cumulative 8.3 billion yuan ($1.15 billion) and reaching a pre-IPO valuation of 24.4 billion yuan ($3.4 billion). Beyond prominent venture firms including Sequoia China, Hillhouse Capital, and Qiming Venture Partners, Zhipu AI attracted two critical investor categories.

Industrial capital from Meituan, Alibaba, Ant, Tencent, TAL Education, and Kanzhun provided strategic application scenarios. Simultaneously, state-backed funds including Beijing Artificial Intelligence Industry Investment Fund, Zhuhai Huafa Group, Hangzhou Urban Construction Investment, Zhongguancun Science Park, and China Information & Communication Technology Group formed what insiders describe as a provincial state capital "directory."

Zhipu AI's IPO attracted cornerstone investors representing 68.6% of the offering, including Tsinghua University Education Foundation, Beijing Financial Holding Group, Taikang Life Insurance, and China Fortune-Land Development.

MiniMax pursued a different capital strategy across seven funding rounds totaling $1.56 billion (approximately 11.06 billion yuan), reaching a post-money valuation of $4.24 billion (approximately 30.2 billion yuan) in its August 2025 Pre-B++ round. Industrial investors including Alibaba, miHoYo, Tencent, Xiaohongshu, and Kingsoft Office collectively hold over 25% of the company, significantly higher concentration than Zhipu AI's industrial backers.

MiniMax's cornerstone investors demonstrate international diversification, featuring Abu Dhabi Investment Authority, Prudential's Eastspring Investments, and Korea's Mirae Asset alongside China-based funds including China Asset Management and E Fund Management.

Technology Architecture and Business Models

Capital structures shaped technical and commercial choices. Zhipu AI built its core offering around dense large language models, progressing from GLM-130B to the GLM-4 series using "full-parameter activation" architecture. The company engineered its models for deployment on domestic supercomputing infrastructure including Sunway systems and Huawei's Ascend chips, achieving compatibility with Chinese heterogeneous computing frameworks through mixed-precision operators.

Zhipu AI's competitive advantage centers on "trustworthy large model delivery packages" that bundle model compression, instruction fine-tuning, and algorithm encapsulation for localized deployment. According to its prospectus, localized deployment generated 84.8% of total revenue in the first half of 2025, with Chinese clients accounting for 88.4% of localized deployment revenue. As of December 2025, Zhipu AI's models supported over 12,000 institutional clients and more than 80 million devices, including government agencies, state-owned enterprises, Samsung, and Xpeng Motors.

Government and large state-owned enterprises consistently contributed approximately 60% of Zhipu AI's revenue between 2023 and 2025, with the top five clients accounting for over 40% of total revenue. Media reports indicate that each enterprise or government client requires six to nine months of on-site deployment, creating substantial delivery costs but enabling gross margins of 59.1% for localized deployment services in the first half of 2025.

MiniMax adopted Mixture of Experts architecture from inception, with its M2 model maintaining 230 billion total parameters while activating only 10 billion per inference. Dynamic routing and model quantization techniques reduced single-inference costs to 8% of Claude 4.5's cost, according to company materials. This lightweight design enables efficient operation on mainstream GPU clusters, with inference costs declining 60% within the first year of product launch.

MiniMax generates 71% of revenue from AI-native consumer applications including its namesake agent platform MiniMax, the Hailuo AI visual generation platform, and virtual companion platforms Talkie and Xingye. Talkie employs gacha mechanics adapted from mobile gaming, packaging AI capabilities as entertainment products. Within 18 months of launch, Talkie reached top-ten positions in overall app store rankings across six countries including the United States, Japan, and Brazil.

The application achieved 75 minutes of average daily user engagement, exceeding TikTok's 55 minutes, with average revenue per user of $3.2. Between January and September 2025, MiniMax's native products served 212 million users, approximately 27.62 million monthly active users, and roughly 1.77 million paying subscribers.

Profitability Challenges and Risk Profiles

Business model differences manifest in contrasting margin profiles. While Zhipu AI's localized deployment achieves nearly 60% gross margins, MiniMax's AI-native products generated only 4.7% gross margins in the first half of 2025. Sales and marketing costs consumed 124.7% of revenue in 2023, 87.8% in 2024, and 76.7% in the first nine months of 2025, reflecting substantial user acquisition expenses despite improving efficiency.

MiniMax faces two principal risk factors. Geopolitical uncertainties create regulatory exposure, with products vulnerable to sudden compliance requirements or removal from app stores. User acquisition costs, while declining as a percentage of revenue, remain elevated and constrain near-term profitability.

Zhipu AI confronts different constraints. Heavy delivery requirements for each government and enterprise client create operational scalability challenges. The company's dependence on policy-driven procurement and concentration among top clients exposes revenue to shifts in government IT spending priorities. Its domestic technology stack, while addressing compliance requirements, may limit access to cutting-edge international AI infrastructure advances.

The contrasting profiles represent rational adaptations to China's AI market structure. Zhipu AI converted academic credentials and regulatory compliance into government contract certainty. MiniMax traded architectural efficiency for consumer growth velocity and international expansion capability. Both companies now face post-IPO scrutiny of their ability to achieve sustainable profitability while maintaining technological competitiveness in a rapidly evolving sector.