Preface: The next three years will witness transformative shifts in the competition for megamodel dominance.

According to iResearch Consulting, China's NLP market is projected to reach approximately 50 billion yuan by 2023, with an annual growth rate of 30%, surpassing 100 billion yuan by 2025.

In terms of competition, although there is no global market leader and the market is not monopolized by a single enterprise, the intensity has escalated to a national level.

Author | Fang Wensan Image Source | Network

First Domestic Megamodel Company to Initiate IPO

On April 15, the China Securities Regulatory Commission's online service platform revealed that Beijing Zhipu Huazhang Technology Co., Ltd. completed registration with the Beijing Securities Regulatory Bureau on April 14, 2025, with China International Capital Corporation Limited as its advisor.

Zhipu thus becomes the first innovative enterprise to officially commence the listing process in the "Hundred Megamodel War" and is poised to be the sector's first listed company.

Since its establishment in 2019, Zhipu has raised over 15 billion yuan in funding, with a valuation of 20 billion yuan in February 2025.

As a Tsinghua University-originated megamodel unicorn, Zhipu has developed a comprehensive blueprint spanning trillion-parameter base models, dialogue systems, code generation, multimodal processing, and agent technology over six years.

Zhipu has not only made significant breakthroughs in megamodel fundamental research but also spurred original advancements in multiple key areas.

Its GLM architecture and phased models, including GLM-10B, Wudao (a trillion-parameter MoE architecture-based model), and GLM-130B, showcase its independent large-scale pre-trained model research and development capabilities.

Zhipu's API revenue surged over 30 times in 2024, with daily average token consumption increasing by 150 times.

Its Machine-as-a-Service (MaaS) megamodel platform boasts over 700,000 enterprise and developer users.

In the industry, Zhipu collaborates with numerous enterprises in sectors like new energy vehicles, manufacturing, consumer goods, finance, government, healthcare, culture, and entertainment.

For instance, Zhipu supports automakers like Xpeng with AI technology for intelligent cockpits and partners with financial institutions such as the Postal Savings Bank of China and China Merchants Bank for applications like intelligent customer service and risk assessment.

Additionally, Zhipu has achieved notable success in the international market, securing overseas orders worth tens of millions of dollars and signing a cooperation agreement with Samsung, generating over 100 million yuan in revenue in less than a month this year alone.

With the addition of Beijing, Hangzhou, Zhuhai, and Chengdu, Zhipu's "[Local + Technology + Capital]" ecosystem is expanding rapidly.

On March 9, 2025, the Chengdu High-tech Zone announced a 300 million yuan investment in Zhipu to jointly build Sichuan's base megamodel [Zhipu AI Zhuge] and construct a trinity AI infrastructure.

This initiative aims to support over 1,000 enterprises in achieving AI transformation and incubate 20 ecological enterprises within the next five years, marking Zhipu's third round of funding within the month.

On March 3, the Hangzhou City Investment Industry Fund and Shangcheng Capital jointly invested over 1 billion yuan in Zhipu to promote the implementation of the GLM megamodel in the Yangtze River Delta region.

On March 13, Zhuhai's largest state-owned enterprise [Huafa Group] announced a strategic investment of 500 million yuan in Zhipu, jointly establishing the city's first GLM megamodel space with the Zhuhai High-tech Zone - [Zhipu + Zhuhai Huafa Space].

Tsinghua Alumni Team Builds Multiple Megamodel Versions

Founder Zhang Peng holds a Ph.D. in Innovative Leadership Engineering from Tsinghua University's Department of Computer Science and Technology, specializing in knowledge graphs and large-scale pre-trained models, current AI frontiers.

His founding team, primarily composed of Tsinghua alumni, is China's earliest academic team focused on data mining, semantic analysis, machine learning, and algorithms.

Before founding Zhipu, his team developed AMiner, China's first large-scale data mining and service system platform with independent intellectual property rights.

Core team members include CEO Zhang Peng (Tsinghua Computer Science graduate), Chairman Liu Debing (disciple of Chinese Academy of Engineering academician Gao Wen), and President Wang Shaolan (Tsinghua Ph.D. in Innovative Leadership).

Based on the internally iterated GLM pre-training architecture, Zhipu has launched multiple megamodel versions since 2023.

In 2020, Zhipu developed the GLM pre-training architecture; in 2021, it trained the GLM-10B model with 10 billion parameters and a converged trillion-sparse model using the MoE architecture; in 2022, it developed and open-sourced GLM-130B, a Chinese-English bilingual trillion-scale pre-trained model; and in 2023, it launched ChatGLM, a trillion-base dialogue model, upgraded twice, and optimized for different vertical scenarios, serving both ToB and ToC directions.

