Hong Kong Exchanges and Clearing Limited and The Stock Exchange of Hong Kong Limited take no responsibility for the contents of this announcement, make no representation as to its accuracy or completeness and expressly disclaim any liability whatsoever for any loss howsoever arising from or in reliance upon the whole or any part of the contents of this announcement.

MININGLAMP

TECHNOLOGY

Mininglamp Technology

明略科技

(A company controlled through weighted voting rights

and registered by way of continuation in the Cayman Islands with limited liability)

(Stock Code: 2718)

ANNUAL RESULTS ANNOUNCEMENT FOR THE YEAR ENDED 31 DECEMBER 2025

FINANCIAL HIGHLIGHTS
	For the year ended 31 December
	2025 (RMB’000)	2024 (RMB’000)
Revenue	1,425,775	1,381,382
Gross profit	789,621	712,694
Operating loss	(15,721)	(132,347)
Adjusted operating profit (non-HKFRS measure)	24,984	580
Adjusted net profit/(loss) (non-HKFRS measure)	42,043	(45,113)

The Board is pleased to announce the audited annual consolidated results of the Group for the year ended 31 December 2025, together with comparative figures for the year ended 31 December 2024. These annual results have been extracted from the audited financial statements of the Group and have been reviewed by the Audit Committee.

CHAIRMAN'S STATEMENT

Dear Shareholders,

In 2025, Mininglamp Technology completed a strategic transformation: shifting from helping clients understand data to enabling them to realize tangible business outcomes. This shift is not merely an expansion of business lines, but a reflection of our view on the essence of AI commercialization: AI is a data-driven computing paradigm that must form a closed loop within real-world business scenarios to achieve continuous evolution. We deliver measurable business results to customers, and their real-world feedback, in turn, continuously drives the iteration of our models and Agents. This two-way closed loop is, in our view, the core path to scaled AI commercialization across industries. It is also the fundamental logic behind Mininglamp's evolution from Data Intelligence to Agentic Services – distinct from traditional software licensing or project-based delivery.

Unlike traditional software licences or project-based delivery, under our Agentic Services model, customers pay for measurable business results – improved marketing ROI and multiplied content-production efficiency – while AI-driven automation lowers our marginal cost of delivery. Building on our Data Intelligence business, we officially launched Agentic Services as a new business line in 2025 to elevate our closed-loop value delivery.

In 2025, the Company recorded revenue of RMB1,425.8 million, representing a year-on-year increase of 3.2%. Revenue from the Data Intelligence business line grew steadily to RMB1,260.4 million. This growth was primarily driven by improved performance in the Operational Intelligence business line, stemming from two key factors: first, iterative upgrades to our conversational intelligence products better met clients' needs for real-time data access, alongside continued expansion of our sales channels, which drove an increase in related revenue; second, ongoing enhancements to AI capabilities of the smart store operating system and a broader set of its application scenarios, which drove revenue growth for this product line. The Agentic Services business line generated revenue of RMB100.2 million. Revenue from this business line during the year was primarily attributable to the pay-for-performance model, which effectively met market demand for optimizing return on investment. It also provided initial validation of the Company's operational advantage in lowering marginal costs through Agent-based empowerment and demonstrated potential for scalable growth. Benefiting from the above-mentioned sustained growth in quality revenue and significant operational leverage from AI-enabled initiatives, the Company achieved an adjusted net profit (non-HKFRS measure) of RMB42.0 million for the year, marking a successful turnaround to profitability.

Agentic AI-driven Upgrades of Delivery

In 2025, we further invested in DeepMiner, our trusted business intelligence agent – an enterprise-grade AI Agent platform, developed by Mininglamp – powered by our proprietary models Cito and Mano. An upgrade to Version 2 has now enabled DeepMiner to deliver stronger task planning, data-ingestion and tool use capabilities. Leveraging DeepMiner's capabilities, we improved the operational efficiency of our Data Intelligence business line. Specifically, delivery efficiency within the marketing intelligence business improved by up to 400%, and task resolution time in operational intelligence fell by over 30%.

Rooted in the continuous operation of Data Intelligence, together with its large and growing accumulation of marketing data and customer business scenarios, we have built the Agentic Services business line from the ground up. In marketing, DeepMiner now supports a full business flow that spans end-to-end services across insight and planning, multimodal content production and distribution, and marketing placement – directly delivering measurable marketing outcomes to customers. On average, we helped customers achieve a 20% improvement in marketing effectiveness with nearly three times the operational efficiency. In content, the extension of our marketing placement services enabled us to identify an even greater opportunity in the generation of digital content. Leveraging AI-driven technical service capabilities, we entered through advertising placement services and progressively expanded into the intelligent planning and technical production of digital content, successfully establishing a complete closed loop system that spans from digital asset management to full domain monetisation – marking a natural extension of the Agentic Services model into the digital content ecosystem.

Deepening and Expanding Customer Value

As we extended offerings from Data Intelligence to Agentic Services, both the breadth and depth of our customer relationships grew substantially. In 2025, the KA clients renewal rate reached 96.0%, and our core customer base remained stable. Over 30% of new KA clients were acquired through the Agentic Services business, validating this model's applicability across customer types and scenarios.

In terms of our service depth, we achieved full-link extension from data insights to content production and marketing execution with multiple leading customers. For example, a renowned automotive brand advanced from data insights to entrusting us with content production and marketing placement, achieving a complete upgrade from insights to result delivery. In terms of customer base, we broadened our service scope to facilitate the cross-border operations of both domestic brands and multinational corporations. For instance, a renowned smart consumer-electronics customer extended our services to multiple global markets, enabling cross-regional tracking and analysis of brand communication effectiveness.

Building Data and Technology Moats

We have long held that true differentiation comes from proprietary data accumulation and building industry benchmarks. In 2025, we continued investing in our core data capabilities: AdEff, our AI-driven creative testing platform, converts consumers' subjective perceptions into quantifiable indicators to help our clients filter out underperforming creatives before launch; Brand AI Perception Assessment (Brand GEO Index) tracks brands' competitive positions across mainstream AI search engines; Firstdata, our global authoritative data knowledge base, converts 1,000+ authoritative sources into traceable, structured facts; and WebRetriever, our web-Agent evaluation benchmark, covers 800 websites and 1,500 tasks to assess Agents' real-world execution capabilities.

Fueled by the continuous accumulation of professional task data, our self-developed models have achieved significant breakthroughs. As of October 2025, our proprietary GUI Agent large model Mano ranked second overall on the OSWorld leaderboard, behind only Anthropic. As a core evaluation benchmark trusted by top-tier global AI companies such as OpenAI and Anthropic for new model releases, our performance in OSWorld serves as a validation that Mininglamp is at the global forefront in its foresight and investment in Agent technology. Mano also topped the Mind2Web leaderboard, achieving global dual-benchmark leadership. Cito, the reasoning model responsible for deep planning and decision-making within DeepMiner, ranked first among small-sized models on the BFCL leaderboard. Mano and Cito form the core technology foundation of DeepMiner; their performance on authoritative benchmarks directly defines the maximum potential of our trusted business intelligence agents' real-world capabilities. The accumulation of these data assets and model capabilities underpins the ongoing evolution of Mininglamp's trusted business intelligence agents, and constitutes the bedrock of our long-term competitiveness.

