Analysis of the Sustainability of Zhipu AI's B2B Model: Challenges and Opportunities Behind a Gross Margin Exceeding 50%

Based on the newly collected prospectus and market data, I provide you with a systematic analysis of the sustainability of Zhipu AI’s B2B business model against the backdrop of its gross margin exceeding 50%.

---

:

Indicator	2022	2023	2024	H1 2025
Revenue (CNH 100 million)	0.57	1.25	3.12	1.91
Gross Margin	54.6%	64.6%	56.3%	50%
Adjusted Net Loss (CNH 100 million)	0.97	6.21	24.66	17.52

Zhipu AI’s overall gross margin has indeed maintained a relatively high level of over 50%, reaching 50% in H1 2025[0][1]. This is mainly due to the

, which includes software licensing and technical services and has a relatively high premium space[2].

---

Zhipu AI adopts a typical enterprise service approach, with revenue mainly coming from:

• MaaS (Model-as-a-Service)
  : API calls, cloud deployment
• Localized Deployment (Privatized Deployment)
  : Targeted at central and local state-owned enterprises (SOEs) and large institutions
• Industry Solutions and Computing Power Rental Services

In H1 2025, the MaaS business contributed

, with localized deployment revenue reaching CNH 162 million[2][3].

• Has served a total of over
  8,000 institutional customers
• 9 of China’s top 10 internet companies
  are Zhipu AI users
• The top five customers contributed
  45.5% of revenue
  in 2024, with the largest customer contributing 19%

: Among the disclosed lists of top five customers each year, , which reflects the drawback of the one-time delivery AI deployment model – unsustainable charging, requiring continuous acquisition of new customers[1].

---

The prospectus shows that the

, resulting in gross profit losses. Zhipu AI admitted in the prospectus that this is due to “reducing service prices in line with market trends”[2].

In 2025, leading internet giants including

launched a fierce price war, pushing API call prices to “floor levels” or even free. This has formed a dimensionality reduction strike against independent vendors like Zhipu AI[3].

Indicator	2022	2023	2024	H1 2025
R&D Investment (CNH 100 million)	0.84	5.29	22	15.9
R&D Expense Ratio	146%	423%	705%	835.4%

In H1 2025, Zhipu AI’s revenue was CNH 191 million, while its R&D expenditure during the same period reached as high as

, meaning that for every CNH 1 earned, it invested more than CNH 8 in R&D[2][3]. This “oversaturated” investment is mainly used for computing service fees (CNH 1.145 billion) and R&D personnel salaries.

As of June 30, 2025, Zhipu AI’s cash and cash equivalents on books were approximately

, but compared to its annual cash burn rate of billions of yuan, this fund may not last long[3]. This is also the core reason for its eagerness to go public for financing.

---

Despite facing challenges, Zhipu AI’s B2B model still has the following

:

• Privatized deployment means that once customers adopt it, the migration cost is extremely high
• Central and local state-owned enterprises (SOEs) have rigid demand for “security, controllability, and privatization” and are willing to pay a premium
• 9 of the top 10 internet companies are Zhipu AI users, forming a strategy of
  binding leading players and radiating the industry
  [4]

According to Frost & Sullivan data:

• In 2024, the market size of China’s large language model (LLM) market was approximately CNH 5.3 billion, of which institutional customers contributed CNH 4.7 billion
• The
  enterprise-level market size is expected to reach CNH 90.4 billion by 2030
  , with a compound annual growth rate (CAGR) of
  63.7%
  from 2024 to 2030[1]

• Originated from Tsinghua University’s Knowledge Engineering Laboratory (KEG), and launched the development of the GLM pre-training framework in 2020
• The first company in China to release the GLM-130B, a large model with 100 billion parameters
• The cumulative downloads of the GLM series models in the open-source community have exceeded
  45 million times
  [2]

---

:

• Price wars have led to losses in the cloud business, while localized deployment maintains high gross margins but has limited scale
• Sustained high R&D investment and losses have put the company under cash flow pressure
• High customer concentration and lack of repurchase stickiness make the one-time delivery model unsustainable

:

1. Technological Generational Gap
   : Whether it can maintain a leading position in model capabilities and avoid being completely replaced by internet giants
2. In-Depth Customer Development
   : Shifting from a one-time delivery model to a continuous service model to increase customer lifetime value
3. Cost Control Capability
   : Whether it can improve the profitability of the cloud business as model inference efficiency increases

Zhipu AI’s B2B model has advantages in

, but are core risk points. This model can still be maintained in the short to medium term, but to achieve sustainable profitability, it needs to cross the two thresholds of technological iteration and cost control.

---

[0] Wall Street CN - “In-Depth Analysis of the Prospectus of AI LLM Unicorns: MiniMax to C, Zhipu AI to B” (https://wallstreetcn.com/articles/3761823)
[1] China Business Network - “Going Overseas and Going Public: China’s First Batch of Large Models Finally Break Through” (https://m.cbndata.com/information/294851)
[2] Wall Street CN - “In-Depth Analysis of the Prospectus of AI LLM Unicorns” (https://wallstreetcn.com/articles/3761823)
[3] OFweek AI Network - “CNH 6.2 Billion Loss in 3.5 Years: Tsinghua Affiliated Unicorn Zhipu AI Goes Public ‘Bleeding’” (https://m.ofweek.com/ai/2025-12/ART-201700-8400-30677243.html)
[4] CNFOLL Finance Channel - “Two Bell-Ringing in 48 Hours: China’s AI Large Models Face Valuation Reassessment” (http://mp.cnfol.com/50789/article/1768355697-142212748)