Contributor | Qianwen, Doubao

Recently, China's AI large-model sector has witnessed a much-anticipated clash: Zhipu AI and MiniMax both cleared Hong Kong Stock Exchange hearings within a mere two days, setting the stage for their market debuts.

These are not isolated IPO events but rather a collective response after three years of China's AI industry navigating through a "hundred models blooming, billions burned" phase.

From the fierce competition in the inaugural year of large models in 2023 to the commercial differentiation and breakthroughs in 2024, these two companies, now approaching the capital market, serve as mirrors reflecting the industry's evolution. Both are backed by top-tier capital but have pursued vastly different survival strategies.

Interestingly, amidst this billion-dollar funding frenzy, banks—traditional financiers in the hard tech sector—have neither taken early equity stakes nor provided initial financial support. Instead, they have adopted a "differentiated entry" approach, becoming core clients and liquidity providers.

 01 

Shareholder Landscape: State Capital Powerhouses vs. Industry Titans

While both companies have attracted China's top capital forces, their shareholder compositions and strategic intentions reveal stark differences.

Zhipu AI is backed by a formidable "national team." Its core leadership boasts Tsinghua University pedigrees. CEO Zhang Peng, a Tsinghua doctorate, began his career at Tsinghua's Knowledge Engineering Laboratory and has led the development of the AMiner tech intelligence platform since 2006—the technological precursor to Zhipu.

Chief Scientist Tang Jie is a professor and deputy head of Tsinghua's Computer Science Department; co-founder, executive director, and chairman Liu Debing previously served as a senior engineer at Tsinghua.

Prior to its public listing, Zhipu secured state capital funds from Beijing, Shanghai, Hangzhou, Chengdu, and Zhuhai. This "five-city blessing" is rare among hard tech companies, underscoring the national-level emphasis on general-purpose large model infrastructure.

Simultaneously, tech giants like Alibaba, Tencent, Meituan, and Xiaomi joined the fray. These investments were not merely for financial returns but to secure a "seat at the table" in the foundational model layer, preventing technological monopolies.

In contrast, MiniMax's shareholder list exudes a more "grassroots" and "industry-centric" vibe. While Shanghai state capital (e.g., Guofang Mother Fund) and Hangzhou Capital participated, their roles were primarily financial.

MiniMax's founder, Yan Junjie, previously served as SenseTime's Vice President, Deputy Dean of the Research Institute, and CTO of the Smart City Business Group, leading the construction of deep learning toolchains and general AI technology systems.

Beyond top-tier venture capitals like Sequoia and Hillhouse, MiHoYo's investment (holding approximately 6.4%) stands out. As a heavy user of gaming and content generation, MiHoYo shares natural business synergies with MiniMax in voice and video generation technologies.

Alibaba and Tencent are also shareholders, but more noteworthy are platform companies like Xiaohongshu and Kingsoft Office, which boast massive C-end user bases. They value MiniMax's rapidly monetizable multimodal capabilities.

 02 

Capital Allocation: Heavy Investment vs. Extreme Frugality

Both companies have secured substantial funding.

Zhipu AI has raised over RMB 8.3 billion in cumulative financing, with over RMB 5 billion secured in just six months since December 2024, valuing it at approximately RMB 24.3 billion.

MiniMax has raised over USD 1.55 billion (approximately RMB 12.4 billion) in cumulative financing, with a latest valuation of USD 4 billion (approximately RMB 32 billion), demonstrating relatively higher financing efficiency and valuation.

However, their spending approaches differ significantly, shaped by their core business models and revenue structures.

Zhipu AI positions itself as a foundational large model provider. For government (ToG) and enterprise (ToB) clients, localized deployment of large models is its primary revenue source. As of 2024, localized deployment contributed 84.5% of revenue, with cloud revenue accounting for only 15.5%.

This necessitates substantial GPU purchases for training, followed by packaging this "software-hardware integration" solution for clients or reserving dedicated computing resources in its data centers. Over 70% of its R&D expenses in 2024 and the first half of 2025 went towards computing service fees.