Capital Foundation and Strategic Confidence

In 2025, Mininglamp Technology successfully listed on the Main Board of the Hong Kong Stock Exchange, laying a solid capital foundation for long-term investment in core technologies and new markets. While we are fully aware of the intensifying competition in the AI industry and rising client expectations for outcome delivery, we also recognize the challenge in overcoming operational complexity brought by scaling Agentic Services model. We are equally convinced, however, that increasing investment in technology and delivery capabilities during this critical period is the right path to building sustainable competitive advantage.

Outlook for 2026

In early 2026, OpenClaw emerged at record speed as one of the world's most closely followed open-source AI Agent frameworks, marking the arrival of the inflection point where Agentic AI (Autonomous Artificial Intelligence Agents) truly transitions from a technical concept to large-scale application. Meanwhile, this shift has also brought new challenges: as AI becomes deeply embedded in business decision-making and daily tasks, security, controllability and interpretability will become core industry priorities.

This is the vision to which Mininglamp has long been committed. At Mininglamp, our mission is to build trusted AI. We believe AI's ultimate value lies not in replacing humans but in evolving toward truly trustworthy AGI by gradually and steadily extending and amplifying human intellectual activity. Our long-term investments in trusted business intelligence agents – from data traceability to model interpretability, from the transparency of task planning to the quantifiable verification of results – are our response to that challenge. Above all, each of our commitments testifies to our belief that using AI to streamline work, improve the quality of life for humanity, and ultimately enable human-AI collaboration with shared purpose is the most meaningful task of our era, and Mininglamp will dedicate itself fully to it.

In 2026, our core task is to scale up our proven Agentic Services delivery capabilities:

Scaled replication: Systematically replicate proven Agentic Services models across a broader range of scenarios and customer groups, continuously deepen capabilities in content marketing and content entertainment, and boost operational efficiency through AI-driven business process transformation.
Expanding Agentic Services boundaries: Actively explore more Agentic Services opportunities in digital white-collar work.
Edge-side AI deployment: Upgrade Lingting (smart ID badge hardware) into an edge-side AI device serving as a real-time multimodal data ingestion point; collaborate with hardware vendors to develop compact edge-side models so that Agents can run on-premise at enterprises, addressing data sovereignty and inference cost issues.
Global expansion: Leverage domestic enterprises' overseas services and digital content as anchors to systematically expand the overseas footprints of products and services.

We extend our sincere gratitude to all employees for their unremitting efforts and dedication over the past year, and to all shareholders and partners for your trust and support. We look forward to witnessing Mininglamp's growth in this historic opportunity together with you.

Minghui Wu
Chairman of the Board and Executive Director

March 2026

MANAGEMENT DISCUSSION AND ANALYSIS

In 2025, the global professional services industry was undergoing a period of structural transformation driven by artificial intelligence (AI) technology. With the proliferation of large model technologies, AI applications are evolving from being mere efficiency-enhancing tools to becoming foundational components capable of handling complex business processes. This technological trend is reshaping the cost structure and delivery models of knowledge-intensive industries. Against this backdrop, companies that can integrate vertical industry expertise, proprietary data assets, and AI architecture are gradually establishing differentiated competitive advantages.

Business Model Evolution: Shifting Towards Quantifiable Business Returns

Against the backdrop of refined macro-budget management, the procurement logic of professional services on the enterprise side is changing. Client demands are gradually shifting from standardized process services to achieving quantifiable business outcomes. The increasing emphasis on return on investment is particularly evident in areas like marketing and commercial conversion. At present, the application of AI technology in the marketing sector has progressed beyond the "proof-of-concept" stage to formally become an indispensable core infrastructure. According to the survey in the IAB's 2026 Outlook, five of the top six priority areas most focused on by advertising buyers are directly related to AI. Empowered by AI technology, the core requirements of brand advertisers are evolving from broad exposure to quantifiable business returns. According to relevant studies by Bain & Company, global marketing budgets are accelerating their migration from display advertising to performance- and outcome-based advertising. Such a substantial migration in budget structure signifies that service providers equipped with AI attribution modelling and cross-channel budget optimisation capabilities, and capable of delivering verifiable results, will assume a dominant position in the reallocation of market share.

Shifting Information Distribution: Generative Interaction Reshapes Commercial Visibility Rules

The upgrade of underlying technology is reshaping the interaction pathways between enterprises and their audiences. As large model technology transforms various information platforms, the way users acquire information is progressively shifting from "keyword-based web indexing" to "intent-driven direct response generation". Data from third-party institutions indicates that traffic to traditional search engines will gradually decline over the next 2 to 3 years. This shift in interaction paradigm signifies that brand visibility and recommendation weightings within AI-generated content have emerged as the core focal point of competition for next-generation traffic gateways. In the new digital ecosystem, the reach efficiency of corporate information no longer depends solely on page ranking, but on whether the large models can accurately retrieve that company's commercial corpus when generating answers. Service providers capable of leveraging trustworthy data assets to transform a company's deep business knowledge into "native decision-making references easily adopted by large models" are becoming powerful enablers for enterprises to maintain their commercial competitiveness in this new digital landscape.

Redefining Competitive Barrier: Democratisation of AI Tools Accentuates the Barrier of "Data + Know-how"

As the cost of accessing general-purpose large models decreases and basic algorithms become widespread, the barriers based solely on technological tools are diminishing. This trend inversely reflects the scarcity of proprietary operational data and industry expert experience within specific vertical business scenarios. The 2026 B2B Content Marketing Trend Report indicates that, among the factors enhancing marketing efficiency, the weightings of content quality and relevance (65%) and professional skills of the team (53%) outweigh that of technology and tools themselves (43%). This reflects the advanced path for enterprise service providers to build their moat: encapsulating comprehensive data and deep industry knowledge into AI systems is becoming the core driver for service providers to establish a long-term business moat.

The resonance of the aforementioned industry trends defines the entry standards for the next generation of professional service providers. Against this backdrop, Mininglamp Technology is undergoing a strategic leap from providing data insights to delivering business outcomes. Leveraging its proprietary data assets accumulated through years of deep industry engagement, validated AI capabilities in vertical fields, and an end-to-end delivery system ranging from insight to execution, the Company is transforming its technological accumulation into resilient business growth through business model upgrades.

The Company's business growth is founded upon a clear and highly synergistic value chain. The overall business architecture is structured around "one core technology foundation + two major business engines":

Core Technology Foundation: Based on the implicit commercial knowledge and holistic quantitative benchmarks accumulated for 20 years of long-term services provided to top-tier brands, the Company deeply integrates its proprietary large model technology to construct DeepMiner, an enterprise-grade trustworthy AI Agent platform. Serving as the unified infrastructure underpinning the upgrade development of the Company's end-to-end commercial operations, this platform transcends the simple integration of traditional functionalities, innovatively constructing a rigorous three-tier collaborative architecture of "Orchestration - Decision-making - and Execution" (Foundation Agent, Cito and Mano) within the industry. This structure design directly enables a modular delivery model and mitigates the uncertainties of large models in commercial applications at the physical mechanism level. The platform uniformly incorporates the high interpretability of systemic decision-making, data isolation at the physical level, and cutting-edge Agent execution computing power into its delivery standards, thereby circumventing compliance frictions and AI hallucination risks during enterprise-grade deployments. This enables the Company to progress beyond the sheer tool-assistance stage and achieve the scaled output of trustworthy productivity characterised by "auditable processes and verifiable results" to enterprise customers with more stringent requirements for compliance and commercial returns, thereby laying a solid cornerstone of trust for the commercial application of industry-grade AI.

