Zhipu AI: Can China's AI darling sustain its valuation premium?

• Zhipu AI has become China's AI market darling, with its share price soaring after listing as investors seek one of the few pure-play listed opportunities in China's large language model (LLM) industry, pushing its valuation to a substantial premium over traditional technology peers.
  
  
• The premium is driven by future growth expectations rather than current earnings. While Zhipu is a leading Chinese foundation model developer with strong technological capabilities, the company remains heavily loss-making as it invests aggressively in AI models, computing infrastructure and talent.
  
  
• Commercialisation will be the key test. Sustaining its valuation depends on successfully converting AI leadership into recurring enterprise revenue through API services, model licensing and AI applications, while demonstrating a credible path towards profitability.
  
  
• Competition is intensifying. Zhipu faces pressure from well-funded rivals such as Alibaba, ByteDance, Tencent, Moonshot AI and DeepSeek, while the planned IPOs of other AI unicorns could erode the scarcity premium currently supporting its valuation
  
  
• Despite its long-term AI potential, execution risks remain high. Heavy AI investment, persistent cash burn and a rich valuation leave little margin for error. We initiate a sell rating with a target price of HKD 968 by end-2028, implying 20.6% downside as current valuations appear ahead of fundamentals.  

As the world's first listed large language model stock, Zhipu has attracted intense market attention since its debut. As of 21 July, its share price stood 949% above the IPO price, a clear reflection of the market's enthusiasm for AI large language models. The question, however, is whether the underlying fundamentals are sufficient to justify such a significant valuation premium.

Scarcity and Model Performance Ignite the Rally

The title of 'world's first' has not only raised Zhipu's profile, but also highlights the scarcity of listed AI large language model companies in the market. Although technology giants both inside and outside China — including Microsoft, Alibaba, and Google — all develop their own AI models, large language models represent only one part of their broader ecosystems and are therefore unable to provide pure-play AI model exposure. Zhipu's listing filled that gap. Moreover, since the vast majority of offered shares were held by cornerstone investors, the freely tradable H-share float is extremely small, further amplifying the stock's scarcity premium and pushing the share price to elevated levels.

Table 1: Comparison of the top 10 AI models

	Model	Country	Company	AI Index¹	Price (USD/1M tokens)	Response Time² (seconds)	Context Window
1	Claude Fable 5	U.S.	Anthropic	60	7.7	110.5	1M
2	GPT-5.6 Sol (max)	U.S.	OpenAI	59	4.4	149.8	1M
3	Claude Opus 4.8 (max)	U.S.	Anthropic	56	3.9	52.9	1M
4	GPT-5.6 Terra (max)	U.S.	OpenAI	55	2.2	163.0	1M
5	GPT-5.5 (xhigh)	U.S.	OpenAI	55	4.4	102.7	922k
6	Grok 4.5 (high)	U.S.	SpaceXAI	54	1.4	17.8	500k
7	Claude Sonnet 5 (max)	U.S.	Anthropic	53	1.5	227.8	1M
8	GPT-5.6 Luna (max)	U.S.	OpenAI	51	0.9	118.7	1M
9	GLM-5.2 (max)	China	Zhipu	51	0.9	13.6	1M
10	Muse Spark 1.1 (xhigh)	U.S.	Meta	51	0.8	21.9	1.05M
Note 1: The AI Index is developed by Artificial Analysis. It combines results from nine demanding evaluations: covering mathematics, science, coding, and reasoning, into a single score that reflects the overall intelligence level of an AI model. Note 2: Refers to the time taken to output 500 tokens, calculated based on the time to the first output character, the reasoning model's thinking time, and the output speed.
Source: Artificial Analysis, iFAST Compilations Data as of 13 July 2026.

In terms of business progress, Zhipu's models have not disappointed the market. Its latest GLM-5.2 leverages the advantages of a Mixture-of-Experts (MoE) architecture to address pain points in long-horizon tasks. It achieves a one-million-token (1M) context window and low response latency at a price below other US AI models, placing it near the top of Artificial Analysis rankings.

Indeed, Zhipu's GLM models have undergone three iterations since the start of the year, with the shortening intervals between releases reflecting the intensity of the global AI model race. As a leading Chinese large language model developer, Zhipu's research output has long been used by the market as a benchmark against US counterparts such as Anthropic and OpenAI. The progress shown by GLM-5.2 suggests that the technology gap between China and the United States is narrowing, which has further fuelled investor sentiment.

Revenue doubles, Yet Losses Remain Entrenched

1. On-Premise Deployment Drives Revenue Growth

Figure 1: Zhipu 2022-2025 Revenue

On-premise deployment — primarily serving enterprise clients in technology and public services — contributed 73.7% of Zhipu's revenue last year, making it the dominant revenue stream. These clients require on-premise solutions due to the sensitive and confidential nature of their data, as well as the need for customisation to meet local requirements.

