When large model capabilities keep advancing and calling costs keep falling, what exactly do businesses truly want to pay for over the long run? Is it simply another model API, or rather a production-grade system that understands their own operations, manages tool access, enforces permissions, and takes accountability for results? This question has become unavoidable as enterprise AI moves from concept to actual deployment, and the interim report from 海致科技集团 (ASX: 02706) offers a telling case study for the industry.
Examining the first half of 2026, the company reported revenue of RMB 295.56 million, representing a 70.4% increase year-on-year, with gross profit amounting to RMB 132.08 million, up 97.8% against the prior corresponding period. The overall gross margin improved to 44.7%, a rise of 6.2 percentage points year-on-year. More notably, revenue from its Atlas intelligent agent arm hit RMB 114.59 million, surging 135.6% and contributing 38.8% of total revenue, while the intelligent agent gross margin stood at 52.9%, comfortably above corporate averages. Net losses narrowed to RMB 78.97 million, improving by 38.1% year-on-year, and adjusted net losses came in at RMB 19.60 million, a 31.5% improvement. Based on segment disclosures, intelligent agents accounted for roughly 54% of total incremental revenue. In other words, this report signals not just project expansion, but a concurrent shift in enterprise AI demand, product architecture, and operational quality. 海致科技集团's value should therefore be assessed within the broader industrial chain spanning foundation models, enterprise data and semantics, execution governance, and industry outcomes.
The procurement bar is moving from “can answer” to “can execute”
In the past, businesses evaluating AI products tended to focus first on model parameters, response accuracy, and demonstration effects. Yet in high-compliance environments such as government services, finance, and energy, the real hurdle is not generating text, but enabling systems to understand a company's own objects, processes, and permissions, and then completing traceable tasks within those boundaries. This means the evaluation criteria for enterprise AI are shifting from "what models can do" to "what enterprises can reliably accomplish": can data be computed consistently, can business logic be accurately captured, can actions be safely orchestrated, and can outcomes be audited and reviewed. AI is evolving from a point-solution Q&A tool into a system embedded in everyday corporate operations.
Guosheng Securities noted in a recent research report that the long-term opportunity for industrial AI can be understood through the lens of restructuring enterprise knowledge-driven labor costs. Around the same period, Huachuang Securities published a deep-dive report describing Harness as the control layer connecting models with enterprise data, rules, and process actions. Common to both institutional views is the recognition that the operational layer above the model is becoming the critical enabler for capturing enterprise AI value. The changing business structure visible in 海致科技集团's interim results offers quantitative evidence of this trend. While graph solution revenue grew 45.0% year-on-year, reinforcing the core franchise, intelligent agent revenue expanded at 135.6%, establishing a more dynamic incremental driver. The next question then emerges: once model capabilities become broadly commoditised, who can organise proprietary knowledge and business workflows into durable execution capabilities that companies keep using?
Beyond “computable” to “actionable”: the emerging control layer outside the model
The technology roadmap disclosed by 海致科技集团 can be summarised as a three-layer progression: multi-modal data, business ontology, and an intelligent agent runtime. The first layer constitutes a multi-modal data foundation. Different data types, including graphs, time-series, and vector data, are organised and computed around the same business objects, reducing the cost of repeated movement and integration between different systems. The second layer is the Ontology, which organises enterprise objects, relationships, states, rules, permissions, and actions into a machine-interpretable business semantic framework. A company is no longer merely "holding data", but becomes describable as a running system. The third layer is Atlas Harness, responsible for context construction, task state management, tool and skill invocation, workflow orchestration, chain tracing, replay, and outcome evaluation, enabling intelligent agents to progress from one-time responses to observable, diagnosable, ongoing operation.
