Zhipu AI(2513.HK): Fell Below the HKD 1,000 Mark. Where Does Zhipu Go From Here?

Summary

• Zhipu’s share price has fallen since September and broken below the HK$1,000 level, driven by API price-war fears after DeepSeek’s cuts, increased share supply from two placements, and the fading scarcity premium of being the first listed large-model stock.
• Its flagship models are priced in the mid-to-high range, but open-source competition and low switching costs leave cloud revenue growth with weak pricing power; gross margin stands at only 26.4%, well below OpenAI and Anthropic, while training and capex pressures remain high.
• Valuation has corrected sharply, yet the P/S based on the latest ARR is still above overseas peers; the fair multiple is cut to 18x with a target price of HK$832, but competition and sentiment risks remain elevated, so investors should wait for clearer signals before taking a position.

Amid the confluence of these factors, Zhipu — hailed as the world's first large-model stock — has now retraced more than 70% from its peak.

In September, Zhipu broke below the HKD 1,000 level once again and has drifted lower ever since. The slide can be traced back to early September, when DeepSeek cut the pricing of its latest V4.1 Flash model. Fearing that Zhipu would be caught up in a price war, the market sent the stock down 10.3% that very day, and it has continued to decline since. Worries over the second share placement since listing and a global model price war have further capped any rebound, leaving the stock still unable to climb back to the HKD 1,000 level.

Against the current investment backdrop, and faced with a share price showing little sign of improvement, where should Zhipu go next?

The scarcity is fading

As we noted before, the scarcity of free-float shares in the secondary market was the main factor amplifying Zhipu's price gains. As more time passes since listing, OpenAI and Anthropic overseas have begun advancing their IPO plans, while DeepSeek and Moonshot AI have also been rumoured to be eyeing listings in the Mainland and Hong Kong markets. The market may soon see more investable targets, and the favourable factors supporting Zhipu's premium are beginning to dissipate.

Related article: Zhipu AI: Can the World’s First LLM Stock Sustain Its Valuation Premium?

On the other hand, as Zhipu needs to invest substantial capital in building data centres and paying training costs, and its earnings and cash flow remain negative, it has conducted two share placements since listing, leaving more shares in free float than before.

The most recent placement took place in September. On 13 September, Zhipu announced a placement of new shares at HKD 714 per share (a discount of approximately 10% to the previous trading day's closing price), which dealt a fresh blow to the already-battered share price. Interestingly, when the first placement was announced, Zhipu's share price actually rose, as the market generally believed the funds would be used for model iteration and to enhance competitiveness, and sentiment was largely positive.

We believe these starkly different reactions precisely reflect a shift in how the market views Zhipu. Although the proceeds of the second placement will also be used to train models, the market has begun to recognise that competition in AI models is expected to remain intense and could turn into a price war. Even if Zhipu continues to invest large sums, the investment returns it generates may be relatively limited as long as constrained pricing power and low switching costs persist. As such, with its operational pain points unresolved, Zhipu's investment sentiment will be hard to lift.

The Dilemma of price hikes and chasing the frontier

The fading of scarcity is not the only reason for the decline in Zhipu's share price — the root of the problem lies in its fundamentals.

Table 1: Pricing of Mainstream Chinese AI Models

Company	Model	Input (CNY/million tokens)	Output (CNY/million tokens)	Cache hit1
DeepSeek	DeepSeek V4-Pro	Off-peak: 4.5 Peak: 9	Off-peak: 13.5 Peak: 27	0.15–0.30
	DeepSeek V4-Flash	Off-peak: 1 Peak: 2	Off-peak: 4 Peak: 8	approx. 0.02/0.04
Alibaba	Qwen3.8-Max	12	36	Discounted
	Qwen3.8-Flash	0.8	2.7	Discounted
Zhipu	GLM-5.3	8	28	2
	GLM-5.3 Flash	0.8	2.8	0.23
MiniMax	MiniMax-M3 (≤512K)	2.1	8.4	0.42
	MiniMax-M3 (>512K)	4.2	16.8	0.84
Moonshot AI	Kimi K3	20	100	2
Source: Company announcements, iFAST Compilations. Note: 1.     A cache hit means that a segment of the input in a given request has already been processed by the model and stored in cache, and is charged at the lower cache-hit price only. 2.     AI model cost = uncached input tokens × input price + cached input tokens × cache price + output tokens × output price.

As shown in Table 1, after several rounds of price increases, Zhipu's flagship GLM 5.3 model is priced in the mid-to-high range among major domestic models, indicating that the company favours capturing mid-to-high-end model market share and establishing its industry positioning through premium pricing. However, this type of pricing strategy requires Zhipu's models to maintain a leading position over the long term in order to persuade users to keep subscribing; otherwise, it will struggle to compete with its peers.

As discussed before, once Zhipu enters the cloud services market, competition will intensify accordingly. Zhipu is currently fighting on three fronts at once: competing with Kimi on model performance, with Qwen on the AI cloud ecosystem, and facing DeepSeek's pressure on cost-performance.

Related article: Zhipu 1H26 Results: Narrower Losses, But Have the Storm Clouds Really Cleared?

The more fundamental problem lies in its business model. Open source has become the mainstream of China's AI industry, the iteration cycles of leading models have been compressed by fierce competition, and the cost of calling models has fallen sharply as model capabilities converge — all of which means Zhipu will struggle to build a durable moat. In light of this, although Zhipu's interim revenue grew by nearly 400% on the back of rising API prices and usage volumes, most of this incremental growth came from the cloud business, where pricing power is fragile. Once model performance loses its clear edge, the existing momentum could collapse, so a temporary surge in revenue does not, by itself, provide Zhipu with any protection.