Capturing enterprise market share is a critical step in the commercialisation of large language models. On-premise deployments require significant upfront capital expenditure for the related infrastructure. Once installed, the system is deeply integrated with the client's internal data, creating high switching costs that encourage long-term contracts and strengthen revenue stickiness. In terms of gross margin and product mix, enterprise-grade AI agents and general-purpose large models account for the largest share of revenue and carry higher margins than other business lines, providing a degree of earnings support for Zhipu.

2. Reliance on Third-Party Cloud Providers Is a Persistent Earnings Drag

Figure 2: Zhipu R&D Expenses as a Percentage of Revenue

In order to keep pace with major competitors' development progress and compete for market share through higher model iteration frequency, Zhipu's R&D expenses climbed sharply in 2024 and reached RMB 3.18 billion last year — more than four times its operating revenue. This ratio is also significantly higher than peers, illustrating how intense competition among large language model developers severely compresses profitability, making sustained losses an inevitable outcome.

Figure 3: Zhipu R&D Expenses as a Percentage of Revenue: Zhipu vs. Peers

The reason this enormous R&D burden has become such an intractable problem for Zhipu is that the company does not own its own cloud infrastructure. As a result, Zhipu must use cloud platforms and computing services provided by third-party cloud vendors — such as Alibaba Cloud and Huawei Cloud — when training models. In an environment where computing resources are constrained, training costs become even higher, and cloud computing service fees consequently represent the primary component of R&D expenses.

Furthermore, when providing cloud-based private deployment and API services, Zhipu must also rely on third-party cloud vendors to deliver these services. This means a portion of its operating profit is captured upstream by these providers. This also explains why Zhipu's cloud business revenue grew 292.6% y-o-y last year, while the gross margin improved by only 15.6%. However, Zhipu announced on 21 July the completion of a 1GW-scale domestic AI computing data centre. With part of the costs previously paid to third-party vendors now internalised, this development provides a foundation for gross margin improvement going forward.

3. Losses and Leverage Coexist

Figure 4: Zhipu Operating Cash Flow and Total Debt Ratio

Capital constraints are a key factor shaping Zhipu's development. The relentless pace of R&D investment and the uncertainty surrounding the commercialisation timeline make it difficult for Zhipu to generate sufficient cash flow to fund operations within a few years, leaving external financing as its only viable option.

Zhipu's interest-bearing liabilities grew 51% y-o-y last year, driven by an increase in bank borrowings. However, the persistent net loss position has kept operating cash flow consistently negative, suggesting that the company relies more on financing activities than its own operations to meet day-to-day running costs — a dynamic that makes the balance sheet increasingly fragile. In particular, Zhipu's bank borrowings are predominantly short-term in nature. Yet the current ratio has declined steadily since 2022, falling to 0.3 times last year, indicating limited capacity to cover short-term liabilities with short-term assets. The cash ratio stands at just 0.2 times, reflecting further room for improvement in overall solvency.

Long-Term Upside, Near-Term Pressure

1. Cloud Deployment Is the Next Step in Commercialisation

Revenue from the open platform and API business — primarily comprising cloud deployment and model subscription plans — grew 292.6% last year, while gross profit surged 2,150%. In the consumer segment, model choices are plentiful and switching costs for individual users are very low, requiring large language model developers to continuously offer high-performance models to prevent churn. The significant increase in API revenue reflects growing recognition of Zhipu's models among consumer users, and the improvements in GLM-5.2 may attract a new cohort of users. When the US government banned Anthropic from selling its two most powerful AI models to foreign users, Zhipu announced the full rollout of GLM-5.2 to subscribers of its Coding Plan, targeting professional users with a demonstrated willingness to pay and a need for high-performance models. Once model performance is validated by the developer community, the expanded user ecosystem will simultaneously improve paid conversion rates and retention, providing support for the long-term customer base of the API business.

While enterprise clients currently generate substantial revenue for Zhipu, excessive reliance on this segment also limits the company's ability to scale over the long term. Long-term contracts lock in revenue over a fixed period, but without new revenue streams — such as retail subscriptions and high-frequency API renewals — long-term growth may be constrained, making it increasingly important to grow cloud deployment revenue. Although Qwen, Doubao, and DeepSeek have already achieved meaningful penetration in the consumer segment, large-scale user data will help Zhipu optimise its models. Expanding the cloud deployment business would also help Zhipu diversify its revenue concentration risk and reinforce its long-term competitive position.

2. China–US Technology Gap and the Xinchuang Environment

As shown in Table 1, GLM-5.2 is the only Chinese model in the global top ten, which in itself reflects Zhipu's leading position both domestically and internationally. According to Artificial Analysis, although Qwen3.7, MiniMax-M3, and DeepSeek V4 also support a 1M context window, their output speed and response time are inferior to GLM-5.2. The technology gap relative to peers may accelerate Zhipu's ability to attract new users and build a durable technical moat.

