The mid-year results update posted to the Hong Kong Stock Exchange on the afternoon of August 26 finally answered a question that had lingered in the market for four and a half years.

SenseTime Group Inc (HKG: 00020) reported an IFRS net profit of RMB 617 million for the first half of 2026, marking its first consolidated profit under IFRS standards since its listing in December 2021. While the figure itself is striking, what truly captures the attention of global investors is the subtle yet profound shift in the company's valuation anchor — a transition from being prized for technological leadership to being valued for commercial execution. SenseTime is rewriting the pricing logic for AI stocks on the Hong Kong exchange.

Breaking Down the RMB 617 Million Profit: What It's Really Made Of

Profit is an accounting construct, but recurring revenue and gross margins reveal the true health of a business. To properly interpret SenseTime's 2026 interim report, one must first dissect how the RMB 617 million profit was generated. On one hand, there has been substantial improvement in the core business. Revenue for the first half reached RMB 2.911 billion, up 23.4% year-over-year; gross profit climbed 32.9% to RMB 1.206 billion, with gross margin expanding by 2.9 percentage points to 41.4%. Revenue growth outpacing cost growth indicates that economies of scale are beginning to take effect. Concurrently, R&D expenses fell 17.1% to RMB 1.756 billion, while administrative expenses decreased 13.2% to RMB 518 million. This combination of rising revenue and controlled costs has opened up profit margins from both directions.

On the other hand, the "1+X" strategy has borne fruit through the realization of fair value gains from ecosystem investments. Other net gains for the period totaled RMB 2.127 billion, including a RMB 521 million gain from the disposal of subsidiaries and RMB 1.249 billion in unrealized gains from financial assets measured at fair value. This income stream represents the capitalization of SenseTime's forward-looking investments in ecosystem companies such as Daxia Robotics and SenseTime Medical. Therefore, the company's book profit stems from two distinct sources: a sharp year-over-year reduction in core operating losses, and gains from fair value changes in its AI ecosystem investments. The former measures the slope of operational improvement, while the latter reflects the phased financial release from value appreciation in portfolio companies — a realization of years of strategic technology and industry positioning. Conflating these two would obscure the true picture; examining them separately offers clarity: a 67.3% reduction in core operating losses, a 2.9 percentage point improvement in gross margin to 41.4%, and a 124% surge in recurring revenue. These three indicators improving simultaneously point to a business model in transition and a rising quality of earnings.

The Business Model Shift: Recurring Revenue Disclosed for the First Time

SenseTime has disclosed its recurring revenue (RR) for the first time. In the first half of 2026, RR reached RMB 1.145 billion, up 124.4% from RMB 510 million in the same period of 2025, with its share of group revenue jumping from 21.6% to 39.3%. RR refers to revenue derived from contracts effective during the reporting period that carry a continuous renewal attribute. Its rapid ascent signifies that SenseTime's client engagements are evolving from one-off project deliveries to ongoing service relationships — a transition from "selling systems" to "selling subscriptions." For an AI company still in its investment phase, seeing RR's share of revenue jump from roughly one-fifth to nearly two-fifths within six months is arguably more telling than the profit turnaround itself.

The revenue mix is also undergoing a transformation. Generative AI revenue reached RMB 2.327 billion, up 28.2% year-over-year, now accounting for nearly 80% of group revenue and serving as the primary growth engine. Visual AI revenue came in at RMB 497 million, a 13.9% increase, recovering growth after operational strategy optimization, with existing customers contributing 67% of this segment's revenue and continuing to serve as the gateway for industry penetration and overseas expansion. International business revenue grew 127.0% year-over-year, far outpacing the group's overall 23.4% growth rate, driven by mature visual AI capabilities, unified multimodal models, and localized delivery systems. The full picture of this transition can be summarized by a few key figures: gross margin rising from 38.5% to 41.4%, R&D spending down 17.1% to RMB 1.756 billion, administrative costs down 13.2%, all while revenue continues to grow at 23.4%. Revenue growth outpacing cost growth signals the release of scale effects.

