The Three-Tier Product Matrix of Mininglamp Technology-W (2718.HK) Under the Agentic AI Wave

On June 1, 2026, Jensen Huang announced at the GTC Taipei conference that AI has officially transitioned from the large language model stage into the era of Agentic AI, characterized by the capacity to autonomously observe, reason, plan, and invoke tools.

From Models to Hardware: A Three-Tier Product Matrix

The foresight regarding Agentic AI shared by Wu Minghui, CEO of Mininglamp Technology, formed an "East-West resonance" with Jensen Huang. In December 2025, at the 9th Marketing Science Conference, he pointed out: "Agentic AI gives artificial intelligence a brain while equipping it with hands and feet in the virtual world. Only after being equipped with hands and feet do they truly become a working class and generate productivity, rather than remaining as an AI that only talks without acting."

Since the beginning of 2026, Mininglamp Technology has consecutively placed three critical chess pieces in the direction of Agentic AI.

First, open-sourcing edge-side models and inference frameworks. The company officially open-sourced its edge-side GUI-VLA intelligent agent model, Mano-P, which supports running locally on Mac. By leveraging a pure vision approach to comprehend and operate graphical user interfaces, it bypasses reliance on API integrations and is capable of handling desktop software and complex workflows. The 72B version of this model ranked first globally on the OSWorld dedicated model leaderboard with a success rate of 58.2%, while its 4B quantized version achieved a pre-fill speed of 476 tokens/s on M4 Pro with a peak memory consumption of only 4.3GB. Concurrently open-sourced was the Cider inference acceleration SDK, which boosted operator speeds by 1.4 to 1.9 times compared to native MLX under the W8A8 mode, unlocking the computing power of edge-side INT8 hardware from the foundational layer.

Second, releasing the AI Native hardware Octic. On May 13, Mininglamp Technology introduced its first AI recording hardware tailored for meeting scenarios. Designed in a magnetic card form factor, it is positioned to "accompany you in meetings." Its core innovation lies in "Private AI"—data is processed locally on the device, becoming more understanding of the user over time. Driven by edge-side models and enterprise memories loaded prior to meetings, it executes tasks such as real-time fact-checking, opinion questioning, and argument supplementation during ongoing meetings, while seamlessly connecting to the multi-Agent collaboration platform, Octo.

Third, the Octo platform transforms Agents into organization-level digital workforces. Mininglamp Technology's open-source collaboration platform, Octo, has been deployed and implemented internally among approximately 1,400 employees, with over 2,900 AI Agents running efficiently within actual workflows. The company treats AI agents as organizational members equal in status to humans, transforming them from isolated individual productivity assistants into collaborative, orchestratable, and evaluable organization-level digital workforces.

These three initiatives completely assemble Mininglamp Technology's product matrix into a three-tier architecture of "Model → Platform → Hardware," making it the only AI company on the Hong Kong stock market to simultaneously possess the four-layer capabilities of edge-side models, inference engines, Agent collaboration platforms, and hardware entry points.

Financial Turning Point: From "Scale Expansion" to "Efficiency Driven"

Mininglamp Technology presented three sets of noteworthy data in its 2025 annual report. During the reporting period, the company achieved an operating revenue of RMB 1.426 billion, a year-on-year increase of 3.2%; gross profit reached RMB 790 million, a year-on-year increase of 10.8%; and the adjusted net profit stood at RMB 42.04 million, successfully turning a profit compared to a loss of RMB 45.11 million during the same period in 2024.

Furthermore, the company's gross margin leaped from 51.6% to 55.4%, its operating loss narrowed substantially from RMB 132 million to RMB 15.72 million, and its adjusted operating profit margin grew by approximately 330% year-on-year. The gap between a mere 3.2% increase in revenue and a 10.8% increase in gross profit reveals a pivotal shift: the turnaround in profitability relied not on expanding the scale of revenue, but on the systematic reconstruction of internal efficiency.