China's Information Technology Application Innovation (Xinchuang) initiative also provides Zhipu with room to grow. Given China's ongoing push for technology self-reliance, Zhipu — having already demonstrated compatibility with domestically produced computing hardware — may find it easier to sell its products to state-owned enterprises and government entities compared with peers such as MiniMax, which procures overseas chips for model training. The State-owned Assets Supervision and Administration Commission (SASAC) launched its 'AI+' initiative for central state-owned enterprises at the start of the year, and demand for AI model deployment from government and SOE clients is expected to increase substantially. According to Zhipu's published pricing, a single cloud private deployment contract alone can generate RMB 1 million in revenue, while on-premise private deployment contracts command even higher prices. Having already established a foothold in the enterprise market, Zhipu's client base will continue to reinforce revenue visibility going forward.

3. Custom Chip Development Could Reduce Computing Costs

Reports indicate that Zhipu is exploring the development of a custom Application-Specific Integrated Circuit (ASIC) for its models. Such chips not only deliver higher performance, but are also able to meet bespoke requirements. As Zhipu is at a critical stage of the model development race, the company requires high-performance hardware to manage model subscription and training costs. Developing its own chips would help Zhipu break through vendor capacity constraints and support its high-frequency model iteration cadence, while also improving API call efficiency and lowering the cost per token. However, it is worth noting that the upfront R&D investment for custom chips is substantial, which may add further capital expenditure pressure in the near term.

4. Potential Impact of Equity Financing

In addition to debt financing, Zhipu is also meeting its capital needs through equity fundraising. Just months after its Hong Kong listing, Zhipu announced in June that it plans to apply for a listing on Shanghai's STAR Market, targeting proceeds of RMB 15 billion. The company also announced a plan to place over 19 million new shares at HKD 1,588 per share. Although the placement was oversubscribed and the share price did not fall in response, the series of consecutive fundraising activities clearly underscores Zhipu's urgent need for capital. Given the scale of future capital expenditures and the elevated uncertainty around R&D outcomes, further equity fundraising cannot be ruled out. At the same time, the lock-up period for strategic independent investors and certain major shareholders is scheduled to expire in early next year, and the combination of potential future fundraising and the release of locked-up stock could dilute existing shareholders and exert downward pressure on the share price.

Investment Merit Exists — But Not Right Now

Figure 5: 12-Month Forward P/S of Zhipu

After a sharp rally, Zhipu's price-to-sales ratio (based on 2025 revenue) exceeded 700 times as of 21 July 2026, while the average 12-month forward P/S since listing has reached 93.6 times. we believe Zhipu's current P/S ratio has limited reference value at this stage, and we therefore use its closest peers, Anthropic and OpenAI, as valuation benchmarks.

Anthropic and OpenAI completed equity financing rounds in May and March 2026 respectively, with implied equity values of USD 965 billion and USD 852 billion. Based on annualised run-rate revenue (ARR), the corresponding price-to-sales ratios are 20.5 times and 34.1 times. Meanwhile, as Chinese AI company Moonshot and DeepSeek is planning to IPO on Hong Kong and A-share markets respectively, and Anthropic and OpenAI are also pushing their IPO plans further. As a result, the scarcity premium that Zhipu is enjoying now might get declined in coming future. To sum up, as Zhipu's model performance and market reach are below those of both companies, and given that near-term capital constraints and ongoing losses are likely to weigh on its valuation, we set a fair P/S ratio of 20 times to reflect these risks.

Based on this assumption, we arrive at a target price of HKD 968 by end of 2028, implying a potential downside of only 20.6% from the closing price on 21 July 2026. This reflects our view that, while Zhipu's long-term growth momentum remains intact, the valuation has reached elevated levels after the sharp share price rally, and further upside is relatively limited in the absence of new catalysts. In addition, the competition among Chinese AI models is intense, Moonshot and Alibaba released model with higher capability, which further put stress on Zhipu.

Investors should also be mindful of Zhipu's liquidity risk. Due to the limited free float and a higher board lot entry cost compared with other technology companies such as Alibaba and Tencent, the low liquidity increases share price sensitivity to new information and amplifies price volatility, making the risk profile more pronounced than that of conventional internet stocks.

Table 2: Valuation and revenue-per-share forecast of Zhipu

	2025A	2026E	2027E	2028E
Revenue per share (HKD)	1.7	8.4	21.7	48.4
Revenue growth rate	131.9%	397.1%	158.1%	123.4%
P/S ratio	722.1	145.2	56.3	25.2
Target price at the end of 2028 (Based on 20x Forward P/E)				967
Potential upside				-20.6%
Source: Bloomberg L.P., iFAST Compilations. Data as of 21 July 2026.

Declaration: For specific disclosure, at the time of publication of this report, IFPL (via its connected and associated entities) and the analyst who produced this report hold a position in the abovementioned securities.

This research report was prepared with the assistance of artificial intelligence (AI) tools. iFAST Financial Pte Ltd does not rely exclusively on AI for content generation; the content of this report – including all investment theses, ratings, price targets and conclusions – has been independently reviewed and verified by the research analyst(s) to ensure accuracy and professional integrity.