Data Intelligence Business: Serving as the cognitive foundation of the entire business ecosystem, this business line deeply integrates the native AI capabilities of DeepMiner. Relying on the AI capabilities of DeepMiner and the Company's cross-platform data assets, the business provides enterprises with comprehensive solutions, including independent third-party marketing effectiveness measurement and verification, cross-platform consumer insights and strategy optimisation, as well as the digitalisation of offline sales pipeline and the intelligent operation of chain stores. During the processing of real-world commercial transactions, the business continuously cleanses and constructs a dynamically evolving system of over 380 high-granularity data tag taxonomies and a massive volume of effectiveness benchmark data. To date, the Company has served over 2,000 brand customers and more than 240,000 enterprise clients. By continuously leveraging the data assets that are progressively enriched alongside business operations, it effectively assists enterprises in achieving precise business attribution, optimising commercial decisions and enhancing intelligent operational efficiency.
Agentic Services Business: Addressing the structural constraints of the traditional marketing services model, which is hindered by vertical experience barriers and the linear expansion of manpower, the Company systematically infuses proprietary data and native AI capabilities into key nodes of the marketing value chain. This has achieved a structural improvement in delivery efficiency through chain reshaping: algorithmic orchestration drives efficiency gains in core links, breaking the linear dependence of output on manpower; a multi-modal architecture underpins the programmatic supply of content, amortizing marginal costs; and profound implicit business experience is encapsulated into professional judgments for scaled output, ensuring the certainty and high standard of complex business decisions. Such underlying reconfiguration integrates the end-to-end delivery closed loop. Building upon this, the service team has dismantled the experience boundaries and data silos of a single industry, significantly expanding the number of serviceable customers and industry coverage. Concurrently, the comprehensive integration of the DeepMiner platform compresses workflows that previously required weeks into mere hours or even real-time responses. The significant release of production capacity drastically reduces the marginal delivery costs of expansion, directly driving the structural improvement of per-employee efficiency and gross margin space. By breaking away from the traditional manpower-based billing model and deeply aligning with customer delivery outcomes, the Company is progressively transforming this business into a cross-market, multi-business format, and highly replicable "Agentic Agency".

The two major business lines, Data Intelligence and Agentic Services, form a rigorous logical closed loop and mutually support each other atop the DeepMiner foundation. During long-term, high-frequency services, the Data Intelligence business continuously accumulates and compliantly cleanses a massive volume of cross-platform and cross-cycle effectiveness benchmark data. Concurrently, the Agentic Services business systematically and continually transforms frontline operational experience into industry knowledge, whilst continuously transmitting back authentic conversion feedback from holistic touchpoints. The aforementioned bi-directional data and experience are continuously fed back into the underlying technology, directly driving the reinforcement training and the evolution of judgement capabilities of DeepMiner. The evolved AI capabilities, in turn, enhance the product precision and delivery ROI of the two major business lines. Such synergistic mechanism of insights guiding execution,

execution verifying insights, and data uniformly feeding back into the foundation ultimately fosters a virtuous cycle of “data accumulation – model iteration – business growth – further data accumulation”. This has become the core driving force for the Company to consolidate its long-term endogenous growth and broaden its leading competitive moat.

DeepMiner

DeepMiner is an enterprise-grade trusted AI Agent platform independently developed by the Company, primarily targeting the realm of commercial data analysis and decision-making. The research and development of the platform commenced in early 2025. Following the interconnection of core chain connectivity and multiple rounds of iteration, Version 1 (V1) was officially launched in September 2025, and a significant leap to Version 2 (V2) was accomplished within the year (accumulating over 70 major and minor version iterations throughout the year). Serving as the core carrier for driving the large-scale implementation of Agentic AI, the platform systematically infuses AI capabilities into enterprise business workflows, realising a paradigm shift from a mindset of incremental process-stacking to one of systemic ecosystem reimagining.

Addressing the industry pain point whereby a single Agent struggles to underpin complex closed-loop business operations, DeepMiner adopts a Multi-Agent collaborative architecture, and leverages an innovative AI Agent technological support system comprising “one central hub and two major models” to establish a rigorous hierarchical collaborative architecture of “Orchestration – Decision-making – Execution”:

Foundation Agent (Multi-Agent collaborative hub): Serving as the intelligent hub of the entire system, it is responsible for cross-task resource scheduling, allocation, and result integration. This layer adopts a MoA (Multi-Agent) collaborative architecture, supporting the automatic decomposition and parallel scheduling of complex tasks. Supplemented by a Human-in-the-loop intervention mechanism, it effectively addresses the issues of access control and complex requirement alignment in enterprise-grade deployment, thereby ensuring the high execution accuracy of the system within vertical domains. DeepMiner has currently been broadly integrated into mainstream marketing and e-commerce ecosystems, and its efficacy in task decomposition and collaborative scheduling within complex business environments has been substantially validated across multiple authoritative industry evaluations and technological awards.
Cito: Focuses on the in-depth application of industry knowledge and professional reasoning. The model incorporates tacit commercial knowledge accumulated by the Company for 20 years, spanning a diverse range of industries: encompassing holistic quantitative benchmarks, competitive strategies, quantitative models for media assets, and analytical frameworks for consumer decision-making, ranking first among small-sized models in the vertical domain evaluation of the Berkeley Function Calling Leaderboard (BFCL). Its core technical barrier lies in converging the uncertainty of large models in quantitative analysis; by dynamically constructing reasoning chains tailored to specific business scenarios, it provides highly deterministic and stable strategic outputs for complex commercial problems.
9 -

Mano: Focuses on bridging the “last mile” of system execution. Leveraging advanced multi-modal visual understanding technology, the model can accurately parse the front-end interface elements of heterogeneous software under strict user authorisation and security compliance controls. This technology substantially lowers the technical barriers to cross-platform information flow and task execution. With a mere 72B parameters, the model ranked first place globally among specialised models in the OSWorld international evaluation. Its support for on-premises private deployment safeguards the stringent security of core enterprise data at the physical level, thereby drastically reducing computing power costs and IT adaptation costs for customers.

DeepMiner V2 has achieved a transformation in its underlying logic from being function-driven to capability-driven, establishing a platform ecosystem wherein “everyone can develop, everything can be connected”. On the data access front, the platform has restructured the traditional system integration paradigm. Leveraging the MCP (Model Context Protocol) and the capability of the Mano model, DeepMiner is equipped with rapid connectivity to navigate fragmented, heterogeneous IT ecosystems. This lightweight data integration approach substantially shortens the adaptation cycle for cross-platform interfaces, making extensive coverage of mainstream marketing and e-commerce ecosystems a reality.

On the development front, DM-Builder achieves plug-and-play connectivity for heterogeneous systems based on the MCP, and offers visualised Agent construction and workflow orchestration capabilities. This empowers non-technical personnel to seamlessly complete data access and Agent configuration, thereby drastically shortening the cycle from requirement to delivery. On the application front, the DM-Skill space modularises business capabilities, encompassing multiple business scenarios such as social media marketing reviews, advertising efficacy research, and monitoring report generation. It supports business personnel in independently building and deploying scenario-based AI applications, facilitating the reuse of capabilities across teams and business lines.

As at the end of the Reporting Period, DeepMiner has been deeply embedded into the principal business lines of the Company, achieving a leap from technical verification to commercial application at scale. The platform provides robust underlying support for the Company’s businesses: in the Data Intelligence business, AI has systematically reconstructed the processing pipeline for massive data, substantially reducing marginal labour costs while achieving end-to-end efficiency enhancements in the measurement and monitoring systems; in the Agentic Services business, the platform deeply couples the profound commercial insights accumulated by the Company for 20 years with the automated execution capabilities of AI Agents. This compound leverage of “industry cognition + AI computing power” is widely deployed across the value chain, encompassing multi-modal content production, influencer selection, and other critical operational nodes, driving a transformative leap in execution productivity to deliver quantifiable commercial results to customers. This systematic transformation methodology has been rolled out and validated across multiple traditional marketing agencies, establishing a standard paradigm for Agentic AI to reconstruct industry production relations, thereby formulating a standardised and scalable path for business transformation.