In 2024, Zhipu remained active: in January, it introduced GLM-4, a new-generation base megamodel with significantly improved performance; in June, it open-sourced GLM-4-9B and the visual model GLM-4V-9B with multimodal capabilities comparable to GPT-4V; in July, it launched CogVideoX, a video generation model with 6x faster reasoning speed; in August, it introduced GLM-4-Plus, a newly upgraded base megamodel with world-class capabilities; in October, it released GLM-4-Voice, an end-to-end emotional speech model, and AutoGLM, an interactive megamodel agent; in November, it upgraded the agent with GLM-PC, initiating technical exploration of [unmanned] PCs; and in December, it released GLM-Zero-Preview, excelling in handling complex mathematical logic, code, and deep reasoning tasks.

Entering 2025, Zhipu launched GLM-Realtime in January, supporting a cappella, 2-minute memory, and Function Call functions; and in March, it introduced AutoGLM Meditation, the first AI agent with deep research and operational capabilities.

Racing Ahead of Megamodel [Five Tigers] to IPO to Ease Competitive Pressure

Last year's aggressive AI megamodel [Six Tigers], committed to Artificial General Intelligence (AGI) research, have now differentiated, pursuing diverse development paths.

Kai-Fu Lee's Wandao announced the launch of the Wanzhi enterprise-level large model one-stop platform in mid-March, providing enterprises with customized DeepSeek deployment solutions, becoming the first [Six Tiger] to fully embrace DeepSeek.

Baichuan Intelligence, committed to building fundamental large models and considered a strong competitor to China's OpenAI version, is currently undergoing strategic adjustments.

MiniMax to B, another [Six Tiger] member, has attracted over 30,000 enterprise customers on its open platform, becoming the most widely used AI large model platform in the domestic public cloud market, though it is also adjusting its strategic direction, particularly for ToC services targeting the consumer market.

In February 2025, MiniMax integrated DeepSeek-R1 into its overseas AI assistant platform MiniMax Chat, offering it for free for a limited time.

While Zhipu leads in the IPO process, the road ahead for large model companies remains long.

As 2025 marks a pivotal year for AI application implementation, large model companies may face industry [reshuffling].

Zhipu must continue to withstand pressure, control costs, and accelerate product R&D and implementation exploration.

According to Smart Hyperparameters' "China Megamodel Bidding Project Monitoring Report (2024)," China's top six general megamodel vendors in 2024 were iFLYTEK, Baidu, Zhipu, Volcano Engine (ByteDance), Alibaba Cloud, and Tencent Cloud, in order.

Zhipu ranks third, after iFLYTEK and Baidu.

Another report indicates that in the AI megamodel solution market, Zhipu ranks fourth with a nearly 8.8% market share, after Baidu (17.0%), SenseTime (14.8%), and iFLYTEK (14.1%).

However, in the MaaS market, Zhipu does not appear among the top five players, with platform giants like Tencent (Yuanbao), Alibaba, and Baidu having established full-stack AI technology independent R&D and popularized models like DeepSeek R1, Doubao, and Kimi dominant in the C-end market.

Conclusion:

From a competitive perspective, Zhipu's pressures will gradually increase, especially with the launch of DeepSeek, which benefits Zhipu by achieving an earlier IPO.

DeepSeek significantly impacts Zhipu's AI business commercialization, directly challenging its core business model.

This impact is first seen in B-end market customer diversion.

By offering the open-source R1 model and an extremely low API pricing strategy (e.g., R1 model calling cost is only 1/30 of similar market products), DeepSeek has eroded and will continue to erode Zhipu's customer base.

Some enterprises previously reliant on Zhipu's MaaS platform may switch to DeepSeek's privatized deployment or directly use its API services.

Since the beginning of 2025, over 200 enterprises nationwide have completed localized deployment, covering government agencies, central enterprises, state-owned enterprises, financial institutions, medical institutions, and technology enterprises.

Partial reference materials: LP Spectrum: "[Zhipu is going crazy]", Geek Park: "The Six Little Dragons of Megamodels, the First IPO is Coming", Caijing Tianxia WEEKLY: "The Six Little Tigers of Megamodels, Zhipu is the First to Reach the IPO Ticket", Zhang Tongshe: "The AI [Unicorn] in Zhangjiang Starts IPO", Direct to IPO: "Just Now, Zhipu, Valued at Over 20 Billion Yuan, Starts IPO", ECNS: "Beijing's Hottest Unicorn is Going Public", Internet News: "Tencent, Xiaomi Participate in Investment! The [Six Little Tigers] of Megamodels, Valued at Nearly 30 Billion Yuan, is Going Public", Pencil Dao Pro: "Breaking News! A 20 Billion Super Unicorn is Born: Lightning IPO, Breaking Records"