Frost & Sullivan: Chinese large model companies face an IPO wave, with commercialization being the real dividing line

MiniMax and Zhipu AI's success in passing the Hong Kong Stock Exchange listing review is not only a milestone for both companies but also an important sign that the Chinese AI large model industry is maturing. This wave of listings sends several key signals to the market: First, the industry has entered a new cycle of commercial validation and large-scale implementation from the early stage of technology exploration; second, capital markets have begun to provide institutional tolerance to AI enterprises with high R&D investment, which are not yet profitable but possess core technical barriers, making the Hong Kong stock market a crucial bridge connecting China's artificial intelligence industry with global capital; finally, domestic large models have entered a capital empowerment phase of global competition. After going public and raising funds, they will have more sufficient capital to support global layouts, which also indicates that Chinese large model companies are transitioning from local technological competition to a new stage of competing on the same stage as global AI giants.

The market size of Chinese large models exceeded 20 billion yuan in 2024 and is expected to maintain an annual compound growth rate of over 40% for the next three years. The main drivers of growth come from three aspects: first, the urgent need for enterprises to transform digitally and reduce costs while increasing efficiency; second, continuous policy support for AI infrastructure construction; third, technological progress has driven the continuous expansion of application scenarios, deepening from general dialogue to specialized fields such as finance, healthcare, and energy.

The current market structure shows clear differentiation: on one hand, internet giants (such as Baidu, Alibaba Tongyi, Tencent, ByteDance, etc.) dominate general scenarios with their data, computing power, and ecosystem advantages; on the other hand, professional large model startups like MiniMax and Zhipu AI focus on technical breakthroughs and vertical field deep cultivation.

Among them, Alibaba Tongyi leads far ahead in the open-source model field, not only continuously launching high-quality open-source large model series (such as the Qwen series) but also actively building a developer ecosystem to significantly lower the threshold for enterprises. Its open-source model usage volume is globally leading. In addition, ByteDance's DouPao large model also performs excellently.

Both MiniMax and Zhipu AI are core members of the 'Six Tigers' of large model industry. MiniMax's technical strength is internationally recognized, with its open-source model MiniMax-M2 performing excellently in global authoritative evaluations, while Zhipu AI stands out in developer ecosystem construction.

However, at the foundational model level, their substantial differences from large model companies are actually limited. Current mainstream large models generally train on public datasets, and their architectures mostly adopt standard Transformers or their improved versions. Coupled with the fact that many models are open-source, the technical path is highly convergent, making 'building foundational models' itself difficult to form a real barrier.

As a capital and technology-intensive track, large models require continuous massive funding for underlying architecture research and development, computing power investment, and data accumulation. Global companies like OpenAI and Anthropic have also experienced long-term investment phases. However, if they rely on external fundraising for 'blood transfusion' and lack a clear revenue loop, this burning money model is essentially unsustainable and cannot be considered healthy. Listing fundraising can only alleviate financial pressure in the short term, supporting technology iteration and market expansion; it is an accelerator, not a cure. If they cannot verify large-scale monetization capabilities within 1-2 years after going public, relying solely on capital to sustain will be difficult to support long-term competition. The truly healthy path is to make revenue growth outpace the rate of burning money.

The key path to profitability can be summarized into three points: first, strengthen the revenue loop of APIs and customized services; second, deeply cultivate high-adhesion niche scenarios; third, optimize cost control by innovating architecture and adapting domestic chips to reduce dependence on high-priced imported computing power and reduce cost losses in the inference phase.

Listing fundraising plays an indispensable supporting role in this profit path. On one hand, in terms of research and development, funds raised through listing can support their continuous model iteration and expansion of the open-source ecosystem to cope with the competition for computing power and technology from large companies; on the other hand, in terms of commercialization, funds can help expand overseas markets, build channels, and improve customer service systems.

Looking ahead, there are four decisive trends:

1. Parallel development of open-source and closed-source: Enterprises build developer ecosystems and enhance their technical influence through open-source foundational models while retaining closed-source versions of core capabilities for commercial monetization.

2. Accelerated implementation of multimodal and intelligent agent integration, with models shifting from single text and image interactions to cross-modal, manipulable intelligent agent forms, deeply penetrating into physical industries such as manufacturing and intelligent driving.

3. Accelerated process of domestic substitution, with more mature adaptation of domestic chips, frameworks, and models, and the localization rate of computing infrastructure is expected to rise above 40%, reducing industry dependence on overseas hardware.

4. Compliance becomes a basic threshold, with compliance requirements such as data security and model alignment forcing enterprises to improve their technology and management systems.

The biggest risks they may face include industry internal strife caused by homogenization competition and supply chain risks related to core technologies and computing power. Currently, high-end computing chips still rely on imports, and if overseas restrictions intensify, it may affect the model iteration progress of some enterprises.