The industry significance of this approach lies in separating long-term enterprise AI value from any single model. Models can be updated and replaced, while accumulated data, terminology, business rules, permission hierarchies, skill libraries, and execution trails can be preserved within the semantic and runtime layers independent of the model. This forms what could be called "enterprise AI assets"鈥攁 set of accumulative, transferable, and reusable business capabilities, not an accounting-level item. Atlas intelligent agents are already compatible with more than 100 large language models. Public disclosures highlight partnerships with companies including Zhipu AI, Hygon Information Technology, Bank of Ningbo, and JD Technology, alongside upgrades to academic workstations focusing on the integration of knowledge graphs with large models, and participation in setting relevant national standards, all underscoring broader ecosystem connectivity and engineering depth. From an industry-chain perspective, 海致科技集团 is not primarily engaged in the general-model parameter race, but rather fills the layer of data governance, business semantics, and execution control between computing power, models, and enterprise production systems. The next question is whether this technical foundation can translate into industry standing. Ultimately, this may come down to whether clients are willing to pay, upgrade, and reuse these capabilities over time.
From a single project to an entire industry: client upgrades convert know-how into growth assets
According to the interim disclosures, 海致科技集团 served 59 graph solution clients with an average contract value of roughly RMB 3.1 million, and 29 intelligent agent clients with an average contract value of approximately RMB 4.0 million. Among clients generating intelligent agent revenue, 69% had previously deployed graph solutions. This ratio should not be taken as a blanket conversion rate for all graph clients, but it clearly outlines an observable client upgrade pathway: starting with data and business semantic foundations, then expanding toward higher-value intelligent agents and broader business processes. If graph projects address "what an enterprise has and how it is connected", intelligent agents tackle "what actions to take under what conditions". The linkage between the two allows a single project delivery to accumulate into ontology models, skill libraries, and scenario components, which then feed additional purchases from existing clients and replication into new industries. With over 430 industrial clients served historically, this "enter first, expand next" path rests on a solid client base.
Beyond its core government, finance, and energy markets, public disclosures show 海致科技集团 has extended its capability set to smart mining, oil and gas, telecommunications, manufacturing, and pharmaceutical scenarios, while establishing a Hong Kong entity and a dedicated team. Cross-industry replication does not mean transplanting the same solution into new contexts, but rather continuously "compiling" industry knowledge gained from on-site delivery into reusable ontologies, skills, and process components. As component reuse, delivery efficiency, and client upgrades improve, project-based revenue should gradually display platform-like characteristics. The operational figures in the interim report offer supporting evidence: research and development expenses rose 81.5%, while sales and marketing costs grew only 15.1%, well below revenue growth, indicating that while the company deepens base-layer investment, operational efficiency is beginning to show clear improvement. As of the period end, cash and cash equivalents combined with financial assets measured at fair value through profit or loss totalled approximately RMB 972 million, with a debt-to-asset ratio of 26.3% and no external borrowings, providing ample financial buffer for continued R&D and industry delivery.
Conclusion: the next phase of enterprise AI competition is turning experience into infrastructure
Based on the interim results, 海致科技集团 is building a chain that runs from graph foundation, through intelligent agent applications, client upgrades, component accumulation, and cross-industry replication. Its industry position need not be defined by "number one" claims or a single analogy, but can be assessed across four dimensions: depth in high-compliance scenarios, cross-industry migration ability, ecosystem synergy, and standardised replication capability. The real dividing line in enterprise AI may not be who owns more models, but who can transform enterprise data, business semantics, permission rules, and execution experience into continuously reusable productivity.
For 海致科技集团, the 2026 interim report reads as a validation point. Its path from project deliverer toward infrastructure provider connecting models with industrial production systems is becoming progressively clearer. In addition, on 28 August, the Hong Kong Stock Exchange announced the first quarterly adjustment after optimising the technology 100 index methodology, adding 海致科技集团, along with WeRide, Pony AI, and Insilico Medicine, as constituent stocks, further highlighting its representation in the AI technology sector. For the market, this provides a fresh reference point for observing enterprise AI samples among Hong Kong-listed technology companies; for the company itself, it is a visible signal that its technology credentials, growth performance, and industrialisation progress are gaining external recognition.