10 -

Data Intelligence

Data Intelligence business comprises two main sub-business lines, being marketing intelligence and operational intelligence, which cover the core needs of brand clients through multiple product systems. By analyzing massive user behavior data from various channels, Miaozhen Systems help our clients optimize digital and out-of-home advertising, enhance marketing efforts on social media platforms, and manage and utilize data to drive client engagement and growth. Jinshuju provides zero-coding business data management and data collection platform services to enterprise clients. Private domain tools based on the Tencent ecosystem provide enterprises with customer interaction management for both private and public domains through WeCom and WeChat. The operational intelligence service's conversational intelligence helps enterprises digitize their offline sales processes and optimize them with AI analysis. The smart store operating system aims to achieve end-to-end intelligent operations by deeply digitizing and automating the service processes.

Through core product systems covering online omni-channel marketing and offline physical operations, this business provides enterprise clients with a full-scenario digital foundation. In the strategic evolution in 2025, relying on the integration of underlying large models and Agentic AI technology, the business has achieved a fundamental leap in business logic, moving from empowering and improving efficiency to ecosystem reconstruction.

The marketing intelligence services has achieved a leap from passive data measurement to proactive intelligent decision-making. By encapsulating complex strategy analysis, content flow and private domain operation actions into autonomous Agents, the business not only unleashes non-linear productivity gains in end-to-end delivery, but also successfully expands commercialization boundaries of the product, raising the ceiling of single-customer value.
The operational intelligence services focuses on intelligent restructuring of offline physical businesses. Facing labor-intensive and complex offline sales and supply chain networks, the business deeply penetrates omni-channel business data, and transforms highly non-standard industry tacit knowledge into algorithmic models available for scalable invocation, thereby substantially optimizing the unit economic model of chain enterprises in a zero-sum competitive landscape.
11 -

Marketing Intelligence

Miaozhen Systems

As the core marketing intelligence brand under Mininglamp Technology, Miaozhen’s business has consistently ranked first in China’s digital marketing technology and data service market. In 2025, despite pressures from the macro market, this business demonstrated significant structural resilience thanks to its robust underlying data assets and full-stack technical architecture. The Company’s business is highly focused on core nodes of the enterprise marketing value chain, and builds a complete closed loop for clients from strategic insights to growth monetization through three main product lines:

Media spending optimization: It covers the entire matrix of media scenarios including PC, mobile, internet TV and outdoor advertising. Leveraging artificial intelligence and real-time data analysis, it provides coordinated budget allocation, precision advertising placement and performance monitoring to enhance the scientific rigor of media strategies and improve ROI.
Social media management: It focuses on marketing effectiveness evaluation and strategy enablement within the social media ecosystem. Based on data analytics models, it provides marketing attribution, influencer selection and audience insights to help clients optimize interaction strategies and use quantitative data to drive product innovation.
Customer growth: It focuses on the accumulation and application of underlying customer assets. By integrating customer data management, marketing automation and analysis modules, it supports personalized reach and refined operation, demonstrating strong business stickiness in cross-cycle member management.

During the Reporting Period, driven by cost control and efficiency improvement and optimized business structure, the three core business lines demonstrated operational resilience in a challenging macroeconomic environment, further increasing the gross profit margin.

12 -

During the Reporting Period, relying on DeepMiner's trusted enterprise AI Agent platform, Miaozhen Systems comprehensively promoted the transformation of Agentic AI for workflow. The project achieved end-to-end connectivity within 60 days. The team completed 99% of the underlying data access in 45 days, laying the foundation for efficient orchestration at the application layer and systematically addressing the challenge of integration of social media data processing with AI applications. While retaining necessary Human-in-the-loop, the end-to-end AI automation completion rate reaches 90%, achieving seamless integration from underlying data to business conclusions.

Taking the highly structured and data-intensive scenario of "social media insights and report delivery" as an example: the Company adopts a maturity-level intelligent orchestration strategy to encapsulate data search, statistical analysis and chart generation into standardized AI skills.

This technological upgrade has systematically reshaped the business cost structure and effectively improved output quality:

Non-linear efficiency improvement: Taking one in-depth report as an example, the delivery cycle under the traditional model was as long as 10 to 15 working days (equivalent to 15 to 20 person-days of actual labor input). The labor efficiency improvement represents a 20-fold improvement under the current model. Business teams are thus freed from repetitive data processing and can shift to high-value strategy development.
Insight quality improvement: Standardized workflows eliminate subjective human biases. The report output achieves substantial increase of 40, 30, 20 and 15 percentage points in insight stability, framework completeness, depth of insights, and novelty in micro-trend identification, respectively.

In addition to end-to-end report delivery, internal AI-driven efficiency improvements have been deeply integrated into the Company's other high-frequency business nodes:

OOH manual monitoring service: For the review of outdoor advertising surveillance photos, it has built a workflow of "AI pre-screening + manual review". The introduction of visual large model technology has increased the overall review efficiency to four times the original, and through iterative optimization, the re-review rate in the re-inspection stage has been reduced from 30% to 5%.
SEO tagging project: The percentage of automated processing jumped from 0% to over 90%. After model replacement, only about 10% of manual quality inspection is retained, achieving near-total automation while ensuring quality.
Demand research and preprocessing: In the early stage of research, structured statistical results are generated directly through natural language commands, replacing the traditional pivot and filtering process. This increases the overall testing efficiency by 40% and avoids errors caused by manual statistics.
13 -

Automatic generation of label rule sets: It changes the traditional model that relies on subjective experience of experts. With the use of AI to automatically generate classification rules based on samples and assign confidence scores, the rule generation speed is accelerated to five to ten times the original speed.
Large-scale data quality verification: A collaborative model of “AI initial labeling + manual verification” is built in the massive label testing stage. Taking the quality inspection of thousands of data points as an example, the task time is reduced from more than ten hours to less than three hours, which increases efficiency by 70% while eliminating inconsistencies from manual judgment.

Transcending the limitations of single-point efficiency tools, Miaozhen has successfully embedded its Agentic AI capabilities deep into the underlying business chain. Beyond optimizing our unit economic model, this strategy translates specialized, non-standard expertise into reliable business decision-making capabilities at scale – solidifying our position as the central hub and the value engine for next-generation business operations.

In the future, Miaozhen will focus on three strategic directions: Firstly, it will consolidate and expand its high-growth business lines to deepen customer loyalty through technological iteration and maintain growth momentum. Secondly, it will constantly deepen internal AI transformation to substantially translate efficiency improvement gains into increased gross margins. Thirdly, it will explore the extended applications of data assets in new scenarios such as digital content, e-commerce and overseas marketing, thereby establishing a synergistic value chain between Miaozhen Systems and other business lines of the Company.

Jinshuju

As a zero-coding data collection and management platform under Mininglamp Technology, Jinshuju provides enterprises with a foundational base covering data collection, flow and management. The platform relies on a drag-and-drop editor and a vast library of templates to support high-frequency business scenarios such as questionnaires, appointments, collections and quizzes. The platform achieves multi-terminal coverage across PCs, Apps and mainstream office collaboration applications, which is designed to upgrade lightweight forms into a transferable business data system with an extremely low skill threshold.

