On the morning of August 3, shares of MiningLamp-W (02718) staged a strong rebound, surging over 20% at one point to a high of HK$55.9. By the close, the stock settled at HK$53.0, up 15.62%, giving the company a market capitalization of HK$8.103 billion. This rally came just days after the company's first major lock-up share expiration since its listing on July 31.
With the selling pressure largely cleared, market attention is now shifting back to a more fundamental question: as Agentic AI transitions from a technical concept to large-scale commercial deployment at an industry inflection point, can the technological closure and policy positioning that MiningLamp-W has already achieved support a re-evaluation of its medium- to long-term value? The rules of competition in enterprise services are being fundamentally rewritten—the value anchor is moving from model parameter scale to tangible business outcomes, and pricing power is shifting from feature benchmarking to profit-sharing based on results.
MiningLamp-W has made multi-dimensional preparations across technology, products, and business. It has open-sourced the on-device model Mano-P and the inference framework Cider, officially launched the AI-native hardware Octic, and introduced Octo, the world's first open-source, trusted agent collaboration network. This creates a complete loop covering perception, reasoning, execution, and collaboration. At the same time, the implementation of a filing system for on-device AI and the first inclusion of intelligent agents in the national industrial roadmap signal that the industry has entered a new phase of compliance and scaled development. The convergence of policy frameworks and technological first-mover advantages is transforming MiningLamp-W's growth narrative from a continued deepening of its data intelligence core to an accelerated unlocking of its second growth curve in agent services.
On July 29, the company announced a plan to acquire approximately 19.07% of Puyuan Software, incorporating scenarios like financial shared services and treasury management for large group enterprises into its business scope. This further expands the potential for deploying intelligent agents in the deep verticals of enterprise management.
Technology Positioning: A Full-Stack Software-Hardware Layout and Open-Source Strategy Securing a Core Position in the Agentic AI Infrastructure
Agentic AI is moving from technical validation to a critical inflection point for large-scale commercialization, with the industry's pace clearly accelerating. MiningLamp-W has built a technology chain covering on-device models, inference engines, agent collaboration platforms, and hardware entry points. The foundation of this technological depth lies in breakthroughs at the model layer. The company's open-source on-device GUI-VLA model, Mano-P, and its supporting inference framework, Cider, together form a closed-loop infrastructure for on-device AI, moving it from "runnable" to "mass-producible" and laying the deployment groundwork for the upper-layer collaboration ecosystem.
In June 2026, the company launched Octo, the world's first open-source, trusted agent collaboration network. Licensed under Apache 2.0 and other protocols, it supports private deployment. Octo is positioned as an interconnection infrastructure for agents. Currently, the network covers approximately 1,500 internal employees, supporting over 4,200 agents in actual business processes. Octo's most significant long-term advantage lies in its design: every human-machine collaboration is not just about completing a task, but also about structurally encoding organizational knowledge. Through an asset-accumulation flywheel of "assigning tasks → receiving feedback and verification → accumulating preferences and skills → increasing efficiency for the next iteration," Octo creates a unique positive feedback loop. This allows the organizational productivity infrastructure to naturally enrich itself with each collaborative interaction, enabling true capability accumulation and intelligent upgrades.
This design elevates AI from a mere "tool" to a "vessel for organizational memory." For enterprises, their organizational knowledge is systematically recorded, verified, and passed on. As the number of collaborations increases, the system's understanding of business context deepens, and switching costs rise, creating a compounding barrier. The launch of Octo signifies that MiningLamp-W has built a collaboration platform capable of orchestrating, coordinating, and amplifying the effectiveness of agent services. Through Octo, different agents can work together orderly within complex business flows, integrating fragmented AI capabilities into a coherent, controllable, and evolvable productivity loop. This supports the scaled deployment of agent services in enterprise scenarios.
Policy Catalysts: On-Device AI Filing and the Agent Industry Roadmap Accelerate Commercial Penetration and Build Compliance Barriers
The formal implementation of the on-device AI filing system and the first inclusion of intelligent agents in the national-level industrial roadmap mark the industry's transition from technical exploration to a phase of compliance and scaled development within an institutional framework. This policy window effectively creates institutional entry barriers for vendors with private deployment capabilities and full-stack proprietary technologies. The policy framework has taken shape rapidly. In July, on-device AI was officially brought under national standardized management, requiring models to be deployed with traceable, auditable, and controllable technical characteristics.
