In-Depth Investment Value Analysis Report on Zhipu AI (09691.HK)

Now I will present a systematic and comprehensive investment analysis report on Zhipu AI.

---

On January 8, 2026, Zhipu AI officially listed on the Hong Kong Stock Exchange, becoming the “world’s first large model stock”, marking a new stage of capitalization for China’s AI large model industry. This report systematically evaluates the investment value of Zhipu AI from multiple dimensions including competitive position, technological advantages, commercialization paths, ToB/ToC differentiation, and open-source/closed-source strategies[1][2].

---

Zhipu AI listed on the main board of the Hong Kong Stock Exchange on January 8, 2026, with stock code 09691.HK (the last four digits are a homophone for “AI for my lifetime”). In this IPO, 37.4195 million shares were offered globally, with 20% allocated to Hong Kong public offering and 80% to international offering. The offer price was set at HK$116.20 per share, with net proceeds from the global offering reaching approximately HK$4.173 billion[3][4].

Indicator	Figure
Offer Price	HK$116.2
Opening Price	HK$120.0 (+3.27%)
Closing Price	HK$131.5 (+13.17%)
Intraday High	Approximately HK$129
Total Market Capitalization	HK$57.89 billion
Over-Subscription Ratio	Approximately 1159x for Hong Kong public offering

Based on the estimated 2025 revenue of approximately US$100 million (about HK$780 million), Zhipu AI’s price-to-sales ratio (P/S) is approximately

. This valuation reflects the market’s recognition of its high growth expectations, while also embodying the common high-valuation characteristic of the large model industry[5][6].

Company	Valuation	Business Model
Zhipu AI	Approximately US$7.5 billion	ToB-focused, MaaS + localized deployment
Moonshot AI	US$4.3 billion (Series C)	ToC-focused, Kimi AI Assistant
MiniMax	Approximately US$4.5 billion	ToC subscription + multimodal content
OpenAI	Approximately US$300 billion	Hybrid ToB+ToC
Anthropic	Approximately US$60 billion	ToB-focused

---

According to the views of four core figures at the AGI-Next Summit on January 10, 2026, a clear tripartite competitive landscape has formed among China’s first-tier AI large model players[7][8]:

1. Zhipu AI
   - Tsinghua University-affiliated, GLM series models, ToB-focused
2. Moonshot AI
   - Founded by Yang Zhilin, Kimi AI, ToC-focused
3. Alibaba Tongyi
   - Qwen open-source ecosystem, led by Lin Junyang
4. Tencent AI
   - Joined by Yao Shunyu, strategic adjustment of the tech giant

Tang Jie, founder of Zhipu AI, clearly stated at the summit:

[7]

This judgment reveals several key investment implications:

• Technological competition remains the core; parameter scale does not equal technological leadership
• The strategic choice between closed-source and open-source will profoundly impact the competitive landscape
• Chinese large models need to pursue original breakthroughs beyond “fast-paced iteration”

According to Caixin Securities’ analysis, with the successive listings of Zhipu AI and MiniMax, the competitive landscape of the “Six Tigers of Large Models” has significantly differentiated[9]:

Company	Status	Strategic Positioning
Zhipu AI	Listed	ToB foundation model + MaaS
MiniMax	Listed	Multimodal + ToC applications
Moonshot AI	Cash reserves exceeding RMB 10 billion	Focus on Agents, not in a hurry to list
StepFun	Continued R&D	Basic model exploration
Baichuan Intelligent	Strategic adjustment	Vertical track focus
01.AI	Shift to applications	Industry implementation

---

Zhipu AI’s GLM series models complete a foundation iteration every 2-3 months, maintaining a globally leading level. The newly released

has achieved breakthrough results in multiple authoritative evaluations[10][11]:

• Ranked first among both open-source models and domestic models with a score of 68 in the
  Artificial Analysis Intelligence Index
  comprehensive evaluation
• Ranked first among open-source models and domestic models in the
  Code Arena
  coding rankings
• In the
  RAG Hallucination Ranking
  , GLM-4.5 has the second-lowest hallucination rate globally and the lowest among domestic models
• Outperforms competitors such as GPT-5.2

Zhipu AI’s technological roadmap presents three notable features:

1. Focus on Coding Capabilities
   • Went all-in on the Coding field in early 2025
   • The annual recurring revenue (ARR) of the GLM Coding Plan exceeds
     RMB 100 million
   • Paid overseas developer users exceed
     150,000

2. Intelligence Efficiency Concept
   • Tang Jie proposed the new paradigm of “Intelligence Efficiency”
   • Focuses on achieving equivalent intelligence improvement with fewer resources
   • Addresses the industry pain point of high investment but low efficiency in current large models

3. First-Mover Advantage in Agents
   • AutoGLM is the world’s first AI Agent with “Phone Use” capability
   • Launched 14 months earlier than Doubao’s mobile version
   • Has been commercialized in long-process scenarios such as food delivery ordering and flight ticket booking

Indicator	2022	2023	2024	H1 2025
R&D Expenditure (RMB 100 million)	0.84	5.3	22	15.9
R&D-to-Revenue Ratio	147%	424.7%	702.7%	835.4%
Proportion of Computing Power	-	58.9%	70.7%	71.8%

The high-intensity R&D investment reflects Zhipu AI’s strategic choice of a “technological positional warfare”, but also means that profitability is difficult to achieve in the short term[12].