In 2025, in the face of rising traffic costs and intensifying competition for existing market share in the industry, Jinshuju shifted its focus to optimizing its business structure and reconstructing its underlying technology. During the Reporting Period, on the basis of maintaining a stable paid conversion rate for its core subscription system, the platform’s underlying technology underwent an upgrade to an Agent-driven model.

14 -

In product distribution and iteration, the deep integration of AI computing power has substantially changed users' usage paradigm. By deploying a full-stack Builder Agent matrix, core business processes have evolved from traditional manual editing to intelligent generation. Currently, the monthly average call volume of various business Agents on the platform is approaching one million, and the proportion of forms directly generated by AI has increased from $4.68\%$ to approximately $15\%$ . Coupled with a systematic upgrade of the mobile terminal's underlying architecture, this intelligent generation capability has been seamlessly deployed across multiple terminals, which further strengthens the platform's long-term ecosystem compatibility.

The in-depth penetration of AI technology into internal R&D and operation processes has systematically optimized the cost structure and productivity model of Jinshuju:

Risk control system: The system-level risk control Agent has taken over more than $90\%$ of the deep-level detection of forms, images and hidden links on the entire site, which effectively reduced the cost of manual review.
R&D efficiency: After introducing AI intelligent programming assistance tools, the size of the R&D team was reduced by approximately $14\%$ while the frequency of system launches increased by $64\%$ during the year despite the headcount reduction.
Customer service operations: The AI-powered customer service system, based on our self-developed RAG technology, is deployed 24/7. During the Reporting Period, the AI-powered customer service system saved approximately $30\%$ of labor costs, reducing operational expenses while maintaining stable service capabilities.

At the same time, the Company is actively exploring ways to diversify its revenue streams. The transaction monetization system, represented by "Xiaojin Merchants" (小金商户) completed its business loop from scratch during the year, with a total of 900 merchants registered. This business operates with a very lightweight staffing, exploring new revenue streams for the traditional subscription model. Data shows that over half of the new collecting merchants have simultaneously converted into paid subscribers to the platform, which initially verified the feasibility of the "subscription + transaction" synergistic customer acquisition model, laying a foundation for the platform's subsequent revenue model upgrade.

16 -

Private Domain Tools Based on the Tencent Ecosystem

Leveraging Tencent’s WeCom ecosystem, the Company has constructed a digital operation foundation, providing enterprises with a full-chain SCRM (Social Customer Relationship Management) solution that spans “Marketing, Sales, and Service”. In 2025, this business achieved significant improvement in core operating metrics and enhanced business health through back-end cost reduction and front-end efficiency gains. While maintaining stable average revenue per user within its core subscription base, the Company not only significantly reduced cloud marginal costs through technological optimization of its server architecture but also unlocked commercial potential by restructuring the product foundation towards Agentic AI. As of now, this business maintains a solid position among the top three in the industry, with over 240,000 enterprise clients and 790,000 enterprise users, and continues to sustain a high renewal rate of over 60%.

In business implementation, the platform has successfully deployed business Agents covering seven core scenarios, including customer insight, conversation analysis, intelligent strategy, and automatic generation. Currently, this intelligent infrastructure has deeply penetrated the enterprise base across these nearly 80 vertical industries, indirectly underpinning precise outreach to over 300 million WeChat end consumers. Data from the platform indicates that since the fourth quarter of 2025, the proportion of calls to the “customer insight” Agent, based on deep data mining, has surpassed that of the basic AI intelligent reply agent. This marks that client demand for AI applications has formally extended to core decision-making in their business operations. By proactively identifying high-intent business opportunity signals, the platform has substantially enhanced resource conversion efficiency.

Meanwhile, the Company’s official account editor has steadily maintained a market share of approximately 30% in the relevant sector. During the Reporting Period, this business achieved a strategic upgrade by transforming its workflows through Agent-based reconstruction. The product has evolved beyond its single-tool origins and has established a full-process Agent system that spans “topic selection, writing, formatting, and publishing”. This reconstruction of productivity has directly driven a counter-cyclical expansion of the business: during the year, paying users and monthly active users have grown by 21% and 44% year-on-year, respectively, solidly validating the scalable penetration potential of fully-process automation tools in work scenarios.

This successful practice based on Agentic workflow is accelerating integration into the entire digital operation foundation of the Company. Looking ahead, the technological evolution of this business will rapidly advance from assisted analysis to autonomous decision-making, leveraging customized intelligent workflows to further take over complex execution tasks in enterprises’ private domain operations. On this basis, through the progressive deployment of standardised Agent capabilities and tailored deployments for high-net-worth clients, the business will build a stabilizing force against market fluctuations and continuously capture high-value market share in the increasingly competitive SCRM market through technological barriers.

Operational Intelligence

Conversational Intelligence

The conversational intelligence business primarily focuses on four scenarios: sales process analysis, marketing campaign analysis, employee comparison analysis, and customer demand analysis. It converts massive unstructured voice data generated during sales processes into structured insights with high commercial value. In 2025, this business substantially completed the digital closed loop for full-sales-cycle management through generational leaps in its technology foundation and forward-looking integration of software and hardware. During the Reporting Period, this business completed a deep transition from self-developed small models to large-model architecture, achieving comprehensive automated penetration of sales compliance and conversion logic through extensive training on massive sales scripts from vertical industries. This reshaping of the technological pathway not only reduced reliance on manual labeling and re-inspection at the foundational level but also propelled the business model transition from traditional labor-intensive services to algorithm-driven products. This qualitative change from unstructured voice data to standardized business insight signifies that the Company has successfully converted AI capabilities into verifiable value output, establishing a technological leadership position in deep decision-making scenarios.

While consolidating its technological advantages, the Company has built a strong competitive advantage through an integrated ecosystem of software and hardware. The 4G "Lingting" (靈聽) device launched in early 2025 achieved significant optimization in outdoor adaptability, ease of operation, and remote control costs, achieving a time-to-market notably faster than the industry average. Through the deep integration of self-developed hardware and cloud systems, the Company has achieved omnichannel capture of online (WeCom, conferencing systems, call centers) and offline (in-store sales recordings) data. This capability to capture multi-dimensional sales conversation data not only strengthens the proprietary nature and completeness of data acquisition but also provides enterprise customers with a rare, holistic perspective through cross-period and cross-channel correlation analysis of online and offline behaviors, solidifying the Company's core position as the foundational platform for decision-making in offline sales process management.

While maintaining competitiveness in core sectors such as automotive and real estate, this business demonstrated cross-industry expansion momentum in 2025. Leveraging the rapid deployment of standardized Agent capabilities, it has successfully penetrated multiple high-potential fields, including 3C, fast-moving consumer goods (FMCG), medical aesthetics, healthcare, and education and training. Currently, more than 18 benchmark projects in key industries are in the proof of concept (POC) stage, highlighting the product's versatility and strong replicability across different depth decision-making scenarios. This structural expansion from core sectors to full industry scenarios not only validates the growth resilience of the conversational intelligence business but also provides solid strategic support for the Company to consistently capture high-value market shares in the increasingly competitive digital transformation market.

17 -

18 -

Smart Store Operating System

As the physical retail sector enters a period of competition for existing market share and refined operations, the Company has established the strategic positioning of the smart store operating system. This system focuses on the front, middle and back-office management of offline chain stores, as well as the "people, merchandise, and space" full-chain management, providing chain enterprises with a Level 3 to Level 4 intelligent operation operating system. In terms of system architecture, the Company has integrated business decision-making applications with offline venue IT operation and maintenance. By unifying complex business flows and physical spaces under management, the system upgrades traditional operation models to an automated business foundation.

