By Wang Chunhao

Source: Jiedian Finance

Looking back at the end of 2025, China's artificial intelligence sector is undergoing a paradigm shift.

If 2023 was the wild outbreak of the 'Hundred-Model War' and 2024 the differentiation of technological routes, then 2025 has undoubtedly entered the most brutal and critical 'capital realization phase.'

As the hype around generative AI technologies fades and the urgent need for commercialization lands, the once-noisy market quickly cools and solidifies. Three 'unicorns'—Zhipu AI, MiniMax, and Yuezhi'anmian—have broken through, forming the first tier in China's large model track and collectively making a final sprint toward the Hong Kong Stock Exchange.

Jiedian Finance believes this is not just a race among three companies but the ultimate showdown between two business models, two technological beliefs, and the capital wills of two internet giants.

MiniMax is becoming the strongest dark horse for IPO with its strong monetization capabilities in overseas markets and a unique product matrix, aiming for January 2026. Zhipu AI, representing the 'Tsinghua faction' national team, abandoned the Science and Technology Innovation Board but swiftly turned to Hong Kong stocks, attempting to establish valuation anchors as infrastructure. Meanwhile, Yuezhi'anmian, despite capturing user mindshare in long-text with Kimi, struggles to balance computing costs and commercialization, delaying its IPO timeline.

Meanwhile, the financing window in the primary market is closing. Investors are no longer paying for mere 'parameter scale' but demanding clear 'unit economic models.' IPOs are not just financing channels but 'get-out-of-jail-free cards' for these unicorns to survive the next round of reshuffling.

01 The Three-Way Showdown: How Are They Performing?

Among the three contenders, MiniMax is the most unique and likely to cross the finish line first.

Headquartered in Shanghai, the company was founded by Yan Junjie, former vice president of SenseTime. Unlike its academically inclined competitors, MiniMax has had a strong commercial bloodline since its inception.

MiniMax's shareholder list is stellar, including Alibaba, Tencent, and Sequoia China. Sources reveal the company plans to list on the Hong Kong Stock Exchange as early as January 2026, aiming to raise hundreds of millions of dollars. As the largest revenue-generating and clearest business model target, MiniMax is seen by investment banks as the most easily understandable and acceptable 'first stock' for the secondary market.

MiniMax's core asset is not some underlying hundred-billion-parameter model but a consumer-facing app called Talkie. This app avoids the red ocean of ChatGPT-style 'productivity tools' and enters the blue ocean of 'AI companions.'

According to Bloomberg, MiniMax's expected revenue for 2024 is as high as $70 million, with the vast majority coming from Talkie. This figure stands out in China's AI startup circle, which generally lacks self-sustaining capabilities.

More crucially, 70% of MiniMax's revenue comes from overseas markets, especially the U.S. Talkie is extremely popular among American teenagers, consistently ranking high in app store charts. Its user stickiness and willingness to pay far exceed domestic counterparts. By introducing gamified payment mechanisms like 'gacha,' MiniMax successfully converts AI computing power into high-margin virtual goods rather than cheap API calls.

If Talkie solves the 'survival' problem, Hailuo Video embodies MiniMax's developmental ambitions. In the second half of 2024, MiniMax released the video generation model Hailuo AI, directly competing with OpenAI's Sora and Kuaishou's Kling.

Jiedian Finance analyzes that video generation is seen as the next Explosion point ( Explosion point translated as 'breakout point') after text, also the most compute-intensive and technically demanding field. Hailuo AI quickly topped global video generation product charts after release, proving MiniMax not only understands products but also possesses the underlying technological prowess to compete with giants.

If MiniMax is the agile 'ranger,' Zhipu AI is the steady 'heavy armor division.' With its pure lineage from Tsinghua University's Computer Science Knowledge Engineering Lab, it occupies an irreplaceable ecological niche in China's large model landscape.

Zhipu AI's strategic core is to build a 'Wintel'-like alliance ecosystem at the bottom layer ( bottom layer translated as 'foundation') level. By open-sourcing small and medium-parameter models, Zhipu quickly captured developer mindshare, then guided enterprise users to migrate to more powerful closed-source commercial models.

According to Tech in Asia and related reports, Zhipu AI's total revenue for 2024 is expected to reach $42 million, with annual recurring revenue from AI development tools exceeding $14 million. This means Zhipu has Preliminary run through ( Preliminary run through translated as 'preliminarily established') the SaaS model.

Zhipu's greatest moat lies in its B2G and B2B businesses. With data security becoming a national strategy, Zhipu has secured numerous privatization deployment orders from governments and state-owned enterprises with its 'autonomous and controllable' label, giving its revenue extremely high certainty and resilience to cycles.

