Deep Financial Analysis of Zhipu AI: Assessment of Commercialization Path and Profitability Prospects

Based on the obtained information, I will conduct a systematic analysis of Zhipu AI’s financial status and commercialization path from multiple dimensions.

---

Founded in 2019 by the core team of the Knowledge Engineering Group (KEG) at the Department of Computer Science, Tsinghua University, Zhipu AI is one of the representative enterprises in China’s large language model sector. The company’s technological development history is as follows:

Time Node	Key Technological Breakthrough
2021	Initiated R&D of the GLM framework
August 2022	Launched the 100-billion-parameter large model GLM-130B (released earlier than ChatGPT) [1]
2023	Launched ChatGLM, one of China’s earliest conversational large language models
Recent Years	Continuously iterated GLM-4 to rival GPT-4

The company’s core technical team originates from the KEG Lab, which has been deeply engaged in research on natural language processing and knowledge graphs since 1996, and is a top academic institution in China’s AI field. This “industry-university-research” integration model has laid a solid technical foundation for Zhipu AI, enabling it to gain a first-mover technical advantage in the competition among domestic large language models.

---

According to data from the Hong Kong Stock Exchange prospectus [1][2]:

Financial Indicator	2022	2023	2024	H1 2025
Revenue (RMB 100 million)	0.57	~1.5	3.12	1.91
Net Loss (RMB 100 million)	1.44	7.88	29.58	23.58
Loss-to-Revenue Ratio	252.3%	~525%	948%	1234.6%

• Accumulated losses of RMB 6.2 billion over three and a half years
  , with only RMB 685 million in revenue, resulting in a severe imbalance between revenue and losses [2]
• The scale of losses is expanding at an accelerated pace, with the loss-to-revenue ratio reaching as high as
  1234.6%
  in H1 2025 [1]
• Although revenue growth remains high (more than fivefold growth in three years), it still cannot cover the high operating costs

Zhipu AI’s cost structure shows a distinct characteristic of “heavy R&D, light marketing”:

Expense Item	Amount in 2024	Percentage of R&D Expenditure
Computing Power Service Fee	RMB 1.553 billion	70.7%
Total R&D Expenditure	RMB 2.195 billion	100%

1. Extremely high R&D investment intensity
   : The R&D expenditure to revenue ratio in H1 2025 reaches
   8.4:1
   , which is more than 5 times that of OpenAI and more than 8 times that of Anthropic [1]
2. Rigid computing power costs
   : Over 70% of R&D expenditure is used for computing power service fees, and this part of the cost grows rigidly as the parameter scale of large models continues to expand [1]
3. High talent costs
   : As of June 2025, R&D personnel account for as high as 74% of the total workforce

Indicator	Amount	Explanation
Book Cash as of the end of June 2025	RMB 2.55 billion	—
Cash Outflow from Operating Activities in H1 2025	RMB 1.33 billion	—
Cash Runway (Based on Operating Cash Flow)	~1.9 years	—
Cash Runway (Based on Adjusted Net Loss)	~1.5 years	—
Available Funds (Including Credit Lines) as of the end of October 2025	RMB 8.943 billion	Including RMB 2.8 billion in cash + short-term investments and RMB 6.1 billion in bank credit lines

This explains why Zhipu AI is rushing to launch its IPO at this time — at the current cash burn rate, the company needs to complete a new large-scale financing round within 1-2 years to maintain operations.

---

Zhipu AI’s revenue mainly comes from two business segments [1][2]:

Business Type	Revenue Share in 2024	Feature Analysis
MaaS (Model-as-a-Service) Platform	15%	Cloud-based API service, with low unit profit but scalable
Local Deployment	85%	Customized solutions for large enterprises, with high gross profit but limited scalability

• Gross profit margin
  : 54.6%, 64.6%, and 56.3% in 2022-2024 respectively, and 50% in H1 2025 [1], remaining consistently above 50%
• Revenue structure adjustment
  : The decline in gross profit margin reflects a strategic transformation from “high gross profit, low scale” to “medium gross profit, high scale”
• Platformization trend
  : The CEO stated that the company plans to increase the revenue share of API business to 50% [1]

Zhipu AI has chosen a development path different from that of OpenAI and Anthropic:

Comparison Dimension	OpenAI	Anthropic	Zhipu AI
Primary Revenue Source	Consumer Subscription + Advertising	Enterprise API Services	Enterprise Private Deployment + MaaS
Customer Structure	Consumer-focused	Enterprise customer-focused	Government and enterprise customer-focused
Expected Break-Even	2030	2027	Unclear
R&D Expenditure/Revenue Ratio	1.56:1	1.04:1	8.4:1

1. Integration of Government (G-end) and Enterprise (B-end) Businesses
   : Integrate government and enterprise business resources to reduce duplicate investment and improve workforce efficiency [1]
2. Focus on MaaS platform
   : Aim to become a “model supermarket” and provide standardized API services [1]
3. Advantage of private deployment
   : Provide customized solutions for large enterprises and institutions, with high customer stickiness

According to IDC data, the top 5 players in China’s MaaS market in H2 2024 [1]:

Ranking	Enterprise	Market Share
1	Baidu	26%
2	Alibaba	19%
3	ByteDance	16%
4	Tencent	10%
5	SenseTime	5%

