On August 24, XUNCE (03317.HK) held its "2026 Interim Results Investor and Expert Exchange Session" at its Shenzhen headquarters. The event brought together authoritative experts from academia and industry associations, institutional investors, and mainstream media, featuring showroom tours, performance presentations, expert panel discussions, and live Q&A sessions to comprehensively showcase the company's commercial progress in AI real-time data infrastructure and the Token business model, while jointly exploring development trends in the AI To B sector and cutting-edge practices in data factor marketization.

Currently, as enterprise AI deployment demand accelerates and data factor marketization progresses rapidly, XUNCE is seizing the moment to fast-track its Token business model, building an integrated AI To B platform that delivers one-stop AI infrastructure from data governance to model deployment. In the first half of 2026, the company not only validated its Token business model—with Token revenue share surging and ARR growing robustly—but also made solid strides in product matrix expansion, diversified industry penetration, and international layout, comprehensively elevating operational quality and underscoring its "real AI" value. On the day, the company's CTO Yang Yang, Chief Scientist Tang Bo, and Board Secretary Chen Wanxie attended the session, joined by five experts from leading institutions in data factors and AI, including Wang Guan, Deputy Director of the Guangdong-Hong Kong-Macao Greater Bay Area Big Data Research Institute; He Bin, Deputy Director of the National Key Laboratory of Autonomous Intelligent Unmanned Systems and Director of the Shanghai Research Institute for Autonomous Intelligent Unmanned Systems; Pan Fei, Deputy Director of the Data Elements Committee under the China Information Industry Association; Lin Zhenyang, Researcher at Beijing Zhongguancun College and Zhongguancun AI Research Institute; and Professor Chen Bo, Director of the AI and Digital Finance Research Center at Central University of Finance and Economics.

Mid-Year Revenue and Profit Hit New Highs, Token Revenue Share Jumps Significantly

At the start of the session, Zhang Kailin, Director of Investor Relations at XUNCE, shared the company's operational results and strategic direction for the first half of 2026. According to the interim results report, the company posted revenue of RMB 967 million, up 389% year-on-year and a record for the period; profit approached RMB 100 million, marking its first-ever profitable first half; gross margin held steady at a high 60.1%; ARPU surged 240%; per-capita revenue generation jumped 379%; and customer retention stayed above 90%. All metrics improved across the board, once again validating XUNCE's deep value and growth potential in the AI infrastructure track.

The company attributed the strong performance to multiple converging drivers: accelerated deployment of enterprise-grade AI real-time data infrastructure, deeper penetration across diverse industries, the rollout of the Token business model, international business expansion, and ecosystem building. Meanwhile, around enterprise AI deployment, the company has formed a full-chain product matrix spanning compute, data, Token, models, and applications.

Full-Chain Product Matrix: "Compute-Data-Token-Model-Application"

Looking ahead, the company has mapped out a four-stage strategic evolution path for AI data infrastructure: starting from data governance (1.0), advancing through Tokenization (2.0), moving toward a global Token exchange platform (3.0), and accelerating enterprise small-model (4.0) training and scaled deployment. The strategy centers on five key pillars—evolving the business model, accelerating cross-industry replication, pioneering cutting-edge applications, building strategic partnerships, and steadily expanding overseas—all aimed at continuously creating value for customers and opening a new cycle of exponential growth.

Experts Discuss AI To B Applications and Data Factor Marketization

During the expert panel, He Bin, Pan Fei, Lin Zhenyang, and Chen Bo engaged in an in-depth dialogue covering digital economy development trends, enterprise data governance pain points, the data Tokenization business model, industry intelligence upgrade directions, data security compliance, and globalization opportunities. In discussing digital economy trends, experts noted that enterprise digital transformation still faces core pain points such as data governance and value realization, underscoring the urgent need for high-quality, scenario-specific data infrastructure. He Bin emphasized that the fundamental driver of enterprise digital intelligence lies in connecting the entire data chain—front-end data sampling, mid-stream processing, and back-end large-model training. With the rise of large models and AI agents like OpenClaw, enterprise demand for data is climbing exponentially. Token, as the smallest unit for large-model training and inference, is becoming a fundamental element of the intelligent era, and Token economics will play a pivotal role in enterprise digitalization.

