--- title: "After continuous share repurchases, when will SF Express's AI efficiency start to reflect in its income statement?" type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/42978068.md" description: "In the nighttime logistics transfer center, a conveyor belt stopping for a few minutes may not cause significant losses. However, if one vehicle arrives late, one unloading dock gets congested, or a route is misjudged, it can cause subsequent personnel, vehicles, and flights to be out of sync. The larger the logistics network, the higher the loss caused by such minor deviations. The application of AI is of great significance to the logistics industry, although it is rarely mentioned at press conferences. On July 23, SF Holding repurchased 436,800 H-shares, investing approximately 14.4321 million HKD, with a transaction price of HKD 32.62 to 33.24 per share..." datetime: "2026-07-28T04:17:30.000Z" locales: - [en](https://longbridge.com/en/topics/42978068.md) - [zh-CN](https://longbridge.com/zh-CN/topics/42978068.md) - [zh-HK](https://longbridge.com/zh-HK/topics/42978068.md) author: "[港股研究社](https://longbridge.com/en/profiles/3199113.md)" --- # After continuous share repurchases, when will SF Express's AI efficiency start to reflect in its income statement? In the logistics transfer center at night, a conveyor belt stopping for a few minutes may not cause significant losses. However, if one truck arrives late, one unloading dock gets congested, or a route is misjudged, it can cause subsequent personnel, vehicles, and flights to be out of sync. The larger the logistics network, the higher the loss caused by such minor deviations. The application of AI is of great significance to the logistics industry, although it is rarely mentioned at press conferences. On July 23, SF Holding repurchased 436,800 H-shares, investing approximately HK$14.4321 million, with a transaction price of HK$32.62 to HK$33.24 per share, and held the shares as treasury stock. From July 20 to 23, the company repurchased H-shares for four consecutive trading days, cumulatively repurchasing 1.7666 million shares, investing approximately HK$58.0689 million. As of July 23, the cumulative repurchase of H-shares under this authorization reached 5.8604 million shares. Almost at the same time, SF's larger A-share repurchase program has concluded. From September 2025 to July 21, 2026, the company cumulatively repurchased 160.16 million A-shares, accounting for 3.04% of the total share capital, with an expenditure of approximately RMB 5.999 billion. The relevant shares are planned to be cancelled and the registered capital reduced. Financial report data shows that in the first quarter of 2026, SF's net profit attributable to shareholders increased by 13.05% year-on-year, and non-GAAP net profit increased by 17.42%. The profit growth rate has already exceeded the revenue growth rate of 6.14%. The repurchase indicates management's attitude towards shareholder returns. The next question is closer to operations: SF's AI and digitalization capabilities, which have been invested in for many years, should leave a sufficiently deep mark on the income statement. **Repurchases enhance the 含金量 (gold content) per share, but operational efficiency still needs to be tested by cash flow** SF's H-share repurchase plan announced in March showed that funds of no more than HK$500 million would be used, sourced from internal resources and self-raised funds. After obtaining shareholder approval on May 8, the upper limit of H-shares the company could repurchase was 24 million shares. Continuous repurchases themselves have clear capital management implications, and combined with the A-share cancellation-style repurchases, they have collectively reduced the circulating share capital. Share capital contraction can directly improve the calculation basis for earnings per share, but SF currently still needs to rely on its main business to provide continuous support. Financial report data shows that in the first quarter of 2026, SF achieved operating revenue of RMB 74.142 billion, an increase of 6.14% year-on-year; net profit attributable to shareholders was RMB 2.526 billion, up 13.05% year-on-year; and non-GAAP net profit attributable to shareholders was RMB 2.317 billion, up 17.42% year-on-year. Profit growth outpacing revenue indicates that the business structure adjustment and lean management promoted by the company previously are still releasing effects. During the same period, the net cash flow from operating activities was RMB 3.365 billion, down 17.16% year-on-year, showing that profit improvement has not yet fully translated into cash flow. The gap between operating cash flow and net profit is also worth noting. In full year 2025, SF's operating cash flow was approximately RMB 27.6 billion. After deducting approximately RMB 9.6 billion in asset-related investments, free cash flow was as high as nearly RMB 17.9 billion, indicating that SF's cash creation capability is not weak. The weakening of cash flow in the first quarter of 2026 may be related to the expansion of business scale, settlement cycles, and the pace of capital expenditures. The semi-annual report needs to provide a clearer explanation. June operating data provided another set of clues. Data showed that SF Express logistics business revenue was RMB 20.017 billion, up 0.28% year-on-year; business volume was 1.389 billion tickets, down 4.86% year-on-year; single-ticket revenue rose to RMB 14.41, up 5.41% year-on-year. Supply chain and international business revenue was RMB 7.863 billion, up 24.97% year-on-year. The combined revenue of these two businesses was RMB 27.88 billion, up 6.19% year-on-year. Based on the above data: SF's revenue structure is tilting towards products with higher single-ticket value, supply chain, and international business. While such changes are beneficial for improving profitability quality, they also increase the difficulty of analysis. Thus, profit margin improvement may come from product price adjustments, changes in customer structure, international business expansion, or cost savings brought by AI scheduling and automation. Therefore, the upcoming formal semi-annual report needs to provide more detailed expense and cost data to identify the actual contribution of