Introduction: A week ago, Zhiyun was still trapped in the pressure from the dilution caused by the share placement, falling more than 46% in September alone; a week later, its model appeared simultaneously on Amazon's cloud shelf, in the default list of overseas programming tools, and on an international investment bank's "buy" list. The quality of this rebound and how far it can go depends on three things that are currently happening.
1. What Happened: The Overseas "Triple Strike" of GLM-5.3
On October 6, Amazon Web Services (AWS) officially announced the integration of Zhiyun's GLM-5.3 into its model service platform Amazon Bedrock, with revenue sharing based on the model's usage volume. According to public reports, aside from AWS, Zhiyun has also established similar revenue-sharing models with multiple overseas cloud providers recently.
This is not Zhiyun's first landing overseas, but the third stop in the last week:
The AI programming tool Cursor has launched GLM-5.3 and GLM-5.3-Flash, revealing that GLM-5.3 achieved the highest score (42.6%) in its proprietary programming evaluation CursorBench 4.0 — this evaluation task comes from real Cursor usage scenarios, covering multi-file processing, code editing, refactoring, and problem investigation;
The German sovereign AI company Aleph Alpha when releasing its new model Kolibri, showed in its technical report that GLM-5.2 and GLM-5.3 were used for supervised fine-tuning data production;
Cybersecurity certification: The evaluation results from CAISI under the National Institute of Standards and Technology (NIST) were cited in an overseas AI security research report released on September 29, listing GLM-5.3 as the strongest open-weight model in terms of cybersecurity capability as of the evaluation time.
From being used to being certified by third parties, GLM-5.3 is transitioning from a "Chinese open-source model" to one of the increasingly utilized options in the global developer toolchain.
2. Why the International Investment Banks Upgraded Their Ratings at This Time
Around October 5, a leading international investment bank upgraded Zhiyun's rating from "neutral" to "buy," giving a 12-month target price of HKD 1,560 based on DCF (neutral interpretation). The three reasons given by this bank deserve individual examination:
The commercialization path has become clearer. The bank raised its ARR forecast for Zhiyun by the end of 2026 from USD 2.7 billion to USD 3.2 billion, based on the growing demand for tokens and the new commercial cooperation terms being executed with major cloud providers in China and globally starting in October; at the same time, it raised its revenue forecasts for 2026 to 2028 by 3% to 12%.
Expansion of computational power and improvement of the balance sheet. Zhiyun completed refinancing of approximately USD 4 billion and USD 5 billion in July and September, respectively, and the supply of computational power may no longer be the main constraint on revenue growth in the short term.
Improvement in cost efficiency. Management disclosed in a September conference call that, thanks to the cost-effectiveness of GLM-5.3, the theoretical gross margin for reasoning could reach up to 80%.
It is worth noting a detail: the rating has been upgraded, but the target price was slightly adjusted down from HKD 1,610 to HKD 1,560, and the net loss predictions also varied (adjusted for RMB 5.1 billion, 5 billion, and 500 million for 2026 to 2028). This seems more like a market reassessment of the clarity of the commercialization path rather than merely an "earnings upgrade" — what the bank is optimistic about is the clarity of the commercialization path, rather than an overall increase in profit expectations.
3. Revenue Sharing with Cloud Providers: Open-source Models Transitioning from Cost Items to Revenue Items for the First Time
The integration of GLM-5.3 into Amazon Bedrock, seen within the context of Zhiyun's commercialization path, may have an even greater significance than a single piece of news.
Previously, the mainstream market concern regarding Zhiyun was that open-source equated to free, and regardless of how strong the model's capabilities, monetization potential was in doubt. The revenue-sharing model with cloud providers may change this equation — GLM series open-source models are no longer just ecological investments, but are products on cloud platform shelves; third-party cloud platforms have transformed from potential substitutes to distribution channels. Domestically, Zhiyun has signed similar revenue-sharing agreements with leading cloud providers like Alibaba Cloud's Bailian platform, and Huawei Cloud has put GLM-5.3 online and reached an intention for similar cooperation.
Some market perspectives draw references from leading overseas AI companies: the combination of models with cloud platforms has validated the capability for large-scale distribution, and related companies' revenue from usage has significantly increased over more than a year. Zhiyun's management stated in a conference call that after starting revenue-sharing with cloud service providers, the company can reduce the computational pressure on some business areas and accelerate the expansion of global business — where partners provide inference infrastructure and sales channels, and the company receives revenue from usage sharing, decoupling the revenue growth from its own capital expenditures on computational power. Whether this path can be replicated or to what extent remains uncertain.
4. A Set of Numbers That Should Be Viewed Together: Increased Investment from the South, Significant Stock Price Drop
In September, Zhiyun received net purchases of HKD 7.884 billion from southbound funds, ranking among the top for net purchases in Hong Kong Stock Connect; the total trading volume through Hong Kong Stock Connect in September reached HKD 93.376 billion, the highest among all targets; the latest shareholding increased to 37.2901 million shares, more than 460% compared to early July.
However, in the same month, Zhiyun's stock price dropped cumulatively by 46.36%.
