XUNCE Technology (03317), a leading provider of AI-powered real-time data infrastructure and analytics services in China, recently participated in a financial roadshow. During the event, the company's Investor Relations Director, Zhang Kailin, engaged in in-depth discussions with investors regarding the company's business model, product evolution, cross-industry diversification, and global strategy.

XUNCE Technology, established in 2016, initially focused on serving the highly regulated and demanding financial asset management sector. Over the past decade, the company has developed comprehensive end-to-end capabilities spanning data acquisition, cleansing, standardized governance, data modeling, and real-time stream computing. Following the explosion of large language models in 2023, XUNCE further integrated intelligent agent development, fully supporting the connection and fine-tuning of these models.

Through continuous collaboration and exploration, the company has successfully diversified beyond finance. By the end of 2025, only 20% of its revenue came from the financial sector, with the remaining nearly 80% derived from diverse fields such as power, advanced manufacturing, and urban operations. XUNCE has also entered cutting-edge sectors like embodied intelligence and the low-altitude economy. The company has accumulated deep expertise across more than 300 functional modules for complex application scenarios in various industries, enabling rapid replication and expansion into new markets.

The company's high product stickiness has resulted in excellent customer retention. By the end of 2025, XUNCE served over 230 enterprise clients, with the average revenue per user (ARPU) experiencing explosive growth, doubling by 105.51% from RMB 2.72 million in 2024 to RMB 5.59 million in 2025.

Within the AI ecosystem, XUNCE positions itself as an enabler and partner rather than a competitor to GPU manufacturers, large model developers, and cloud providers. It processes, refines, and analyzes data on top of computational infrastructure, allows clients to freely choose any large model to process its specialized tokens, and deploys deeply within clients' local, public, or hybrid cloud environments to build secure and accurate data hubs.

**Business Model Evolution: Token-Based Pricing Unlocks New Growth**

Prior to 2026, XUNCE's revenue primarily came from subscription fees and project-based transactions, with the latter accounting for 91% of revenue in 2025. Starting in 2026, the company introduced a disruptive new model: pay-per-Token.

It is crucial to distinguish XUNCE's "Scenario-Specific Vertical Tokens" from the generic tokens used by large model providers. If generic tokens are likened to "water"—a basic utility for entering the AI world—XUNCE's tokens are more akin to "fuel" or "oil." They are specifically processed, standardized, and infused with private enterprise data, local business rules, and industry know-how. This high-quality, pre-processed input is what enables large models to function effectively in complex, real-world business scenarios.

Pricing for these Scenario-Specific Vertical Tokens ranges from $10 to $100 per million tokens, significantly higher than the cost of generic tokens. This premium is justified because enterprises cannot directly feed messy, unstandardized raw data into large models due to poor efficiency, high costs, and a lack of domain-specific context. XUNCE acts as the essential "connector," bridging this gap.

**TokenOS: The Operating System for Data Tokenization**

To operationalize data tokenization, XUNCE launched its core system: the TokenOS operating system. This AI-native, enterprise-grade platform covers nine complete stages of data tokenization across five application layers:

1. **Foundation Layer:** Focuses on data standardization, evolving into a standardized "Token Factory" for rapid, replicable data tokenization across different enterprises. 2. **Middleware Layer:** The core stage for injecting industry know-how, involving deep data modeling and real-time stream computing. Here, enterprise data and business logic are encapsulated into measurable, priced, and exchangeable Scenario-Specific Vertical Tokens. 3. **Application Layer:** Allows intelligent agents (Agents) to call upon vertical tokens. The system is highly flexible, supporting connections to multiple models via MCP, freeing clients from vendor lock-in. 4. **Frontier Layer:** TokenOS is compatible with diverse computational backends (GPUs, CPUs), enabling flexible connection and adjustment regardless of the underlying hardware or chosen large model. 5. **Metering Layer:** Provides clients with a clear system interface to monitor token usage by business process, enabling complete traceability, metering, and compliance auditing.