Zhipu's capabilities in programming have won recognition from developers and have become a key driver of significant growth in API usage. According to Zhipu's disclosures, the company's next step is to move towards 'Co-Work' — integrating models into work environments and maximising commercial value by delivering tasks and charging on a per-task basis. The primary application is the discovery of cybersecurity vulnerabilities, expanding into security scenarios by strengthening autonomous learning capabilities as a highlight of Zhipu's products. However, investors should note that this technology is still in the R&D stage, has not yet generated revenue at scale or been validated by the market, and its impact on valuation remains uncertain.

Back to fundamentals-based valuation

Therefore, as the market comes to terms with the competitive landscape and returns to calm from euphoria, Zhipu's pricing may no longer be excessively anchored to themes, and its valuation will instead be built to a greater extent on fundamentals.

Given that Zhipu's business focus is shifting towards cloud deployment, the company has begun to gradually build distribution systems at home and abroad. Following its partnerships in Mainland China with Alibaba Cloud's Bailian platform and Huawei Cloud, Amazon Web Services (AWS) is also set to onboard Zhipu's GLM 5.3 model through Amazon Bedrock, opening up an overseas revenue channel for its API business. Meanwhile, in September Zhipu's management also raised this year's annualised recurring revenue (ARR) guidance by 5% to USD 3 billion.

While the raised ARR guidance reflects management's optimism about the company's revenue outlook, ARR as of September was only around USD 1.8 billion, meaning average monthly ARR for the remaining months would need to nearly double versus the average of the first three quarters. Moreover, although Zhipu regards Co-Work as the next step in its business expansion, the process from signing contracts with enterprises to revenue recognition takes a relatively long time, so the conversion rate from business targets to recognised revenue remains uncertain in practice.

From a profitability perspective, as we have noted, high training costs and a sharp decline in the proportion of the high-margin on-premises deployment business have pushed Zhipu's gross margin down to 26.4% in the first half of this year. By comparison, the market estimates that the gross margins of Anthropic's API business and OpenAI's paid-user compute both exceed 70% — a significant gap versus Zhipu's performance.

Even with scaled adoption of domestic chips and advances in model architectures cutting the per-token inference cost by 80% from the start of the year, the pressure that training costs and capital expenditure exert on profitability and cash flow cannot be ignored. As Zhipu aims to keep pushing the intelligence frontier and to drive self-reinforcing learning of its models across multiple scenarios, and given the need to maintain its models' leading edge, training costs are expected to remain Zhipu's main expense for some time. In addition, as we noted earlier, the impact of Zhipu building its own data centres on training costs is still unknown, and given that training demand remains enormous, Zhipu may still need to continue relying on third-party cloud service providers. Furthermore, Zhipu is currently acquiring properties and companies to meet its operational needs, so capital expenditure pressure may be hard to eliminate in the near term.

Therefore, with its financial position showing no clear improvement and industry pressure still immense, Zhipu's valuation will inevitably be discounted to some extent.

Figure 1: 12-Month Forward P/S of Zhipu

From a valuation standpoint, as of 16 October 2026, Zhipu's 12-month forward price-to-sales (P/S) ratio stands at 13x, while its P/S based on 2025 revenue has fallen from over 1,000x at its peak to the current 384x. Based on the latest ARR of USD 1.8 billion, Zhipu's P/S is approximately 28.9x. Compared with the P/S ratios implied by the valuations and latest ARR from OpenAI's and Anthropic's most recent funding rounds (see Table 2), Zhipu's P/S is clearly higher. But given the gaps between Zhipu and the two companies in terms of products and user stickiness, the current valuation level suggests Zhipu still has room to fall.

Table 2: ARR of OpenAI and Anthropic

	Company valuation	Latest ARR	Implied P/S
OpenAI	approx. USD 852 billion1	approx. USD 50 billion (as of September)	17x
Anthropic	approx. USD 965 billion2	approx. USD 65 billion (as of July)	14.8x
Source: Company announcements, compiled by iFAST. Notes: 1.     Estimated based on the new funding round conducted in March. 2.     Estimated based on the Series H funding round conducted in May.

In summary, given that Zhipu's operating and competitive environment has shown no clear improvement — with gross margins staying low and formidable rivals in every segment of the domestic AI model market — pressure on Zhipu's development persists. Together with the company's latest new share placement to raise funds and the lock-up expiry of another batch of shares in January 2027, we believe the scarcity premium Zhipu previously enjoyed will gradually dissipate, keeping its valuation under pressure. We therefore lower our fair P/S ratio for Zhipu from 20x to 18x. This level remains above the current forward P/S, reflecting our recognition of Zhipu's technological leadership among domestic models and its first-mover advantage in programming scenarios. Our target price is reduced to HKD 832, implying a potential upside of 60.5% as of 16 October 2026.

Although the potential upside is considerable, investors should note that competition among AI models at home and abroad remains fierce, and the pace of Zhipu's future model iteration and its market acceptance both carry significant uncertainty, leaving a low margin of safety. Meanwhile, global AI model investment sentiment is being weighed down by two major bearish factors: Anthropic's call to slow down frontier models, and the possibility that overseas giants' revenues fall short of expectations. Since reaching the target price requires multiple drivers, while the near-term downtrend is already set, we believe investors should wait for clearer signals before making any move, until the risk-reward becomes more balanced.

Table 3: Valuation and revenue-per-share forecast of Zhipu

	2025A	2026E	2027E	2028E
Revenue per share (HKD)	1.6	8.0	20.7	46.2
Revenue growth rate	131.9%	397.1%	158.1%	123.4%
P/S ratio	321.3	64.6	25.0	11.2
Target price at the end of 2028 (Based on 18x Forward P/S)				832
Potential upside				60.5%
Source: Bloomberg L.P., iFAST Compilations. Data as of 16 October 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 NIL 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.