In-Depth Analysis of Zhipu AI's Technology Licensing Revenue Structure and Long-Term Competitiveness

Based on the latest collected data and market information, I will conduct a systematic analysis for you.

---

According to the prospectus data disclosed by Zhipu AI in December 2024, the company’s revenue features typical “heavy asset, heavy delivery” characteristics:

Time Period	Proportion of On-Premises Deployment	Proportion of Cloud Deployment (MaaS)
2022	95.5%	4.5%
2023	90.4%	9.6%
2024	84.5%	15.5%
H1 2025	84.8%	15.2%

From the data evolution, although the revenue from on-premises deployment shows a downward trend (a cumulative decrease of about 10.7 percentage points over four years), it still occupies an absolutely dominant position. Zhipu AI CEO Zhang Peng clearly stated that the company plans to increase the revenue proportion of API business to 50%, indicating that the current business structure adjustment is still in progress [1][2].

1. High Gross Profit Margin Support
   : The gross profit margin of on-premises deployment business has remained above 50% for a long time, reaching 66.0% in 2024 and 59.1% in H1 2025, which is significantly higher than the industry average. This high gross profit margin reflects the value premium capability of customized services [1].

2. Deep Binding with Government and Enterprise Customers
   : As one of the few enterprises that can continue to launch new models after being included in the US Entity List, Zhipu AI, relying on its advantages of technological neutrality and secure controllability, deeply meets the rigid demands of government and enterprise customers for data security and customization, which is a niche market that internet giants find difficult to cover [2][3].

3. Industry Ecosystem Construction Capability
   : Through the “Z Plan” and the 1.5 billion RMB “Z Fund”, Zhipu AI has invested in multiple startups including Jiliu Technology and Wuxin Xinqiong, forming an ecological network centered on its own technology and enhancing its discourse power in the industrial chain [1].

1. Customer Concentration Risk
   : In 2024, the top five customers contributed 45.5% of the revenue, with the largest customer accounting for 19%, and there is almost no overlap in the annual top five customer lists, reflecting the unsustainability of fees caused by the one-time delivery model, which requires continuous acquisition of new customers [2][3].

2. Restricted Large-Scale Expansion
   : On-premises deployment means long sales cycles, high customization levels, and heavy delivery costs, making it difficult to achieve exponential growth under the API model. Compared with OpenAI’s revenue structure dominated by consumer subscriptions, Zhipu AI’s expansion elasticity is significantly weaker [2][3].

3. Intensified Price War Pressure
   : At the beginning of 2025, DeepSeek impacted the market with an open-source free strategy, promoting the in-depth development of industry price wars. Zhipu AI’s on-premises heavy delivery model is at a disadvantage in terms of cost structure, facing the risk of market share erosion [2][3].

Currently, China’s large language model industry is experiencing a fierce reshuffle. The industry consensus is that it may eventually converge to the “Five Basic Model Leaders” — DeepSeek, Alibaba, ByteDance, StepStone, and Zhipu AI. Kai-Fu Lee is more pessimistic, predicting that only three large language model companies will remain in China in the end [2][3].

Dimension	Zhipu AI	MiniMax	DeepSeek	Alibaba/ByteDance
Revenue Structure	B-end Dominant (84.8%)	C-end Dominant (71.4%)	Open-Source and Free	Ecosystem Collaboration
Overseas Revenue	11.6%	73.1%	Rapid Growth	In Layout
Technology Route	Full-Stack Independent R&D	Product + Model	Open-Source Strategy	Independent R&D
Valuation	24.4 Billion RMB	4 Billion USD	Unlisted	Hundreds of Billions

Zhipu AI’s technological advantage lies in the continuous iteration capability of its GLM series models. GLM-4.7 has tied for first place with OpenAI and Anthropic in global programming blind tests. However, the link from technological leadership to commercial transformation has not yet been fully established. The prospectus shows that its cumulative R&D investment is about 4.4 billion RMB, which is more than 8 times the revenue in the same period [2][3].

• The high proportion of technology licensing revenue does not pose a fatal threat for the time being, as the company has established a solid customer base of government and enterprises and a network of more than 8,000 institutional customers
• The gross profit margin remains healthy, and the level above 50% provides support for continuous R&D investment
• Overseas business is expanding rapidly (the proportion of overseas revenue reached 11.6% in H1 2025), which helps to diversify single-market risks

• Critical Turning Point
  : If the proportion of MaaS business cannot be effectively increased to the target level, Zhipu AI may fall into the trap of “heavy delivery, low growth” and struggle to win in large-scale competition
• Technological Generation Gap Risk
  : The “half-generation gap” in the large language model industry can be quickly closed, as evidenced by the sudden rise of DeepSeek. Zhipu AI needs to maintain technological leadership continuously, otherwise it may be surpassed by the open-source ecosystem
• Capital Efficiency Consideration
  : The cumulative loss exceeds 6.2 billion RMB. If profitability cannot be achieved within a reasonable time, the sustainability of the business model will be questioned

1. Accelerate MaaS Business Expansion
   : While maintaining the high gross profit margin advantage of on-premises deployment, Zhipu AI needs to rapidly expand its user base through the API platform (currently, the number of developer users on the API platform has exceeded 2.9 million) to increase the proportion of cloud revenue [2][3].

2. Deepen Penetration into Industry Scenarios
   : Leveraging the customer relationships and industry understanding accumulated in the government and enterprise market, convert on-premises deployment cases into reusable industry solutions to reduce marginal delivery costs.

3. Differentiated Competition in Overseas Markets
   : Zhipu AI’s first-mover advantage in the sovereign large language model market in Southeast Asia has strategic value. This model can be replicated in countries along the “Belt and Road” to form a differentiated competitive barrier [1][3].

4. Balance Technological Investment and Commercialization
   : While maintaining leading model capabilities, it is necessary to optimize the R&D investment structure and accelerate the conversion of technological advantages into quantifiable commercial revenue.

: The high proportion of technology licensing revenue does pose certain challenges to Zhipu AI’s long-term competitiveness, but this model has also built a unique competitive barrier for the company — high customer stickiness, in-depth industry understanding, and trust in the government and enterprise market. Against the backdrop of accelerated industry reshuffle, whether Zhipu AI can strike a balance between “heavy business” and “light business” will determine its position in the future competitive landscape. The current adjustment direction of the revenue structure is clear. If it can achieve the target of increasing the MaaS business proportion to 50% as planned, Zhipu AI is expected to maintain its technological leading advantage while upgrading and iterating its business model.

---

[1] 21st Century Business Herald - Detailed Analysis of Zhipu AI’s Prospectus: What’s the True Value of the “First Large Language Model Stock”? (https://m.21jingji.com/article/20251220/herald/e4f3f3583d578063b9bc60e79405dda5.html)

[2] MIT Technology Review China - The First Large Language Model Stock is Born! From Tsinghua Garden to the Hong Kong Stock Exchange, Zhipu AI’s Six Years (https://www.mittrchina.com/news/detail/15747)

[3] 36Kr - Zhipu AI and MiniMax Compete for the “First Large Language Model Stock”: Each Has Its Own Challenges Amid High Growth (https://m.36kr.com/p/3608023772447745)