SenseTime (00020) 2026 Mid-term Report: Accelerated Gross Profit Growth, Shift in Business Model, and Improved Profit Quality

On the afternoon of August 26, the Hong Kong Stock Exchange updated its announcement board with a mid-term performance report that finally revealed the answer the market had been waiting for over four and a half years. SenseTime (00020) achieved an IFRS net profit of 617 million yuan in the first half of 2026. This marks the first time SenseTime has recorded consolidated profits under IFRS since its listing in December 2021. The figure itself is already astonishing. But what truly left global investors breathless is the valuation anchor shift curve that is subtly emerging behind this financial reportfrom "technological leadership" to "commercial realization," SenseTime is rewriting the pricing logic of the AI sector in the Hong Kong stock market. Profit Quality: First, Understand How the 617 Million Yuan Was Achieved Profit is merely an accounting figure; recurring revenue (RR) and gross margin are what define a business. To interpret SenseTime's 2026 mid-term report, we first need to understand how the 617 million yuan in profits was generated. On one hand, there has been a substantial improvement in core business operations. In the first half of the year, SenseTime achieved revenue of 2.911 billion yuan, a year-on-year increase of 23.4%; gross profit reached 1.206 billion yuan, up 32.9%; and the gross margin reached 41.4%, an increase of 2.9 percentage points year-on-year. The revenue growth outpaced the cost growth, and the scale effect is starting to be released. At the same time, R&D expenses decreased to 1.756 billion yuan, down 17.1% year-on-year; administrative expenses fell to 518 million yuan, down 13.2%. Between this rise and fall, profit margins were opened up in both directions. On the other hand, the fair value realization of ecological investments under the "1+X" strategy contributed. In the first half of the year, net other gains reached 2.127 billion yuan, primarily including gains of 521 million yuan from the sale of subsidiaries, and unrealized fair value gains from financial assets amounting to 1.249 billion yuan. This portion of income reflects the capitalized results of SenseTime's long-term foresighted investments in ecological enterprises like Siasun Robot & Automation and SenseTime Medical. Thus, SenseTime's reported profits come from two lines. One is the significantly reduced losses in core business operations, and the other is the fair value changes from investments in AI ecological enterprises. The former measures the slope of operational improvement, while the latter represents the phased monetary realization brought about by the rising value of ecological companies under the 1+X strategy, including gains from the sale of subsidiaries and fair value changes of financial assets, reflecting years of technological and industrial deployment. Confusing the two may lead to misjudging the quality, while analyzing them separately provides clarity: the core business saw a 67.3% reduction in losses, the gross margin improved by 2.9 percentage points to 41.4%, and RR surged by 124% year-on-year. These three indicators improving simultaneously indicate a shift in the business model and an enhancement in profit quality. Business Model: RR Disclosed for the First Time, Transitioning from Projects to Services For the first time, SenseTime disclosed its recurring revenue (RR). In the first half of 2026, RR was 1.145 billion yuan, a 124.4% increase from 510 million yuan in the first half of 2025, making up 39.3% of the group's total revenue, up from 21.6%. RR is defined as income derived from contracts in effect during the reporting period with the property of sustained renewals. Its rapid ascent signifies a shift in SenseTime's collaboration with clients from single project deliveries to continuous services, moving from "selling systems" to "selling subscriptions." For an AI company still in the investment phase, the RR share rising from 20% to nearly 40% within six months speaks volumes beyond just a profit coming back to the positive side. The revenue structure is also changing gears. Revenue from generative AI was 2.327 billion yuan, a year-on-year increase of 28.2%, accounting for nearly 80% of total revenue and becoming the main engine; revenue from visual AI was 497 million yuan, up 13.9% year-on-year, recovering growth after optimizing operational strategies, with returning customer contribution at 67%, continuing to serve as the entry point for the company into industries and overseas markets. Revenue from overseas operations grew by 127.0% year-on-year, far exceeding the group's overall growth rate of 23.4%, relying on mature visual AI capabilities, unified multimodal models, and localized delivery systems, thus extending its capabilities outward. The full transition can be summarized with a few figures: gross margin rose from 38.5% to 41.4%, R&D expenses fell by 17.1% to 1.756 billion yuan, administrative expenses decreased by 13.2%, all while revenue continued to grow at a rate of 23.4%. Revenue growth has outpaced cost increases, and the scale effect is beginning to be released. Why Can These Three Indicators Improve Simultaneously? The simultaneous improvement of these three indicators corresponds to SenseTime's system-level capabilities of "a single model, a Token factory, and an intelligent agent control system" successfully running together. These three components interact cohesively. First, look