By deeply integrating the Company's algorithmic cognition, the system transforms unstructured data generated by storefronts into actionable digital models, offering closed-loop solutions covering intelligent marketing, smart stores, and intelligent supply chains. The technology-driven efficiency reshaping is evident in key operational areas:

At the single-store interaction and fulfillment level: The system builds a voice ordering system spanning App and in-vehicle terminals. The original order fulfillment Agent upgrades in-store passive anomaly intervention to minute-level proactive warnings intervention, achieving a 20% automated closed-loop in core processes and reducing customer complaint rates by 20%.
At the operational decision level: The intelligent data Agent shortens complex store site selection calculations from hours to minutes.
At the middle and back-office management level: The visual mid-office achieves a high degree of automated processing, reducing the workload for verification checks by 80%. The intelligent document Agent utilizes multimodal parsing technology to reduce the time for manual review of non-standard contracts by over 95%, cumulatively reducing client compliance operational costs by 70%. The private domain customer service system managed by agents helps benchmark clients achieve a 71% year-on-year reduction in labor costs.

The continuous evolution of the above-mentioned digital and intelligent capabilities provides core support for the scale expansion of the physical space in offline fields. In 2025, as the service reach extended to multiple sectors such as footwear and apparel, and automotive 4S dealerships, the scale of entities managed by the system achieved significant growth. The expansion of management scale has created a virtuous and positive cycle with the reconstruction of the AI Native intelligent service system. Relying on the intelligent order dispatch engine, the underlying delivery platform of the system has completed the transformation from manual management to intelligent automated dispatch.

Looking forward to 2026, the smart store operating system will undergo continuous evolution, anchored by Agent standardization and decision-making automation. At the technical implementation level, the Company is deconstructing the complex operational processes and encapsulating them into flexible pluggable intelligent units. At the system architecture level, these standardized capabilities will be deeply integrated into the operational intelligence operating system, the spatial intelligence platform, and the service intelligence platform. By focusing on strengthening the synergy of the three major platforms, the Company will establish a digital closed loop from physical space management to the full-chain operation of stores. This comprehensive integration from the underlying modules to the top-level architecture, combined with the store knowledge base updated in near real time, will significantly reduce the marginal cost of penetrating into multi-format new retail scenarios. Ultimately, by continuously converting process assets and data assets into measurable financial returns, the Company will further consolidate its competitive advantage in the digital foundation of its chain business and release even greater profit margin expansion during the efficiency pricing cycle.

Agentic Services

As the capabilities of large models expand from information retrieval and text generation to understanding intent and autonomously executing complex workflows, the role of AI has substantially shifted from an auxiliary tool to a system-level execution unit. For quite some time, in professional service fields such as marketing and operations which are highly dependent on human resources, enterprises' revenue scale and human resource investment have exhibited a tight, structural correlation. The traditional model based on scaling through human labor hours has resulted in business growth being constrained by the marginal cost of personnel management. At the same time, the non-standardized nature of expert experience has also constituted a barrier to cross-industry expansion, often trapping enterprises in a single sector and making it difficult to achieve true industry-wide scale benefits.

The Company's development of Agentic Services business aims to address the aforementioned issues. The traditional hourly-based human resource billing model has been a core obstacle limiting the profitability of service providers. Leveraging DeepMiner's enterprise-grade trusted AI Agent platform, the Company transforms its accumulated data assets and industry knowledge into a trustworthy digital workforce that can be standardly scheduled and replicated across industries. This restructuring of productivity allows the Company to break free from limitations of the old billing model and shift towards performance-based pricing that delivers quantifiable business results.

19 -

The Company strategically chose the full marketing and transaction chain, encompassing front-end reach and back-end conversion, as the initial entry point for its Agentic Services business. This decision was not based on business inertia; the core logic lies in the fact that this chain possesses the prerequisites for establishing a viable AI business loop. On the one hand, high-frequency data interaction and quantifiable feedback metrics constitute a data flywheel for continuous optimization of Agents. On the other hand, the relatively clear attributability of the chain's results provides the necessary trust foundation for a pay-per-performance business agreement.

In specific business delivery, the Company has reshaped the traditional workflow by leveraging AI infrastructure:

Efficiency liberation: Algorithms improve the standardization of core operational processes, reduce the linear dependence of revenue scaling on headcount, and realize AI-driven orchestration from single-point optimization to overall business strategy.
Programmatic supply of dynamic content assets: Leveraging a multimodal generation architecture, the system has broken through the constraints of traditional physical production capacity at extremely low marginal costs. By dynamically recombining granular business insights with massive amounts of native materials, the system ensures a precise match between content and audience needs. These high-converting creative matrices not only directly increase the efficiency of the front-end marketing funnel, but the high-frequency interaction signals captured in real transaction environments also continuously flow back and form a high signal-to-noise ratio feedback loop for calibrating the underlying large model.
Quality verification: DeepMiner encapsulates data insights, business experience and strategic capabilities into professional judgments that can be delivered at scale. The system's decision interpretability and data isolation mechanisms, while increasing strategy conversion rates and compliance security, lay a foundation of trust for the pay-for-performance model.

In view of the above, services have gradually extended downstream, establishing an end-to-end pipeline from insight to conversion. This process has expanded the addressable market and created conditions for the Company to increase its proportion in the allocation of clients' overall budgets.

The Company's Agentic Services business has evolved into a scalable alternative to traditional marketing channels and has successfully expanded into the AI-native market. Specifically, it is divided into two areas:

20 -

(I) AI-Driven Reconstruction of Traditional Business Chains

Front-end insights and building omnichannel content influence: To address industry pain points such as the erosion of trust in traditional advertising and the difficulty of large-scale management of user-generated content (UGC), the Company launched the DOMO marketing intelligence product in 2025. Deeply integrated with the DeepMiner platform, this system combines historical marketing data with vertical industry knowledge bases to activate over 200,000 decentralized high-quality communication nodes. Using this as a foundational asset pool, AI can rapidly construct a thousand-person matrix (千人矩陣) tailored to different target demographics and dynamically generate a thousand-word matrix (千言矩陣) of customized content strategies. This systematic “thousand-person, thousand-words” (千人千言) architecture transforms fragmentary communications into precisely deployable brand assets. In practical application, DOMO has increased the efficiency of data insights and proof-of-concept (POC) solution output by 160% and 300%, respectively. Real-world data shows that the system not only supports the concurrent deployment of over 3,000 pieces of content per project but also drives a 40-fold increase in engagement metrics during trending marketing campaigns, while boosting search volume by 700% and maintaining a cost per engagement (CPE) 52% below the industry average. Within its first year of launch, the product was deployed across more than 70 commercial projects, serving six major vertical industries including food and beverage, beauty and personal care, healthcare, automotive, apparel and footwear, and maternal and infant products. The reliability of its delivery results has driven numerous leading brand clients to upgrade their partnerships from one-off trials to annual strategic frameworks, validating the high repurchase rate and scalability of this model.
Multimodal AIGC creative factory: Leveraging DeepMiner’s modeling of social media trends across domains, the Company connects the business chain from marketing insights to multimodal content production. By building an AI-native creative factory, the system breaks through the production capacity constraints of traditional physical photography and manual copywriting. On the production side, the platform can support the simultaneous production and distribution of over 10,000 articles and 10,000 images per day, and maintain a delivery frequency of over 100 streaming videos per week and over 50 creative videos per month. On the operation side, the system uses Agents to stably schedule and cover hundreds of high-quality social media account matrices on mainstream platforms. This high-efficiency production reveals clear boundaries for human efficiency: an average of over 800 pieces of text and image content per person per day, and refined management of 30 cross-platform accounts per person. Leveraging batch generation and real-scene fusion technologies, the platform has compressed delivery cycles from weekly to daily. It now provides AIGC commercial services to over 50 brand clients, with click-through rates (CTR) and cost per unique visitor (CPUV) for ultimate placement both outperforming industry averages. This efficiency breaks the linear relationship between content supply and manpower accumulation, and establishes a business model that meets the needs of massive asset volumes at a lower marginal cost.
21 -

