Company Overview
Zhipu AI was founded in 2019, emerging from the Knowledge Engineering Group (KEG) of Tsinghua University's Department of Computer Science. The core team, led by Professor Tang Jie, is one of the earliest in China devoted to pre-trained large model research. The company is positioned as a core player in the "AI model and middleware layer," focusing on the development of general-purpose large models (GLM series) and building an open model ecosystem and application middleware on this foundation. In 2023, Zhipu AI rapidly accumulated developer credibility through its open-source ChatGLM series models, subsequently launching GLM-4, a hundred-billion-parameter foundational model that demonstrated outstanding performance in Chinese comprehension and reasoning. The company is not publicly listed but has completed multiple financing rounds, with early investors including Hillhouse Capital, Qiming Venture Partners, and Meituan. By the end of 2023, its valuation exceeded US$1 billion, placing it in the top tier of the "Six Little Dragons" of large models.
Zhipu AI provides API call services for models such as GLM-4 and GLM-3 through its open platform, offering developers and enterprise customers capabilities including text generation, dialogue, code generation, and multimodal understanding. It adopts a pay-as-you-go (Token) pricing model and monthly subscription packages, and introduces limited-time free quotas to attract small and medium-sized developers. This business is the company's main source of cash flow. In 2024, benefiting from the explosive growth of large model applications, API call volume increased by over 300% year-on-year. Major customers include online education platforms, SaaS companies, and gaming companies.
For high-privacy and high-customization scenarios such as finance, healthcare, and government affairs, Zhipu AI offers private deployment solutions based on GLM models, including model fine-tuning, knowledge base injection, security auditing, and other features. Typical customers include the intelligent customer service system of a major state-owned bank and the medical report generation platform of a top-tier hospital. This business has a high gross margin (estimated above 70%), but the delivery cycle is relatively long.
Zhipu AI actively embraces open source, having successively open-sourced models such as ChatGLM-6B, GLM-130B, and CodeGeeX, forming a large developer community. Based on its open-source models, the company provides enterprise closed-source enhancement services (e.g., higher concurrency support, proprietary data training), as well as value-added services such as technical training and model evaluation. In addition, the company expands its influence through GitHub sponsorships, technology summits, and other channels, indirectly driving its API and private deployment businesses.
This includes joint research projects with universities and national-level laboratories, as well as government-affairs large model pilots for government agencies. The gross margin for this segment fluctuates considerably, but it helps enhance the technical brand reputation.
| Product Line | Revenue Share (Estimated) | Core Customers | Gross Margin (Estimated) |
|---|---|---|---|
| API Cloud Services | ~40% | Small and medium-sized developers, SaaS enterprises | ~60% |
| Private Deployment & Solutions | ~35% | Financial institutions, hospitals, government agencies | ~75% |
| Open Source Ecosystem and Derivatives | ~15% | Developer community, enterprise IT departments | ~50% |
| Others | ~10% | Universities, governments | ~40% |
Technical Moat
Moat 1: Open-Source Ecosystem and Academic Influence of the GLM Series Models
Zhipu AI's GLM (General Language Model) is built on the Transformer architecture and delivers outstanding performance in long-context Chinese text understanding and multi-turn dialogue. Its open-source models, ChatGLM-6B and GLM-130B, have become among the most downloaded Chinese large models on HuggingFace by domestic developers. The open-source strategy not only lowers the trial-and-error cost for new users but also creates a funnel effect of "API calls → open-source adoption → enterprise upgrade." Meanwhile, the academic team consistently publishes papers at top conferences (e.g., ACL, NeurIPS), building a strong foundation of technical credibility.
Moat 2: Vertical Domain Fine-Tuning Capabilities and Model Safety Alignment
Zhipu AI has accumulated extensive fine-tuning data in vertical domains such as finance and healthcare, and has developed model safety alignment technologies (e.g., value alignment, content filtering) to meet domestic regulatory compliance requirements. Compared with general-purpose APIs, its private deployment solutions integrate industry knowledge graphs, enhancing model accuracy in specific scenarios.
The core team of Zhipu AI originates from Tsinghua University's KEG Lab. The company continuously recruits AI doctoral interns from Tsinghua each year and jointly trains postdoctoral researchers with the university. This "research-first, production-second" model enables Zhipu AI to gain first-mover access to cutting-edge research outcomes (e.g., MoE architecture, long-context expansion), reducing R&D costs.
| Dimension | Data |
|---|---|
| Domestic large model market share (2024 estimate) | Approximately 8% |
| Industry ranking | Fifth (top four: ByteDance Doubao, Baidu ERNIE, Alibaba Tongyi Qianwen, Moonshot Kimi) |
| Main competitors | Baidu ERNIE Bot, Alibaba Tongyi Qianwen, Moonshot AI, MiniMax, Baichuan Intelligence |
| Downstream customers | Finance, healthcare, education, government affairs, software development, etc. |
Competitive background: The domestic large model market presents a landscape where "big tech price wars" and "startup differentiation" coexist. ByteDance and Baidu compete for developers through free or ultra-low API pricing, putting pressure on Zhipu AI's API business. However, Zhipu AI has established barriers in the government and enterprise market through its open-source ecosystem and high-average-transaction-value private deployment solutions. In 2024, it launched the GLM-4 long-context (128K token) version, further narrowing the gap with Kimi.
Financials & Growth
The company has not published financial reports; the data below is compiled based on industry estimates and public funding information.
| Metric | Data |
|---|---|
| Revenue (2024 forecast) | Approximately RMB 500-800 million (of which API accounts for ~RMB 300 million, privatization ~RMB 200-300 million) |
| Gross Margin | Approximately 65% (weighted average) |
| Net Margin | Approximately -30% (still in a high R&D investment phase, loss-making) |
| Core Growth Logic | 1) Government and enterprise digital transformation drives demand for privatized deployment; 2) Open-source community users converted into paying API customers; 3) Multimodal models (image/video generation) expanding revenue streams after launch |
Growth drivers: In 2024, the penetration rate of domestic large models in the financial and healthcare sectors was less than 15%, expected to double by 2025; Zhipu AI has signed contracts with multiple leading banks and tertiary hospitals, securing contract value for the next 2 years. Additionally, the company plans to launch a new generation of MoE architecture foundation model in 2025, with anticipated inference cost reduction of 50%, further attracting small and medium-sized developers.
Key Risks:
- Price war pressure from major tech companies: Baidu, ByteDance, and Alibaba continue to lower API prices or even offer them for free, which may erode Zhipu AI's API market share and compress profit margins.
- Commercialization falls short of expectations: Government and enterprise private deployment cycles are long, and customization costs are high. If customer renewal rates decline, cash flow will be impacted.
- Technology iteration pressure: Overseas vendors such as OpenAI and Google are continuously releasing stronger models, while new domestic players (e.g., DeepSeek) are also catching up rapidly. Zhipu AI needs sustained high R&D investment to maintain competitiveness.
- Financing pace and environment: If capital market enthusiasm for large AI models cools down, the company may face valuation downgrades or increased difficulty in future financing rounds.
Core Investment Thesis / Industry Value Summary:
Backed by Tsinghua's academic background, an open-source ecosystem moat, and first-mover advantages in the government and enterprise market, Zhipu AI has established a "technological backbone" position in the domestic large model track. Despite challenges from major tech companies' price wars and the long commercialization cycle, its highly sticky private deployment solutions and sustained technological innovation capability position it to become a key player in the middleware layer of the AI industry. It is recommended to monitor the rollout pace of its multimodal products and the implementation of government and enterprise orders.