• Encirclement by tech giants
  : Baidu, Alibaba, ByteDance, and Tencent dominate the market with their cloud computing ecosystem advantages
• Competition from startup peers
  : Players like MiniMax, Moonshot AI, and 01.AI are also rushing to go public
• Price war pressure
  : DeepSeek reduced API prices to the industry floor in May 2025, triggering an industry-wide price war [1]

---

1. Favorable revenue growth trend
   : Revenue grew from RMB 57 million in 2022 to RMB 312 million in 2024, with a compound annual growth rate of 134%
2. Relatively stable gross profit margin
   : Remaining consistently above 50%, indicating that the core business has self-sustaining capabilities
3. Vast market space
   : The prospectus predicts that the scale of China’s enterprise AI market will reach RMB 100 billion by 2030 [1]
4. Solid technical barriers
   : Backed by Tsinghua University, with full-stack independent R&D, and has competitive advantages in areas such as Chinese language understanding
5. Recognition from the capital market
   : Its over 1000x subscription rate for the Hong Kong Stock Exchange IPO indicates investors’ recognition of its long-term value

1. Continuous loss risk
   : The scale of losses continues to expand, putting pressure on cash flow
2. Rigid computing power costs
   : Over 70% of R&D investment is used for computing power, and cost pressure persists under international chip restrictions
3. Living space squeezed by tech giants
   : Cloud vendors dominate the MaaS market with their ecosystem advantages, making it difficult for “pure-play” large model vendors to compete head-on
4. Profits eroded by price wars
   : API prices continue to decline, and the industry has seen a phenomenon of “selling cloud services at a loss” [1]
5. Over-reliance on a single business
   : No other businesses to “provide blood transfusion”, fully relying on large model monetization

Indicator	Zhipu AI	OpenAI	Anthropic
R&D Expenditure/Revenue Ratio	8.4:1	1.56:1	1.04:1
Share of Computing Power in R&D Investment	70%+	Low (with strategic investment)	Low (supported by AWS/Google)
Expected Break-Even	Unclear	2030	2027
Backing Parties	Primary Market Financing	Microsoft	Amazon + Google

---

• Track advantage
  : AI large models are one of the most certain tech tracks in the next decade
• Technical barriers
  : Tsinghua University-originated team, with full-stack independent R&D of 100-billion-parameter models
• First-mover advantage
  : One of the earliest domestic enterprises to lay out large models
• Platform potential
  : The MaaS platform has network effects and scale effects

• Continuous loss risk
  : The scale of losses continues to expand, putting pressure on cash flow
• Technological iteration risk
  : AI technology routes are changing rapidly, and the company may be disrupted by new technologies
• Intensified competition risk
  : Price wars and technological catch-up by tech giants squeeze the company’s living space
• Regulatory policy risk
  : Uncertainty in AI industry regulatory policies
• Valuation bubble risk
  : A valuation of RMB 24.3 billion corresponds to H1 revenue of RMB 191 million, resulting in an extremely high price-to-sales ratio

: If API call volume expands rapidly, inference costs decrease, and market share increases, referring to Anthropic’s experience, the company may achieve break-even in 2027-2028

: Maintaining the current investment intensity and commercialization pace, the break-even time may be in 2029-2030

: If price wars continue and R&D investment cannot be converted into revenue, the company may face capital chain pressure and need to continue relying on external financing

---

Zhipu AI’s commercialization logic is clear — providing standardized API services through a MaaS platform while offering private deployment solutions for large government and enterprise customers. This positioning is similar to Anthropic’s and is feasible in the enterprise market.

1. Cost side
   : Computing power costs remain high, and R&D investment intensity far exceeds that of international peers
2. Revenue side
   : Price wars compress profit margins, and the ecosystem advantages of tech giants are difficult to overcome
3. Timeline
   : Referring to international experience, Anthropic expects to achieve break-even in 2027, and Zhipu AI’s path may be longer

For ordinary investors, Zhipu AI, as the “first large model stock”, has important reference value and concept speculation opportunities, but based on current financial data:

1. Short-term
   : Pay attention to market sentiment and capital speculation after the IPO
2. Mid-term
   : Track changes in the revenue share of API business and improvements in gross profit margin
3. Long-term
   : Observe the downward trend of computing power costs and the improvement of market share

: The price-to-sales ratio corresponding to the current valuation is extremely high, and investors need to carefully assess their risk tolerance before investing.

---

[1] 36Kr - “Wanting to Copy Anthropic’s Model, Zhipu Still Faces Many Challenges” (https://www.36kr.com/p/3628998562776324)

[2] 199IT - “Zhipu AI: H1 2025 Revenue of RMB 191 Million, Accumulated Losses of RMB 6.2 Billion” (https://www.199it.com/archives/1803166.html)

[3] ChinaVenture - “Zhipu AI, Ranked ‘Second’” (https://news.pedaily.cn/202512/558915.shtml)

[4] Guancha.cn - “When the Market Prices Zhipu, China’s Large Models Are Being Evaluated Together” (https://www.guancha.cn/economy/2025_12_26_801778.shtml)

[5] MIT Technology Review China - “The First Large Model Stock is Born! Zhipu’s Six Years from Tsinghua Campus to the Hong Kong Stock Exchange” (https://www.mittrchina.com/news/detail/15747)