Agent-based transaction chain and restructuring of the all-domain operation ecosystem: Addressing the highly manpower-dependent commercial delivery chain, the Company has deeply integrated the leading ecosystems of mainstream short-video e-commerce and social media seeding platforms, and leveraged DeepMiner for system transformation. On the short-video e-commerce transaction side, the system connects to the underlying data platform, compressing the review and decision-making from weekly updates to daily updates, which represents a seven-fold efficiency improvement. In the execution phase, the intelligent matching and parsing speed of media resources (KOLs/KOCs) has increased by 24 times and ten times, respectively, and daily video production capacity has increased by three times. In terms of social media content seeding and lead conversion, the system integrates cross-domain data, shortening the strategy generation cycle by seven times. The team, with only 1/5 staffing of its industry competitors, can maintain a high-frequency output, averaging 200 articles per day; relying on smart real-time monitoring and placement scheduling, benchmark clients achieved a 39-fold increase in lead generation and a 44% decrease in order costs within three days. This computing-power-driven productivity leap directly translates into an advantage for business expansion, which supports the addition of ten core brand customers to the business line in a single quarter, and verifies the feasibility of the "AI infrastructure + full-domain intelligent operations" model in reshaping the profitability elasticity of the heavy asset service sector.

(II) Independent Development of AI-native Emerging Businesses

Cognitive positioning of GEO: Among the first batch of service providers in China to enter this sector, the Company has adapted to the structural trend of information acquisition migrating to AI platforms to reshape the brand's cognitive system of the next generation of traffic entry points. Unlike traditional SEO's manual optimization, the Company relies on a comprehensive data foundation to build an intelligent closed loop of "capture-evaluation-optimization". Leveraging its proprietary data products, the Company can accurately capture the original needs expressed behind social media discussions, thereby reconstructing underlying content based on real consumer insights. By tracking the evolution of large-scale model algorithms in real time and matching it with strict compliance mechanisms, the Company has ensured the commercial credibility and asset security of GEO deliverables. Actual measurement metrics show that, for high-ROI target keyword sets, the GEO business helps brands achieve brand visibility of over 85% on the three major AI platforms, with an average ranking consistently within the top three. Simultaneously, by leveraging advanced content strategies and technologies, we assist clients in effectively navigate adverse misinformation while amplifying the reach of their core messaging. This fulfills their comprehensive objectives of deepening brand perception and building long-term trust. On the supply side, proprietary Agents have taken over the writing and precise distribution of AI-friendly content, which increases the labor efficiency of content production and distribution by ten times and two times respectively. At present, the business has covered leading brands in more than ten industries, and in January 2026 alone, it added annual strategic frameworks of more than ten core clients. Furthermore, the Company has simultaneously deployed its GEO business overseas, helping Chinese brands expanding internationally establish an early presence in the mainstream international AI search ecosystem.
22 -

Intelligent creation and placement of digital content: In Q4 of 2025, the Company entered the digital content field. In response to the business characteristics of this track that are highly dependent on materials production capacity and media bidding, the Company built an end-to-end AI production and placement pipeline. The self-developed AI content generation system has greatly compressed the traditional development cycle of several weeks, and can complete the entire process from creative proposal, outline construction to content plan output within 40 minutes. Subsequently, the Agent automatically takes over the underlying editing tools to enable 24/7 batch production of video materials, ensuring front-end capacity at extremely low marginal costs. This business directly leverages the Company's underlying capabilities in data analysis and marketing technology. Based on the system's accumulated placement models, the business deploys an intelligent operation monitoring Agent for automated bidding to capture conversion rate inflection points in real time and dynamically adjust placement strategies, thus significantly increasing the success rate of media purchasing decisions. Relying on the synergistic operation of the aforementioned technologies and data architecture, this new business achieved operational profitability, which initially validated the commercial viability and financial feasibility of the Company's AI technology in this sector within two months of its launch.
Quantitative screening and effects evaluation system for AIGC assets: In response to explosive growth in materials production capacity and rising trial and error costs brought about by AIGC technology, the Company launched the AI creative testing platform AdEff during the Reporting Period. This platform is based on a self-developed multimodal large language model and a Mixture-of-Experts (MoE) architecture, integrates over 100,000 neural measurement data from over 100,000 participants accumulated over more than a decade, and simulates consumers' subjective emotional feedback through quantitative calculations. On the application side, AdEff shortens the traditional research cycle to a mere 15 minutes to complete quantitative assessment and outputs optimization suggestions within seconds, helping clients intercept ineffective creatives before actual placement to avoid budget losses. Data validation shows that the scores of this system's predictions have an $89\%$ correlation with real-world test results, and the core evaluation dimensions match human expert expectations by $76\%$ . Currently, AdEff has been deployed at scale in multiple vertical industries globally, including FMCG, 3C and online retail. Collaborating with the Company's DeepMiner business Agent platform, it has created a complete business loop from content insight and creative generation to pre-effects testing, marking a strategic extension of the Company's business chain from "AI-assisted generation" to "AI-driven decision-making".

The commercial prerequisite for exporting AI productivity is the prior completion of organisational restructuring and workflow reorganisation internally. During the Reporting Period, the Company achieved $100\%$ adoption of the DeepMiner platform among all employees. The core objective of this restructuring is to empower employees with the ability to build scalable Agents. Leveraging the DM-Skill skill space within the platform, employees can independently create and deploy their own dedicated business Agents based on specific business scenarios. The mechanism has directly triggered a structural shift in internal working practices: in response to complex business processes, standard internal operating procedures have evolved from a one-way manual execution approach to human-AI collaborative orchestration. The first step for employees in resolving tasks has shifted from traditional manual processing to prioritising the use of DeepMiner. In this way, AI has moved beyond being merely a tool for improving efficiency and has effectively transformed into a digital colleague that participates in business logic processing. Through this internal practice, the DeepMiner platform has

evolved from a business-application-level product into a base organisational-level infrastructure that supports the daily operations of the Company.