Professor Chen Bo, from a value-realization perspective, pointed out that AI provides an objective yardstick for digitalization efforts that were previously difficult to quantify. Token plays a critical pricing role—both application scenarios and data value are directly reflected in Token consumption. He argued that AI not only fuels the intelligence wave but also dramatically lowers the barrier to algorithm adoption, enabling deep business-technology integration and pushing the digital economy into a new phase with unprecedented market capacity and scenario expansion.

On opportunities and challenges in converging enterprise data assets with AI, Pan Fei noted that since data was designated the fifth production factor, national and local policies have been rolled out intensively, building a "top-down" institutional and organizational framework that delivers sustained policy dividends. Meanwhile, authorities such as the SASAC are pushing central state-owned enterprises to cultivate future industries through scenario-driven approaches, and local governments are publishing AI and data scenario demand lists, creating abundant implementation opportunities for AI and intelligence service providers. However, challenges remain: over the past few years, nearly every link in the data value chain—from infrastructure and institutions to application—has faced bottlenecks. The rise of Token-based service models now offers a breakthrough path. Token-based pricing mechanisms could directly quantify and price data value services, driven by market demand where payment initiative stems from business outcomes. Professor Chen Bo added that data governance is shifting from a "public governance" phase to a "value governance" phase. Governance itself is becoming a business model—in the AI era, data circulation, pricing, auditing, and application form a closed loop, and Token consumption is not just a billing metric but a test of whether the scenario loop actually works. Human-Token interaction has a multiplier effect, but it must ultimately anchor on effective application loops; otherwise, even the best models cannot sustain a viable commercial logic.

On the evolution of data from resource to asset and future business models, Pan Fei cautioned that after the Ministry of Finance pushed data asset recognition policies, the market fell into the misconception that "all data has value." The prerequisite for enterprise data assetization is achieving business value first, then assetization. Long-term value assessment must be grounded in real business enablement and incremental contributions from circulation. The core essence of this round of data factorization lies in "circulation" and "integration." Enterprises should prioritize cross-domain collaboration between internal data and external ecosystems and customer data, avoiding premature "sealing" of data that hinders value flow. Director Lin Zhenyang elaborated on the "dual circulation" theory of data assetization: the internal loop involves enterprises converting data assets formed through business informatization and digitalization into balance-sheet entries after compliant rights confirmation; the external loop emphasizes that data assets must enter the market and trade with other entities to leap from internal cost reduction to society-wide value creation. AI accelerates this process and brings a fundamental shift—data was once for human consumption, but now increasingly for AI consumption. Token, as the smallest measurable, priceable, and outcome-based payment unit, is reshaping business models. The next phase could be a new era of Tokenized AI data, where AI-to-AI architectures spawn one-person companies and AI-native enterprises, with flywheel effects spinning ever faster.

Regarding key industries for future AI infrastructure and intelligence upgrades, He Bin pointed out that Tokenization is the critical step in converting enterprise data into assets callable by large models. Currently, information-intensive industries are progressing faster, while traditional sectors like energy, mining, and manufacturing lag due to weak data foundations. But with the development of physical AI, demand for data-driven and model training will surge across industries, and traditional sectors will catch up. Director Lin Zhenyang highlighted two promising directions: first, the scientific field, where AI is driving research into a data-driven fourth paradigm—international large-scale investment around AI-accelerated scientific exploration will generate huge demand for data processing and intelligent algorithm tools; second, the broader health field, such as emotion, psychology, and brain science, where large language models build collective knowledge bases but still leave vast room in individual mind-body connections and cognitive blind spots. Breakthroughs in these areas will trigger large-scale industrialization, ultimately returning AI to serving humanity itself. In industry expansion, experts noted that beyond finance, manufacturing, power, and new energy—where demand is already clear—emerging sectors such as intelligent vehicles, low-altitude economy, biomedicine, and commercial aerospace are becoming major markets for data intelligence upgrades. On globalization, experts observed that overseas markets impose stricter data sovereignty and security compliance requirements, but China's high-quality scenario Tokens have export capability—XUNCE's TokenOS deployment in the European market, known for stringent data sovereignty demands, serves as proof. Across the panel, all four experts converged on a shared consensus: data is becoming the most critical production factor in the AI era, and high-quality, scenario-specific data infrastructure will be the core foundation for enterprises to unlock AI value and build competitive moats.