---

Zhipu AI’s revenue features

[13][14]:

Period	Revenue (RMB 100 million)	YoY Growth Rate	Remarks
2022	0.57	-	Startup phase
2023	1.25	118%	Growth phase
2024	3.12	150%	Acceleration phase
2025 (Estimated)	>7	>120%	Breakthrough phase

• Localized Deployment
  : Accounted for approximately 85% in 2024, and approximately 84.8% in H1 2025
• Cloud MaaS
  : Accounted for approximately 15.5% in 2024, and approximately 15.2% in H1 2025
• It is expected that the proportion of cloud revenue will continue to increase in the medium term, and cost-effective code tools will become an important growth driver in 2026

Zhipu AI has over

, covering approximately [15].

• Internet and Technology: 38.3%
• Public Services: Approximately 30%
• Finance, manufacturing, energy, power, telecommunications, education, and other industries

Indicator	2022	2023	2024	H1 2025
Adjusted Net Loss (RMB 100 million)	0.97	6.21	24.66	17.52
Gross Margin	54.6%	64.6%	56.3%	50%
Cash Equivalents (RMB 100 million)	-	-	-	25.5

• Losses continue to expand, mainly due to increased R&D investment and computing power costs
• The decline in gross margin reflects the marginal cost pressure of cloud business
• Net proceeds from the IPO are HK$4.17 billion, which can sustain operations for approximately 1.9 years at the current cash burn rate

---

At the AGI-Next Summit, four core figures gave distinct judgments on the differentiation between ToB and ToC[16][17]:

“ToB and ToC may be different. In terms of ToB, Agents are on a continuous upward curve, and there is no sign of slowdown at present.”

“The problem with doing ToC is that DAU or product indicators are often irrelevant, or even inversely related, to the intelligence of the model.”

“For ToC, most people don’t actually need such strong intelligence.”

“Building a model is essentially creating a worldview.”
“We will focus on Agents, ‘not targeting absolute user numbers, but pursuing the upper limit of intelligence’.”

• Revenue is more predictable, and enterprise customers’ willingness to pay is relatively stable
• Localized deployment meets data security and compliance requirements
• High proportion of public service and government-enterprise customers, with certain barriers

• Revenue growth curve is relatively flat
• Payment collection cycle for government-enterprise customers is relatively long
• Customized demands may dilute gross margin

• Surged 109% on debut, with market capitalization exceeding HK$100 billion
• Over 400,000 retail subscriptions, 36x institutional subscriptions
• Business model is easier to understand by the secondary market

• High user acquisition cost, requiring continuous advertising investment
• Weak correlation between DAU and model capabilities
• Facing traffic encirclement from tech giants such as ByteDance (Doubao) and Tencent (Yuanbao)

Dimension	ToB-focused (Zhipu AI)	ToC-focused (MiniMax)
Growth Certainty	High	High but volatile
Valuation Elasticity	Low	High
Profitability Cycle	Long	Medium
Competitive Barriers	Technology + customer relationships	User habits + ecosystem
Suitable Investors	Long-term value investors	Growth-oriented aggressive investors

---

• Alibaba Qwen
  : Ranked first globally in the number of derivative open-source models and download volume
• Zhipu AI
  : GLM-4.5 series open-sourced under MIT License, with cumulative global downloads exceeding 60 million

• OpenAI
  : GPT series closed-source
• Anthropic
  : Claude series closed-source
• Moonshot AI
  : Kimi adopts a closed-source strategy

• Open-source foundation models to expand ecological influence
• Monetize high-performance versions and industry solutions through APIs and privatized deployment
• Form a closed loop of “open-source traffic generation + commercial conversion”

Advantages	Disadvantages
Rapid expansion of developer ecosystem	Technological advantages may be caught up
Rapid growth of MaaS platform ARR	Pricing power is restricted
Enhanced brand influence	Marginal revenue diminishes

According to Wall Street Journal data, the ARR of Zhipu AI’s MaaS platform has reached

, of which overseas revenue exceeds RMB 200 million, taking only 10 months to grow from RMB 20 million to RMB 500 million[18].