The persistent increase in southbound fund holdings while the stock price is falling — one important background for this divergence is the approximately USD 5 billion financing announced on September 13: consisting of a USD 2 billion share placement and a USD 3 billion zero-interest convertible bond, with the placement price of HKD 714 representing a discount of about 9.96% compared to the previous closing price, and if all convertible bonds are converted, it will add approximately 5.3% to the equity. The market generally views the dilution from the share placement as one of the significant factors for the decline in September, while southbound funds choosing to continue buying during the decline may reflect that some domestic funds' pricing anchor is not on the short-term stock price but on the realization pace of the commercialization path — this interpretation still carries uncertainty. Whether the September decline has partly reflected the impact of the share placement dilution remains to be observed; the overseas deployment of GLM-5.3 in October and the opening of the revenue-sharing channels with cloud providers provide new observation anchors.
5. From "Selling Models" to "Collecting Rent"
Some market views summarize the core logic of Zhiyun's recent rise in one sentence: the business model is switching from "selling models" to "collecting rent."
"Selling models" involves project-based work, privatized deployment, one-time revenue, with high gross margins but clear scale limits; "collecting rent" means cloud platforms sharing revenue based on usage, resulting in continuous token billing revenue, with relatively low gross margins but much higher scale ceilings. According to the company’s mid-term report, Zhiyun's open platform and API revenue reached RMB 825 million in the first half of 2026, a year-on-year increase of about 27 times, accounting for 86.5% of total revenue; the API business gross margin improved from -0.4% to 24.6%. The overall gross margin decreased from 50% to 26.4%, which isn't necessarily a deterioration but likely reflects a structural change in the revenue composition from high-margin privatized deployments to scaled APIs — however, the trend of profit quality still needs continuous tracking. If the revenue-sharing model works on overseas cloud platforms, the slope of this revenue curve may be significantly steeper, which awaits data verification.
The bank ranked Zhiyun's ARR performance first among its Chinese peers (neutral interpretation) — according to this judgment, Zhiyun is potentially one of the leading Chinese large model companies at the intersection of "open-source ecology" and "commercial realization," with the competitive landscape still evolving.
6. Monitor These Three Signals
The actual confirmation rhythm of revenue from cloud provider partnerships. Management stated that revenue sharing with cloud service providers would begin to be recognized from October. If the revenue data for October and November can validate the revenue-sharing model’s scale potential, it may become an essential observation point for the market to re-evaluate its valuation logic.
The digestion progress of the share placement dilution. Whether the approximately 46% decline in September has adequately priced in the dilution effects of the share placement and convertible bonds depends on the subsequent inflow rhythm of southbound funds and changes in institutional holdings. The combination of increased southbound holdings and low stock prices has historically shown similar situations, but past performance does not guarantee future outcomes, and this combination itself does not constitute a reliable rule.
The actual usage volume of GLM-5.3 in the overseas developer community. The integrations of Cursor and Amazon Bedrock are "on the shelves," but the real test lies in "the usage volume after being shelved." If GLM-5.3 establishes stable usage habits in overseas programming scenarios, the revenue-sharing ramp-up curve overseas may outpace domestic growth.
Data Notes:
Market Condition: Closed at HKD 665 on October 5 (+6.15%); peaked at HKD 720 during trading on October 6 (+8.3%), reported HKD 703 at 10:08 (+5.7%), total market capitalization approximately HKD 313.4 billion; MINIMAX-WP, Alibaba-W, Baidu Group-SW rose around 10:08 trading data. Public market information.
Rating: A leading international investment bank upgraded its rating from "neutral" to "buy," target price HKD 1,560 (adjusted from HKD 1,610), ARR forecast raised from USD 2.7 billion to USD 3.2 billion, and 2026–28 revenue forecast raised by 3% to 12%. Media references to research reports from October 2 to 5, 2026.
Overseas Development: Cursor launched GLM-5.3 (CursorBench 4.0 score of 42.6%); Amazon Bedrock integration with revenue sharing based on usage; Aleph Alpha Kolibri technical report; overseas AI security research report citing NIST/CAISI evaluation. Company announcements and public reports from September to October 2026.
Financials: H1 2026 revenue of RMB 954 million (+399.7%), API revenue of RMB 825 million (accounting for 86.5%), adjusted net loss of RMB 1.964 billion; ARR approximately USD 1.8 billion, year-end guidance of USD 3 billion. Company mid-term report and conference call from August to September 2026.
Southbound Funds: Net purchases of HKD 7.884 billion in September, total trading volume through Hong Kong Stock Connect of HKD 93.376 billion ranked first, shareholding increased over 460% compared to early July. Public data from October 2026 (Wind standards).
Financing: Approximately USD 5 billion (USD 2 billion placement + USD 3 billion zero-interest convertible bond), placement price of HKD 714, representing a discount of about 9.96%. Company announcement on September 13, 2026.
Disclaimer: This content is provided for general information and market commentary purposes only, based on publicly available data as of the time mentioned in the text. Relevant market data, expectations, and probabilities may change with market conditions. The views and investment strategies cited from third-party organizations, analysts, or others only reflect the perspectives of those third parties at specific times and do not represent the views or recommendations of BIT. This article does not constitute investment advice, investment research, offers, solicitations or recommendations for any securities, investment products, or trading strategies, nor should it be taken as a basis for any investment decisions. Financial markets involve risks, and prices and market performance may fluctuate, and historical performance and past market trends do not represent or guarantee future results. Investors should independently assess related risks based on their own circumstances and seek professional advice if needed.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