Full-Chain Product Layout Opens New Growth Cycle for AI To B

At the session's close, management engaged in in-depth Q&A with attendees, detailing the real logic and breakthrough approaches behind XUNCE's AI To B implementation. Responding to whether closed-source model capabilities have already exceeded enterprise needs, Yang Yang offered a nuanced view: it cannot be simplified into a binary. In everyday general scenarios—like meeting minutes and report drafting—large model capabilities are indeed more than sufficient; but in core production processes such as industrial quality inspection and professional decision-making, relying solely on large models falls far short. Tang Bo, from a foundational perspective, argued that the core issue lies not in the models themselves but in how data is organized. He noted that traditional informatization follows a "building blocks" approach—each new application requires redoing data aggregation, governance, and training, which is costly and unsustainable. The correct path for AI-native applications is to build AI data infrastructure, consolidating multi-dimensional enterprise data on a single platform, using ontology and semantic modeling so data can be repeatedly invoked and continuously mined for value. The primary enterprise goal shifts from developing isolated software to continuously growing new agents and business applications atop the data infrastructure.

On willingness to pay, Yang Yang offered an insightful observation: most customers cannot distinguish how much of a solution comes from models, processes, or data—they only care whether problems are solved. While everyone focuses on the model hype, in specialized scenarios, customer willingness and ability to pay are tied to specific business value and outcomes. This means selling model capabilities alone is not viable. True To B professional services involve hardware, software, hardware-software integration, and scenario-specific data curation, optimization training, and application development. XUNCE's approach is always scenario-driven, designing and integrating all relevant resources, ultimately delivering based on results. Addressing the specialization trend in industry agent deployment and talent bottlenecks, Yang Yang highlighted an often-overlooked fact: traditional software companies pursue standardization and educate users to adapt to products, paying little attention to specialization; but professional enterprise service scenarios are inherently segmented and require deep cultivation. XUNCE's expansion path is to leverage data processing concepts and service models honed in finance, seeking transferable "soil." Entering each new industry—such as healthcare or manufacturing—requires partnering with deep-domain players to jointly extract tacit knowledge and convert it into data, assets, and models.

On the role and bottleneck of FDEs (Front-end Deployment Engineers), Yang Yang stated that FDEs are the "flashpoint" of all bottlenecks, and precisely for that reason, they are the core hub of value creation. XUNCE's R&D teams are being cultivated toward FDE roles, while actively seeking external partnerships in healthcare and other sectors. FDE is a key focus for expansion and deep cultivation, but solving all problems "still comes step by step." Regarding risks like model vendor price hikes, Tang Bo was clear: XUNCE's positioning is "the last mile for large models," combining model capabilities with enterprise business practices, making AI truly usable through private-data Tokenization, training, and inference. Yang Yang used a vivid analogy: "The relationship between large models and us is like components to a finished vehicle, or an engine to an airplane—no one can fly to the US sitting on an engine; you still have to buy a ticket." In other words, XUNCE delivers complete, deployable solutions, with large models serving as key components. Even if upstream model policies shift, XUNCE, as the "integrator" connecting model capabilities with enterprise needs, retains irreplaceable pricing power and customer stickiness. Enterprise AI deployment is no longer about "having models or not," but about "how models truly integrate into business flows and create quantifiable value." The real moat going forward lies not in technical parameters, but in deep cultivation of each niche scenario and accumulated trust. XUNCE is optimistic about the AI industry and direction, yet cautious in taking each step forward.