at the models. In March 2026, the NEO-Unify architecture was introduced, followed by the launch of the natively unified multimodal understanding and generation model, SenseNova U1 in April, and SenseNova U1 Pro, which competes with GPT-Image 2, was launched at the World AI Conference in July. In August, SenseNova U1.5 Lite was open-sourced, supporting native 4K output with lighter specifications. On the long-range intelligent agent front, the May release of SenseNova 6.7 Flash Lite boasts capabilities like tool orchestration and long-term memory, and its token consumption for scenarios like information search dropped by 60% compared to pure text-based agents. The native multimodality determines what an intelligent agent can understand and create, while the long-range intelligence dictates its ability to take sustained actions and achieve goals. The two pathways merged, allowing AI to expand from single-point capabilities to a comprehensive chain covering understanding, planning, execution, verification, and delivery. Next, consider the Token factory. In July 2026, the daily Token service volume exceeded 2.4 trillion, a year-on-year growth of approximately 22 times. The large device has upgraded from self-use computing power to an operational Token factory, providing services to four external foundational model vendors, doubling training speed for Agent model context, with mainstream domestic chips reaching up to 2.5 times the benchmark set by chip manufacturers. The computing-electricity collaboration with Agent achieved a 96% accuracy rate for load forecasting, saving over 12 million yuan in electricity costs in the first half of the year, and contributing to a reduction in carbon emissions of 24,000 tons annually per 10,000 petaFLOPS of computing power, continuously lowering the cost per unit of intelligence. Finally, look at the intelligent agent control system. In the first half of the year, services extended to over 1,000 corporate clients, with more than 100 new additions across over 20 industries; the Xiaohuanxiong has serviced major enterprises like Lenovo, Ping An Technology, the three major telecommunications companies, and JD.com; the Seko system produced 10,000 minutes of video per day with an estimated total view count exceeding 1.5 billion times; the cumulative user base of the personal assistant Kakipi series surpassed 45 million. The frontend product diversity and backend capability are unified, allowing the same task delivery capability to be rapidly applied across different industries. The logic is straightforward: customer and task data and feedback continuously drive iterations of models, workflows, and products; improvements in model capabilities and Token production efficiency lower task delivery costs and expand application boundaries. Value measurement is gradually shifting from Token consumption to Task delivery, creating a positive cycle among technical capability, customer value, recurring revenue, and operational efficiency. Valuation Anchor Shift: From Technological Leadership to Market Leadership The significance of the profit announcement lies not in the figure of 617 million yuan itself, but in the confirmation of a valuation anchor shift. SenseTime is transitioning from a model company that "sells APIs and computing power" to a service-oriented company that "delivers reliable multimodal intelligent agent capabilities." The valuation anchor is shifting from model capabilities to commercial monetization capabilities, which is what the market truly needs to reassess. Brokerages have begun adjusting their pricing. After Goldman Sachs' profit announcement on August 17, it updated its report, raising SenseTime's target price from 2.00 HKD to 2.05 HKD, maintaining a neutral rating, but significantly altering profit forecasts: the expected net loss for 2026 was narrowed from 1.25 billion yuan to 59 million yuan, and from 357 million yuan to 336 million yuan for 2027, indicating that SenseTime is nearing breakeven; profit forecasts for 2028 to 2032 were revised upward by 13%, 5%, 2%, 2%, and 1%, respectively. Goldman Sachs judges that SenseTime's AI model is moving toward "Delegated Intelligence," where users will focus on delivering AI output results, facilitating large-scale commercialization. Within a broader consensus among sell-side analysts, SenseTime's target price over the past 12 months ranged between 2.20 and 2.72 HKD, with an average of about 2.43 HKD, indicating significant upside potential compared to the current price of 1.47 HKD. Support from the financial side and index levels is also accumulating. SenseTime has been incorporated into key indices such as MSCI China and Hang Seng Tech, creating cumulative demand for passive allocation alongside active reassessment logic. When an AI company proves it can achieve positive operating profits, its valuation anchor will naturally shift from "price-to-sales" to "enterprise EV/EBITDA" and "price-to-earnings (P/E)"SenseTime is at this critical point of transition. Conclusion The first profit is the starting line for the shift of the valuation anchor. For an AI company to simultaneously achieve a 67% reduction in core losses, a gross margin surpassing 41%, and a year-on-year doubling in RR, the signal conveyed is no longer "can it make money," but rather "the way it earns money is changing": from one-time project delivery to ongoing services, from selling Tokens to delivering an intelligent agent that gets the job done. The revaluation of SenseTime has only just begun.