Why These Three Metrics Can Improve Simultaneously

The simultaneous improvement of these three indicators corresponds to SenseTime's system-level capabilities — "one unified model, one Token factory, and one agent orchestration system" — beginning to operate cohesively. These three components are interlocking. Consider the model layer first. In March 2026, the company proposed the NEO-Unify architecture; in April, it launched SenseNova U1, natively unifying multimodal understanding and generation; in July, it unveiled SenseNova U1 Pro at the World Artificial Intelligence Conference, benchmarked against GPT-Image 2; and in August, it open-sourced SenseNova U1.5 Lite, a lightweight model supporting native 4K output. For long-horizon agents, SenseNova 6.7 Flash Lite, released in May, offers tool orchestration and long-term memory capabilities, reducing token consumption by 60% in scenarios such as information retrieval compared to text-only agents. Native multimodality determines what an agent can understand and create, while long-horizon capabilities determine whether it can sustain action and accomplish goals. The fusion of these two tracks extends AI from isolated capabilities to a full chain covering understanding, planning, execution, verification, and delivery.

Next, the Token factory. As of July 2026, daily token service volume exceeded 2.4 trillion, an approximate 22-fold increase year-over-year. The AI data center (AIDC) has evolved from self-use computing power into a production-grade Token factory, now serving four external foundational model vendors. Agent model context training speed has doubled, and mainstream domestic chip MFU can reach up to 2.5 times the chip manufacturer's baseline performance. The computing-power-electricity coordination agent achieves 96% load forecasting accuracy, saving over RMB 12 million in electricity costs in the first half, while reducing carbon emissions by 24,000 tons per 10,000 P of computing power annually — continuously driving down unit intelligence costs. Finally, the agent orchestration system: in the first half, it served over 1,000 enterprise clients, adding more than 100 new ones across more than 20 industries. The Xiaohuanxiong agent now serves Lenovo, Ping An Technology, all three major telecommunications operators, JD.com, and other leading enterprises. Seko's daily video production reached 10,000 minutes, with cumulative estimated views surpassing 1.5 billion. The personal assistant Kapi series has accumulated over 45 million users. Front-end products are diverse, but the back-end capabilities are unified, allowing the same task-delivery framework to be rapidly reused across different industries.

The logic is straightforward: customers and tasks continuously generate data and feedback, driving iterations in models, workflows, and products. Improvements in model capabilities and Token production efficiency lower task-delivery costs and expand application boundaries. As value measurement gradually shifts from Token consumption to task completion, the positive feedback loop among technological capability, customer value, recurring revenue, and operational efficiency begins to turn.

Valuation Anchor Migration: From Technological Leadership to Market Leadership

The significance of the profit announcement lies not in the RMB 617 million figure itself, but in what it confirms: a migration of the valuation anchor. SenseTime is transitioning from being a model company that sells APIs and computing power to a service-oriented company that delivers trustworthy multimodal agent capabilities. The valuation anchor is shifting from model capability to commercial monetization ability — this is what the market should truly be re-rating. Brokerages are already adjusting their pricing. Goldman Sachs issued a research note on August 17 following the profit alert, raising its target price for SenseTime from HKD 2.00 to HKD 2.05 while maintaining a neutral rating, but significantly revising its earnings forecasts: the 2026 net loss expectation was narrowed from RMB 1.25 billion to RMB 59 million, and the 2027 loss from RMB 357 million to RMB 336 million, implying SenseTime is nearing its breakeven point. Profit forecasts for 2028 through 2032 were raised by 13%, 5%, 2%, 2%, and 1%, respectively. Goldman believes SenseTime's AI models are moving toward "Delegated Intelligence," where users will focus on the delivery of AI-generated outcomes, facilitating large-scale commercialization. Across the broader sell-side consensus, target prices over the past 12 months range from HKD 2.20 to HKD 2.72, with a mean of approximately HKD 2.43 — representing significant upside from the current price of HKD 1.47. Support is also accumulating on the capital flow and index front. SenseTime has been included in core indices such as MSCI China and Hang Seng Tech, combining passive allocation demand with active re-rating logic. And when an AI company proves it can generate positive operating profit, the natural valuation anchor shifts from price-to-sales to EV/EBITDA and P/E ratios — SenseTime stands precisely at the threshold of this transition.

A New Chapter Begins

The first profit is the starting line for a valuation anchor migration. An AI company simultaneously achieving a 67% reduction in core losses, gross margin breaking above 41%, and recurring revenue doubling year-over-year in the first half sends a signal that is no longer about "whether it can make money," but rather "how it makes money is changing": from one-off project delivery to continuous service, from selling tokens to delivering agents that can actually get work done. The re-rating of SenseTime has only just begun.