It is understood that this stemmed from a large-scale internal AI implementation experiment by Mininglamp Technology. During the reporting period, its subsidiary, Miaozhen Systems, completed a full-link Agentic AI transformation over a span of 60 days. This represented a systematic restructuring of core business segments, including social media data processing, insight analysis, and report generation.

Through AI empowerment, Mininglamp Technology's full-link Agentic AI transformation has boosted the human efficiency of producing deep social media analysis review reports by up to 20 times. It has enhanced marketing intelligent delivery efficiency by up to 4 times and compressed operation intelligent ticket resolution times by over 30%. Secondly, 100% of all employees have connected to the DeepMiner platform, enabling them to autonomously create and deploy exclusive business Agents, which evolves internal standard operating procedures from human execution to human-machine collaborative orchestration.

Recently, Mininglamp Technology publicly announced that Octo, its self-developed open-source multi-Agent collaboration platform, has been deployed and implemented internally among approximately 1,400 employees, with over 2,900 AI Agents operating efficiently within actual workflows. Agent digital workforces are becoming a core element of organizational collaboration.

Guosen Securities pointed out that the AI investment narrative has shifted from the "demand side" to "whether the supply side can sustain subsequent order deliveries." Mininglamp Technology's internal AI-driven transformation serves precisely as a validation of supply-side capabilities—restructuring its own delivery costs with AI before serving clients.

CICC pointed out in its review of Mininglamp Technology's 2025 annual report that the company's adjusted net profit exceeded expectations, benefiting from the operating leverage effect brought by AI empowerment.

According to research by Frost & Sullivan, Mininglamp Technology is the largest provider of data intelligence application software in China based on total revenue in 2024. Building on this solid foundation, the addition of Agentic Services is not merely the introduction of a new business line, but rather the grafting of a new commercial logic onto its original data intelligence infrastructure.

Guotai Haitong Securities pointed out that Mininglamp Technology's new business segment, "Agentic Services," achieved large-scale implementation in 2025, contributing over RMB 100 million in revenue and driving the gross margin up to 55.4%. The company successfully validated a commercial closed-loop based on "performance-based billing," achieving a key account retention rate of 96%, with over 30% of newly acquired key accounts originating from this new business.

Inclusion in Stock Connect Earns Phased Recognition; AI Industry Remains in Its Preliminary Stages

CITIC Securities pointed out that the AI industry is still in its early stages where opportunities outweigh challenges, and the model iteration cycle will continue to shorten. At this node of macro-narrative switching, as the first listed company on the Hong Kong stock market to possess a complete technology stack for Agentic AI, Mininglamp Technology's milestone of achieving profitability in 2025 combined with its dense release of technical products in the first half of 2026 serves as a concrete footnote of the industrial wave materializing within a single enterprise.

On June 1, 2026, Mininglamp Technology was included as an eligible security under the Stock Connect program. Stock Connect enforces strict threshold requirements regarding the market capitalization, liquidity, and compliance of listed companies. Its inclusion marks a phased recognition by the market of its capital volume and trading activity.

Spanning from its major open-source moves in May to its share price hitting a record high, and further to its inclusion in Stock Connect resonating on the same day as the GTC conference in June, Mininglamp Technology's cumulative gain within the month has exceeded 40%. Recently, under the overall correction of the Hang Seng Tech Index, the share price of Mininglamp Technology-W was also affected and experienced a pullback.

TF Securities stated that the volatile conflict situation has impacted market risk appetite and liquidity, which also created expected perturbations in the AI technology industry. Overall, we are actively optimistic that the demand for domestic AI infrastructure will maintain high prosperity in 2026, while applications are expected to continuously bear fruit.

NewTimeSpace Disclaimer: All content herein is the original work of NewTimeSpace. Any reproduction, reprinting, or use of this content in any other manner must clearly indicate the source as "NewTimeSpace". NewTimeSpace and its authorized third-party information providers strive to ensure the accuracy and reliability of the data, but do not guarantee the absolute correctness thereof. This content is for reference only and does not constitute any investment advice. All transaction risks shall be borne by the user.