As foundation AI models become commoditized, the Shanghai-based company is betting structured knowledge will become the next source of competitive advantage.
By Doug Young
As artificial intelligence shifts from general-purpose AI models toward specialized professional applications, Shanghai Able Digital Science & Technology (2687.HK) is betting that the next competitive battleground will be structured knowledge rather than bigger models.
The Shanghai-based company, which has spent nearly two decades serving universities and research institutions, has positioned itself as an intermediary between large language models and specialized academic disciplines. Rather than building its own foundation model, it converts textbooks, research papers, laboratory processes and expert knowledge into knowledge graphs — machine-readable networks that map relationships between concepts and allow AI systems to retrieve specialized knowledge together with the underlying sources.
That strategy has become the company’s fastest-growing business. Revenue from knowledge graph products rose 68.5% in 2025 to 573.5 million yuan ($84.6 million), accounting for 59.2% of total revenue. Deliveries more than doubled to 10,386 projects, while gross margin improved to 65.5% from 61.9%.
Beyond foundation models
Chief Financial Officer Crystal Cao said the company views large language models as only one component of enterprise AI.
While foundation models have become increasingly capable at language generation and general reasoning, she said professional users require authoritative and continuously updated domain knowledge that reflects academic standards and institutional workflows.
Shanghai Able Digital’s Polymas platform combines third-party foundation models with the company’s own structured knowledge assets through its Meta Graph engine, allowing AI-generated answers to be linked back to academic sources and knowledge nodes. The approach is intended to improve transparency and reduce unsupported AI responses.
The company’s strategy is deliberately model-agnostic, allowing it to incorporate advances from multiple foundation-model providers rather than depending on a single platform.
Knowledge graphs have become central to that strategy. Rather than simply storing documents, the company converts academic materials into structured knowledge that can be continuously updated and accessed by AI systems. Cao described the transition as moving from “knowledge storage” to “knowledge computation,” allowing universities to update educational content far more quickly than traditional textbooks.
Partnerships and institutional demand
Shanghai Able Digital has also expanded partnerships with China’s major AI infrastructure providers, including Alibaba Cloud and Volcano Engine, a cloud computing and enterprise digital service platform owned by Chinese tech giant ByteDance.
According to Cao, the cloud companies contribute foundation models, computing infrastructure and AI ecosystems, while Shanghai Able Digital provides discipline-specific knowledge structures and experience working with universities and research institutions.
The company’s collaboration with Alibaba Cloud includes projects involving the Qwen large language model ecosystem, AgentScope multi-agent platforms, and physical AI research. Its partnership with Volcano Engine covers knowledge fine-tuning using ByteDance’s Doubao model, virtual-and-physical training systems and digital talent development.
Shanghai Able Digital is also discussing potential cooperation with other major foundation-model developers, Cao said.
The company increasingly works with what it calls lighthouse customers — universities that jointly develop AI applications in individual disciplines before the products are adapted for wider deployment across other institutions.
Universities seek trustworthy AI
Cao said customer priorities have evolved rapidly as AI adoption has accelerated across higher education.
Universities are now less concerned with whether to adopt AI than with integrating it responsibly into teaching and research. Institutions increasingly want systems that provide trustworthy outputs, incorporate discipline-specific knowledge, integrate with existing platforms and offer clear governance and security.
Student expectations are also changing, she said, with learners increasingly expecting personalized and interactive AI systems capable of explaining concepts in multiple ways, simulating experiments and identifying the academic sources behind their answers.
The company sees similar opportunities in scientific research, where AI can assist researchers with literature reviews, hypothesis generation, experiment design, simulation, data analysis and validation.
One example is its work on a mechanics model developed under China’s national 101 Plan, an initiative to modernize university teaching in strategic disciplines, led in that field by Zhejiang University. The discipline-specific model incorporates more than 2,000 knowledge points and over 4,000 relationships linking concepts, equations, simulations and research tools into a unified knowledge system that can be updated as new findings emerge.
Long-term growth
National priorities under China’s 15th Five-Year Plan (2026-2030) are creating sustained demand for AI-enabled knowledge infrastructure and new opportunities for the company, Cao said.
Higher education, scientific research and talent development are becoming more closely integrated with national innovation and industrial upgrading. New technologies require new disciplines, updated curricula and more complex practical training. At the same time, faster knowledge creation is increasing the operational burden on academic institutions.
But the opportunities are not simply a result of supportive policy, she said. Institutions must absorb new research, update teaching content, redesign practical training and prepare teachers and students for emerging technologies much faster. Traditional, manually maintained content systems cannot keep pace.
Shanghai Able Digital’s products address these challenges by making knowledge structured, traceable, updatable and callable. The Meta Graph helps institutions understand and govern their knowledge, while the Harness engine applies it across teaching, research, experiments and talent development, she said.
Demand for the company’s products currently exceeds its delivery capacity at some institutions, with implementation schedules already extending into 2027 and, in some cases, 2028, providing visibility into future project revenue.
The Insight Asia offers a wide-ranging mix of coverage of the Chinese technology sector and emerging industries, along with Chinese companies operating in those industries, including some sponsored content. For additional queries, including questions on individual articles, please contact us by clicking here.