The core of TokenOS is enabling enterprise data "Tokenization." Since the establishment of China's National Data Administration and the push for data asset recognition, a major industry pain point has been "how to price" and "how to establish rights" for data. XUNCE believes that tokenizing and standardizing private enterprise data to make it measurable, priced, and exchangeable is key to the large-scale, industrial deployment of AI.

**Financial Performance: Doubling Revenue and Path to Profitability**

XUNCE has maintained rapid growth. In 2025, revenue reached RMB 1.285 billion, doubling from 2024. Revenue from diversified (non-financial) industries already constituted 80% of the total and is expected to increase further.

Regarding revenue models, the structure is evolving: subscription fees are expected to remain stable at around 10% of revenue. While the absolute value of project-based revenue will continue to grow, its proportion is projected to decrease to 60-70%. The new pay-per-Token model is anticipated to contribute 20-30% of total revenue in 2026. Notably, the Annual Recurring Revenue (ARR) from token calls in April alone showed a 300% quarter-over-quarter growth.

Gross margins vary by industry. In the initial 1-3 year deployment phase for a new sector, margins are typically 40-50%. After this period, margins converge upward by 10-15 percentage points annually until reaching a steady state of 70-80%. For example, margins in the mature asset management (finance) sector have stabilized around 70%. The company's overall consolidated gross margin remains around 60% as it continues to enter 1-2 new industries each year.

On the profitability front, XUNCE achieved breakeven on an adjusted net profit basis in the second half of 2025 and expects a full-year inflection point in 2026. Operating cash flow improved significantly in 2025 and is projected to turn positive, achieving a self-sustaining ecosystem by 2026 or 2027. The company held nearly RMB 1.1 billion in cash and equivalents at the end of 2025.

**Future Outlook**

Looking ahead, daily token call volumes in the Chinese market are experiencing exponential growth. XUNCE aims to capitalize on regulatory tailwinds like data asset recognition. While currently operating in about 10 industries, it sees significant room for expansion, benchmarking against international giants like Palantir, which serves 17 industries. The company also expects overseas revenue to reach 5-10% of the total in 2026 by following leading Chinese companies in their global expansion. Furthermore, XUNCE is exploring strategic partnerships with top GPU and large model providers to co-develop industry-specific models for vertical sectors, unleashing the full value of high-quality private data.

**Q&A Highlights**

* **On competition from falling generic token prices:** The company clarified that its Scenario-Specific Vertical Tokens and generic model tokens are fundamentally different products with opposite price trajectories. As generic tokens become cheaper utilities ("water"), the value of high-quality, pre-processed data tokens ("fuel") increases. Enterprises are willing to pay a premium for tokens that contain their private data, business rules, and domain expertise, which are essential for effective AI deployment in complex scenarios. XUNCE's token pricing has been trending upward since the model's introduction. * **On competitive landscape:** XUNCE stated that while many vendors focus on specific segments of the data infrastructure chain, it sees few direct competitors offering a full-stack, end-to-end service from raw data processing to model tuning and agent delivery. Its primary global benchmark is Palantir, but XUNCE differentiates itself with the launch of its TokenOS operating system and the proven pay-per-Token business model in 2026, offering faster, more accurate multi-model connectivity and complete commercial metering. * **On expansion via M&A:** Management remains open but cautious regarding mergers and acquisitions, prioritizing organic growth. Potential acquisition targets would focus on technology/product synergy, industry know-how to accelerate new market entry, or assets aiding overseas expansion in regions like Southeast Asia and the Middle East. * **On transitioning clients to pay-per-Token:** The company reported strong client receptiveness to the new model. The "use-more, pay-more" logic allows clients to align costs directly with value and usage, reducing upfront commitment. Confidence in the product's ability to drive client revenue growth or improve risk control is key. The rapid 300% growth in token-based ARR and the projected 20-30% revenue contribution from this model in 2026 demonstrate high market acceptance and confidence in XUNCE's data tokenization capabilities.