In contrast, MiniMax focuses on multimodal models and applications, generating revenue mainly from C-end subscription fees (e.g., Talkie, Hailuo AI) and developer API calls. This model dictates a more "cost-effective" capital allocation.

Prospectus data reveals that in the first nine months of 2025, revenue from Talkie/Xingye reached USD 18.75 million, accounting for 35.1%; revenue from Hailuo AI surged to RMB 17.464 million, accounting for 32.6%; revenue from open platforms and other AI-based enterprise services reached USD 15.42 million, accounting for 28.9%.

MiniMax prioritizes optimizing model architectures (e.g., Mixture of Experts systems) to achieve optimal results with minimal computing power. While burning cash, its open platform maintains a gross margin above 60% and converges losses more rapidly.

MiniMax is a true frugality champion. From inception to September 2025, it spent USD 500 million, achieving global leadership in full modality with less than 1% of OpenAI's budget.

 03 

The Fintech Paradox: Where Are the Banks?

Amidst the billion-dollar funding extravaganza for Zhipu AI and MiniMax, a profound fintech question arises: What role do banks, key financiers in the hard tech sector, play?

Over the past decade, Chinese banks have deeply engaged with leading firms in semiconductor, new energy, and high-end manufacturing sectors through equity investments, specialized loans, and supply chain finance.

However, this model has not materialized in the general-purpose large model sector. Neither Zhipu AI nor MiniMax received bank equity investments or even conventional credit support before 2024.

This "absence" is no accident. On one hand, the general-purpose large model industry is characterized by rapid technological iteration, long profitability cycles, and core assets (model weights, algorithm patents) difficult to quantify as collateral. On the other hand, early-stage financing heavily relied on VC/PE risk capital, while state capital and industrial giants further crowded out bank equity participation.

The National Social Security Fund (Zhongguancun Independent Innovation Investment Fund) appearing as a strategic investor in Zhipu's shareholder list reflects policy-oriented capital's priority in large model infrastructure, while risk-averse bank capital opted for cautious observation.

Nevertheless, banks did not entirely exit but shifted to a "differentiated entry" approach. In 2024, Zhipu AI secured a bank loan repaid in July the following year, followed by 10 new unsecured loans totaling RMB 270 million within three months. As of June 2025, the company held approximately RMB 2.55 billion in cash reserves, with bank credit serving as a crucial short-term liquidity source.

Notably, while banks did not invest as shareholders, they became premium clients, forming another closed loop in the fintech ecosystem. Leading financial institutions like China Merchants Bank and Postal Savings Bank precisely targeted the value of large models in financial scenarios, procuring Zhipu AI's models for private deployment in intelligent customer service upgrades, financial code generation, and risk control model optimization.

For Zhipu, a single national bank's cooperation order can generate millions in revenue, with financial institutions becoming the "anchor" of its ToB business.

In contrast, MiniMax's client base leans more towards internet companies and global developers. While its intelligent analysis platform has been adopted by several banks for risk assessment and customer profiling, revenue contributions remain negligible.

Banks' "observe first, then engage" strategy in the large model sector reflects the challenges emerging tech sectors pose to traditional financial models. As the large model industry transitions from R&D burning to commercial validation, bank capital may deeply participate in value distribution through more flexible risk control models and innovative financial products. The listings of Zhipu and MiniMax could mark a new starting point for comprehensive synergy between fintech and the large model industry.

 04 

Conclusion: Two Paths, One Future

Zhipu AI and MiniMax exemplify two distinct survival strategies in China's AI large model industry.

At the foundational layer, relentless cost investment is necessary, relying on state capital and giants to develop core computing power and general capabilities, trading heavy assets for competitive moats.

At the application layer, "market-driven" efficiency is crucial, achieving survival through extreme technological optimization and a globalized vision, trading high margins for viability.

Currently, Zhipu dominates in technological barriers and domestic government-enterprise markets; MiniMax shows greater imagination in commercialization efficiency and global C-end monetization.

Regardless of who lists first, both signify China's AI large models transitioning from "storytelling" to "financial reporting" era.

-End-