• Maintain technological leadership
• Gain pricing power
• Avoid technology spillover

• Slow development of developer ecosystem
• Faces cost-performance competition from open-source models
• Relatively single commercialization path

1. Open-source is not the goal, but a means
   • Zhipu AI’s open-source strategy effectively lowers the threshold for use and expands the developer community
   • The open-source version GLM-4-9B has accumulated 150,000 GitHub stars
   • The ultimate goal is to convert developers into paying customers

2. Closed-source does not equal backwardness
   • DeepSeek rose to prominence through an open-source strategy
   • The key is to find a balance between open-source and closed-source
   • Zhipu AI’s approach is “open-source foundation models, closed-source high-performance versions”

3. Business model determines route choice
   • ToB-focused players are more suitable for open-source strategies (ecological land-grabbing)
   • ToC-focused players may need closed-source protection (differentiated competition)

---

1. First-Mover Advantage
   : The world’s first listed large model enterprise, with high recognition in the capital market
2. Technological Leadership
   : GLM-4.7 ranks first among both open-source and domestic models
3. Commercial Validation
   : Three-year CAGR of revenue reaches 130%, ARR exceeds RMB 500 million
4. Customer Base
   : Over 8,000 enterprise customers, covering multiple industries
5. Financial Strength
   : Abundant capital after IPO, sustainable R&D investment

Risk Type	Specific Description	Risk Level
Loss Risk	Sustained high R&D investment, long profitability cycle	High
Competition Risk	Dual competitive pressure from tech giants and startups	High
Technological Risk	The technological gap between China and the US may widen	Medium
Valuation Risk	14.6x P/S ratio is at a relatively high level	Medium
Cash Flow Risk	Can sustain operations for approximately 2 years at current cash burn rate	Medium

• The industry is in a high-growth period; China’s large model market size is expected to exceed RMB 70 billion in 2026
• First-mover advantage and brand effect of the “world’s first large model stock”
• Broad prospects of the Coding and Agent tracks

• Sustained and expanding losses
• Relatively flat growth curve of the ToB model
• Lower capital market enthusiasm compared to MiniMax

: Neutral-Cautious (Initiation Coverage)

• Investors who are bullish on the AI large model track in the long term
• Growth-oriented investors who can tolerate short-term losses
• Institutional investors focusing on technological barriers and competitive landscape

1. Short-term
   : Mainly wait-and-see, observe the stock price performance in the first week after listing and changes in institutional holdings
2. Medium-term
   : Focus on the release rhythm of GLM-4.8/4.9 and MaaS revenue growth in 2026
3. Long-term
   : Focus on the emergence of profitability inflection points and the commercialization of Agent products

• Conservative: HK$50-55 billion (based on 8-10x P/S ratio)
• Neutral: HK$60-70 billion (based on 12-14x P/S ratio)
• Optimistic: HK$75-85 billion (based on 15-17x P/S ratio, requiring over-expected growth)

---

The AGI-Next Frontier Summit on January 10, 2026, gathered four core figures of China’s AI large model industry, whose views have important reference value for understanding industry trends and investment directions[19][20]:

Person	Core View	Strategic Implication
Tang Jie (Zhipu AI)	The gap between China and the US has not narrowed; need to explore Intelligence Efficiency	Focus on efficiency rather than mere scaling
Yang Zhilin (Moonshot AI)	Scaling has evolved into improvements in architecture, optimizers, and data	Technological deepening rather than simple computing power stacking
Lin Junyang (Alibaba)	Global first in open-source ecosystem; RL potential not fully released	Continue open-source + reinforcement learning
Yao Shunyu (Tencent)	ToC bottleneck lies in context and environment, not model capabilities	Scenario and application innovation

1. Year of Agents
   : Agent products capable of automating 1-2 weeks of human workload are expected to emerge
2. Paradigm Change
   : The new paradigm of Intelligence Efficiency may bring industry changes
3. Commercial Breakthrough
   : Large-scale implementation of enterprise AI will have in-depth business impacts

1. Technological Roadmap
   : Continue to focus on Coding and Agents, maintain technological leadership
2. Commercialization Paths
   : ToB-focused with ToC exploration, balance growth and profitability
3. Open-Source Strategy
   : Exchange open-source for ecosystem, exchange ecosystem for revenue

---

As the “world’s first large model stock”, Zhipu AI occupies an important position in China’s first-tier AI large model players. Its competitive advantages are mainly reflected in:

1. Technological Leadership
   : GLM-4.7 ranks first among both open-source and domestic models
2. Commercial Validation
   : Three-year CAGR of revenue reaches 130%, ARR exceeds RMB 500 million
3. Financial Strength
   : Abundant capital after IPO, sustainable R&D investment
4. First-Mover Advantage
   : The world’s first listed large model enterprise, with significant brand effect

• The ToB model has high growth certainty but limited valuation elasticity
• Zhipu AI’s ToB-focused positioning is suitable for long-term value investment
• Valuation may be suppressed by MiniMax’s ToB model performance in the short term

• The open-source strategy effectively expands the developer ecosystem but dilutes short-term technological barriers
• In the long term, the “open-source traffic generation + closed-source monetization” model is sustainable
• The key is whether MaaS revenue can maintain high-speed growth