Earlier, the Ministry of Industry and Information Technology released the "Implementation Opinions on Innovative Development of 'AI + Information and Communication,'" which included "intelligent agents" in the national industrial roadmap for the first time. On July 23, Beijing issued its first special policy on agents, "Several Measures to Support the Leading Development of Intelligent Agents," systematically deploying across five dimensions: common infrastructure, native applications, terminal integration, security governance, and open-source ecosystems. MiningLamp-W's Private AI infrastructure, built on the core principles of "open-source, privately deployable, and white-box auditable," aligns with current policy requirements for data security and controllability. This provides a compliance advantage in enterprise procurement within data-sensitive industries.
Market data confirms the commercial conversion efficiency of this institutional dividend from the demand side. Frost & Sullivan predicts that the on-device AI market will grow from RMB 321.9 billion in 2025 to RMB 1.22 trillion in 2029, a compound annual growth rate (CAGR) of 40%. The enterprise-level AI agent market in China is expected to exceed RMB 332 billion by 2029, with a CAGR of 107% from 2024 to 2029. In this high-growth track, the policy framework sets the entry threshold, and the technology base determines the depth of penetration. By leveraging its private, compliant deployment base, MiningLamp-W can meet some of the requirements of enterprise customers in such procurement. This adaptability provides a certain first-mover advantage in project bidding for data-sensitive industries.
Growth Realization: The Compounding Effect of Data Assets and Collaboration Network Customer Stickiness, with Agent Services Catalyzing the Second Curve's Ascent
Enterprise AI competition has shifted from an arms race of model capabilities to a deeper contest of data asset compounding and ecosystem lock-in effects. MiningLamp-W's multi-modal data assets, accumulated over two decades, provide business context for its intelligent agents. The Octo collaboration network, through continuous human-machine interaction, accumulates organizational experience, creating a barrier that "thickens with use." Building on this, the DeepMiner full-stack architecture transforms data insights into a scalable AI workforce, pushing the business model from software subscriptions to outcome-based profit-sharing. The revenue ramp-up of agent services is converting the compounding effect of data assets and customer stickiness into quantifiable growth realization.
On top of the data asset and collaboration network foundation, agent-based services are reshaping the logic of marketing delivery. MiningLamp-W injects its long-accumulated multi-modal data and AI capabilities into key links, using algorithms to improve standardization in core operational processes, multi-modal content generation to support large-scale supply, and packaging business experience and strategic capabilities into professional judgments for mass output. This creates an end-to-end closed loop from strategy to delivery. In its first year of contribution in 2025, agent services generated over RMB 100 million in revenue, validating the model's feasibility. The deeper MiningLamp-W penetrates the client execution chain, the larger the pool of budget it can tap into. A first coverage report from CSCI Investment notes that the company's agent service revenue is approaching an inflection point, forecasting it will expand to approximately RMB 500 million in 2026 and further jump to around RMB 1.2 billion in 2027, representing 5 times and over 10 times the 2025 level, respectively. Recognizing its high-growth potential, CSCI Investment assigns a 35x price-to-sales ratio valuation to the agent business.
When data compounding drives continuous model optimization, the collaboration network's marginal value increases with scale, and outcome-based pricing models earn premium space as ROI is realized, these three forces form a positive feedback loop. MiningLamp-W's transition from a data intelligence provider to an agent-based service provider is pushing its business model towards an outcome-based monetization model. In summary, the market may be entering a window to re-price MiningLamp-W's industrial logic within the current price range. The anchors for medium- to long-term value are: the sustained increase in the external adoption rate of the Octo open-source ecosystem, the compound growth of agent service revenue as expected, and the synchronization of the on-device AI commercialization penetration pace with policy drivers. As short-term selling pressure clears and market attention returns to the core industrial logic, MiningLamp-W's positioning in the agent-based services domain, alongside the exponential expansion of the agent economy, will continue to be the primary driver for a re-evaluation of its valuation system.
Comments