Zhipu AI was once the most touted AI company for the Science and Technology Innovation Board. However, in mid-2025, the company abruptly shifted strategy to Hong Kong. Jiedian Finance believes this decision reflects management pragmatism—the Science and Technology Innovation Board's requirements for profitability and R&D capitalization are too rigid for Zhipu, which still needs massive investment. Hong Kong's flexibility and its status as an international capital gateway better align with Zhipu's 'rooted in China, radiating Belt and Road' globalization strategy.

Yuezhi'anmian, founded by post-90s prodigy Yang Zhilin, is the youngest and most Silicon Valley-esque of the three. It carries China's tech circle's ultimate fantasy for AGI (Artificial General Intelligence).

Yuezhi'anmian rose to fame with the Kimi smart assistant. Kimi was the first to support ultra-long context inputs of 200,000 to 2 million words, precisely addressing knowledge workers' pain points in processing financial reports, legal documents, and academic papers.

As of Q3 2025, Kimi has about 9 million monthly active users. While trailing ByteDance's Doubao in absolute numbers, it excels in core metrics like user time and retention. Its user base skews highly educated, with significant potential commercial value.

However, long-text processing incurs steep inference costs. Processing 2 million words consumes massive video memory and computing power. With C-end users accustomed to free usage, Kimi faces the dilemma of 'the more it's used, the more it loses.'

Yuezhi'anmian has experimented with a 'tipping' model and launched enterprise-grade APIs. But compared to MiniMax's gamified revenue and Zhipu's B-end contracts, Kimi's monetization remains weak. This directly delays its IPO timeline—planned for the second half of 2026, apparently to allow more time to optimize unit economic models.

02 Alibaba and Tencent's 'Double Bets'

Behind this IPO race stand two colossal figures—Alibaba and Tencent.

A deep dive into the three companies' equity structures reveals that Alibaba and Tencent have invested in all three nearly simultaneously.

Unlike the clearly divided camps of U.S. tech giants (Microsoft with OpenAI, Amazon with Anthropic), China's Alibaba and Tencent have adopted a 'shotgun' strategy of full-track coverage.

Under Joe Tsai and Eddie Wu's leadership, Alibaba has made AI its core strategy. Through investments, Alibaba secures not just financial returns but future computing demand for its cloud business. Alibaba's lead investment in Yuezhi'anmian's latest $1 billion funding round exemplifies this strategy.

Tencent's approach focuses more on application-layer connectivity. As the social media titan, Tencent doesn't want to miss the next WeChat. Investing in MiniMax and Yuezhi'anmian serves as double insurance beyond its own Hunyuan large model.

Jiedian Finance argues that large models could disrupt existing internet gateways, including search, e-commerce, and social media. Unable to determine the final winner, the safest bet is to buy the entire track. This means that regardless of whether MiniMax, Zhipu, or Yuezhi'anmian goes public first, Alibaba and Tencent will almost certainly be the behind-the-scenes winners.

03 The Decisive Battle for Commercialization: The Application Layer Explodes

In May 2025, ByteDance released the Doubao large model, pricing tokens by the 'cent' and instantly igniting the market. Zhipu swiftly lowered GLM-4.5 Lite to $0.11 per million tokens.

Giants aim to clear the field. For independent startups without cloud provider backing, this is an asymmetric war. Giants can sell models at a loss and profit through cloud services; startups lose money with every model sale.

This forces MiniMax and Yuezhi'anmian to move up the stack and create super apps. At the application layer, users pay via ads or subscriptions, insensitive to underlying token costs.

By late 2025, the market landscape becomes clearer.

At the foundation level, Alibaba (Tongyi), Tencent (Hunyuan), Baidu (Wenxin), ByteDance (Doubao), and Zhipu (GLM) divide the general-purpose large model market.

At the application layer, MiniMax (Talkie) and Yuezhi'anmian aim to become the 'TikTok' and 'WeChat' of the AI era. Data shows Talkie's user time already surpasses many traditional social apps, proving AI-native applications have the potential to disrupt old orders.

On the other hand, facing an overly competitive domestic market, going overseas isn't an option but a necessity.

MiniMax's Talkie success proves Chinese teams' dimensional superiority in ToC products. With its stellar financials and clear business model, it will likely ring the bell first in January 2026, becoming China's first large model IPO. This will significantly boost industry confidence and set a high valuation benchmark for subsequent Zhipu and Yuezhi'anmian listings.

However, Jiedian Finance cautions that listing is just the beginning. Under the public market spotlight, every research dollar invested must translate into revenue. For these young entrepreneurs, the real test has just begun. They must not only confront technological iterations but also face the ruthless scrutiny of Wall Street and Central investors.

China's AI golden age, powered by capital, is accelerating.

*Featured image generated by AI