technological efficiency. **Logistics AI steps out of the lab, SF begins using network scale to train decision-making capabilities** Peter Drucker emphasized "doing the right things right." For logistics AI, the "right thing" usually refers to finding better arrangements among hundreds of millions of waybills, vehicles, and operation nodes. SF Technology disclosed that the company has established an intelligent decision-making system covering prediction, planning, early warning, scheduling, and feedback, capable of supporting the planning of hundreds of millions of waybills across the entire network. The scheduling system can dynamically adjust transfer plans based on package volume, equipment, vehicle, and personnel status, and can also combine weather, flight delays, and traffic control information to warn of potential timeliness risks. As of the end of 2025, SF's vertical large model for logistics consumed over ten billion tokens daily. AI agents have been applied in scenarios such as prediction, planning, marketing, fulfillment, customer service, customs affairs, and data analysis. Public materials also show that related multi-agent clusters have entered more than 30 internal business scenarios. The persuasiveness of the above data is stronger because, compared to model parameters, they are closer to the real needs of logistics business. SF possesses a large amount of real-time data covering orders, waybills, aircraft, trucks, sorting equipment, and last-mile delivery. In this way, the model can repeatedly receive feedback from real operational results. SF's network scale advantage can thus be further transformed into training data, execution capability, and error-correction speed. Some scenarios have already provided quantitative results. SF disclosed that digital twin products have been applied in over a hundred transfer centers and Ezhou Huahu International Airport. Optimization of small-package sorting plans can increase site capacity by up to 20% and shorten sorting time by more than 10%. The airport digital twin platform reduces the daily scheduling distance of vehicles by 40% and improves overall timeliness by 5% to 15%. The above data belongs to the project effects disclosed by the company and still needs observation regarding their coverage and contribution to the group's overall costs. SF's technical capabilities have also begun to output to customer scenarios. The New Balance China Smart Delivery Center, put into operation in July, uses SF's self-developed supply chain platform and AI Agent to coordinate order timeliness, operation standards, and task loads. SF Technology also explicitly stated that its digital twin commercialized products are now providing customized services to the market. Therefore, a more accurate judgment of SF's commercialization path is that SF AI first completes verification within its internal network, then embeds mature modules into supply chain services. External revenue is unlikely to be listed separately as an AI business in the short term, but technical capabilities may increase the delivery depth, customer stickiness, and contract value of supply chain projects. **The semi-annual report does not need to list AI revenue separately, but must explain why expenses decreased** Traditional logistics companies find it difficult to disclose AI revenue in the manner of software companies. After intelligent scheduling is embedded into the transportation system, path models begin to enter vehicle arrangements, and visual algorithms connect to sorting equipment. Technical benefits will be dispersed into transportation costs, labor costs, management expenses, and asset turnover. Therefore, the focus of observing SF's semi-annual report needs to return to several basic indicators. Under the circumstance of low revenue growth in the express business, whether single-ticket transportation and transfer costs continue to decline; whether the manpower and outsourced capacity saved by unmanned vehicles and automated equipment can cover new depreciation and maintenance expenses; whether the management expense ratio can continue to decline after AI agents enter customer service, customs affairs, and data analysis; and whether profit growth can once again drive the improvement of operating cash flow. In addition, we need to avoid attributing all profit increases to AI. Data shows that in June, single-ticket revenue increased by 5.41% year-on-year, while business volume declined, indicating that changes in SF's product structure and pricing have already impacted revenue quality. The issue is that the rapid growth of supply chain and international businesses will also change the group's profit composition. Therefore, how much of the AI contribution there is requires cross-validation through same-basis costs, unit resource efficiency, and cash returns. Howard Marks often focuses on the current and future profitability of assets when discussing corporate value. Following this line of thought, SF does not need to rely on independent AI revenue for recognition. As long as AI can continuously reduce unit fulfillment costs, improve resource utilization, and reduce profit volatility, SF's overall profitability will improve accordingly. It should be noted that such value is usually not released centrally in one quarter. Because the logistics network contains a large number of offline assets and complex execution links, algorithm upgrades also require organizational processes, automated equipment, and frontline operations to work together. The faster the technological expansion, the more newly added depreciation, R&D expenses, and system transformation costs need to be included in investment return calculations. SF's AI path provides a more realistic sample for traditional industries: technological capabilities first enter the cost structure, then enter customer solutions, and finally settle into more stable profitability and cash flow. Large-scale enterprises possess more data and bear higher organizational complexity. If SF can transform this complexity into algorithmic advantages in the future, AI will become the long-term operating system of the logistics network. ### Related Stocks - [002352.CN](https://longbridge.com/en/quote/002352.CN.md) - [06936.HK](https://longbridge.com/en/quote/06936.HK.md)