The successful closed-loop integration of marketing and e-commerce scenarios has provided initial logical validation for the Company's Agentic Services business model. As the technological base continues to develop, the medium-long term strategic focus is shifting towards deeper penetration into broader commercial horizons. This transition will develop systematically, relying on the following three core pillars:

Evolution of Underlying Technologies: Focusing on system synergy and multimodal structures. The Company will continue to refine the DeepMiner base platform, focusing on enhancing the Cito model's deep cognitive capabilities based on cross-industry knowledge graphs, expanding the execution boundaries of the Mano model, and improving the scheduling efficiency of the Multi-Agent collaboration framework. The Company will maintain a high level of capital investment in AI infrastructure, leveraging system-level synergy to expand the processing capabilities of edge tasks and continue to solidify its foundational technology moat.
Market Expansion: Penetrating labour-intensive and expert-intensive sectors. Leveraging its strategic expertise in delivering end-to-end marketing solutions, the Company is systematically transferring its mature AI service capabilities to traditional industries, such as operation, finance, accounting, human resources and supply chain, which are heavily reliant on manual workflows and specialist expertise. By scale-deploying AI trusted workforce solutions to these sectors via an Agentic Services model, the Company is not only committed to eliminating inefficiencies within complex business chains, but also substantially expanding the accessible market across industries, thereby establishing a new growth engine to underpin the next phase of financial expansion.
Scalable Replication: Driving the transformation of traditional enterprises into Agent-based models. In response to the widespread challenges faced by traditional enterprises, namely fragmented underlying systems and data silos, the Company has developed a standardised methodology for Agent-based transformation. This methodology comprehensively covers key project phases such as use case mapping, metric alignment, data governance and compliance definition, and is supported by a suite of ready-to-use tools. This system significantly reduces the launch cycle for proprietary Agents in traditional enterprises. More crucially, the Company's proprietary compact models – engineered to support high-success-rate Agentic operations – are natively aligned with the stringent demands of traditional enterprises for data privacy and on-premises deployment. This fundamentally helps clients reduce capital expenditure on computing power, providing a robust guarantee of security and compliance at the physical infrastructure level.

Mininglamp Technology's long-term strategy transcends the boundaries of individual applications and specific industries, centring on the large-scale deployment of a trustworthy AI workforce across a wide range of business scenarios. To further empower the entire industry, the Company is opening up the core capabilities of DeepMiner to ecosystem partners, with a commitment to establishing the principles of "verifiable outcomes, auditable processes and knowledge accumulation" as universal guidelines for AI-native delivery across the industry. Within this open and collaborative network, and underpinned by commercial contracts based on results-based payment, the Company will continuously drive the integration of Agents from the foundational execution layer into the core decision-making hub of the enterprise, ultimately facilitating a substantive transformation of the physical industry towards a results-driven and intelligence-led model.

24 -

FINANCIAL REVIEW

The following table sets forth the comparative figures for the years ended 31 December 2025 and 2024:

	Year ended 31 December
	2025	2024
	RMB'000	RMB'000
Revenue	1,425,775	1,381,382
Cost of sales	(636,154)	(668,688)
Gross profit	789,621	712,694
Research and development expenses	(360,555)	(353,047)
Administrative expenses	(244,756)	(362,263)
Selling and marketing expenses	(175,087)	(127,299)
Impairment losses on financial assets and contract assets, net	(34,893)	(24,342)
Other operating income, net	9,949	21,910
Operating loss	(15,721)	(132,347)
Finance costs	(7,917)	(11,703)
Other income/(losses), net	25,613	(34,349)
Share of profits of joint ventures	274	384
Share of losses of associates	(953)	(104)
Fair value changes of preferred shares, warrants and convertible notes	(6,414,012)	185,989
(LOSS)/PROFIT BEFORE TAX	(6,412,716)	7,870
Income tax credit	42	79
(LOSS)/PROFIT FOR THE YEAR	(6,412,674)	7,949
Attributable to:
Owners of the parent	(6,414,969)	4,735
Non-controlling interests	2,295	3,214
	(6,412,674)	7,949
Non-HKFRS measure
Adjusted operating profit	24,984	580
Adjusted net profit/(loss)	42,043	(45,113)

25 -

Revenue

For the year ended 31 December 2025, our revenue increased by 3.2% year-on-year to RMB1,425.8 million. The following table sets forth a breakdown of our revenue by source, both in absolute amounts and as percentages of our total revenue, for the years ended 31 December 2025 and 2024:

	Year ended 31 December
	2025
RMB'000	%	2024
RMB'000	%
Data Intelligence services
Marketing intelligence services	718,156	50.4	730,853	52.9
Operational intelligence services	542,199	38.0	522,813	37.9
Subtotal	1,260,355	88.4	1,253,666	90.8
Agentic Services
Agentic marketing services	95,403	6.7	–	–
Others	4,821	0.3	–	–
Subtotal	100,224	7.0	–	–
Others	65,196	4.6	127,716	9.2
Total	1,425,775	100.0	1,381,382	100.0

For the year ended 31 December 2025, revenue from the Data Intelligence business increased by 0.5% year-on-year to RMB1,260.4 million, driven by growth in the operational intelligence business. The growth in the operational intelligence business derived from an increase of revenue of conversational intelligence products and smart store operating system. The increase of revenue from conversational intelligence products was driven by the product upgrade which effectively addressed customer requirements for real-time data access, coupled with steady expansion in our sales channels. The increase of revenue from smart store operating system is driven by enhanced AI capabilities and expanded store scenario coverage. The decrease in revenue of marketing intelligence service was primarily attributable to the transformation of certain service models within the social media management business. To better align with client needs, the Group has transitioned part of its services from traditional consultancy report delivery to end-to-end, full-chain agentic marketing services focused on outcome-based delivery.

In 2025, we launched our Agentic Services powered by Agentic AI. Leveraging our proprietary AI Agents, agentic marketing services focused on addressing the core needs of our clients' social media marketing, extending AI capabilities to a broader range of marketing functions, including planning and strategy development, content creation and execution, to help clients enhance marketing effectiveness and optimise costs. The service centered on the achievement of clients' key performance indicators as

26 -

its core delivery objective, driving the transformation of the service model from process-based output to outcome-driven delivery. Others primarily derived from AI products and services. For the year ended 31 December 2025, revenue from Agentic Services reached RMB100.2 million.

Other revenue derived from the industry solutions business. For the year ended 31 December 2025, industry solutions business revenue decreased by 49.0% year-on-year to RMB65.2 million, as we decided to phase out this business line in the second half of 2022, choosing not to take on new projects, with the exception of renewals for a few existing projects.

Cost of Sales

For the year ended 31 December 2025, cost of sales decreased by 4.9% year-on-year to RMB636.2 million.

Gross Profit and Gross Profit Margin

For the year ended 31 December 2025, our gross profit increased by 10.8% year-on-year to RMB789.6 million, and our gross profit margin increased from 51.6% in 2024 to 55.4% in 2025. This was primarily driven by the wider adoption of internally developed AI tools within the Data Intelligence business, which significantly boosted productivity whilst reducing the number of staff required for data processing and report delivery.

Research and Development Expenses

For the year ended 31 December 2025, research and development expenses increased by 2.1% year-on-year to RMB360.6 million, primarily due to an increase in our procurement of technical services.

Administrative Expenses

For the year ended 31 December 2025, administrative expenses decreased by 32.4% year-on-year to RMB244.8 million, primarily due to a decrease in administrative-related staff costs, particularly share-based remuneration.

Selling and Marketing Expenses

For the year ended 31 December 2025, selling and marketing expenses increased by 37.5% year-on-year to RMB175.1 million, primarily due to (1) the expansion of the sales team to drive revenue growth, resulting in higher employee benefit expenses; and (2) increased efforts in brand and product marketing, which led to higher marketing and promotional expenses.

Impairment Losses on Financial Assets and Contract Assets, Net

For the year ended 31 December 2025, net impairment losses on financial assets and contract assets increased by 43.3% year-on-year to RMB34.9 million, mainly due to the increased aging of receivables from industry solutions and the increase in revenue leading to a higher balance of trade receivables.

27 -

28 -

Other Operating Income, Net

For the year ended 31 December 2025, other operating income, net decreased by 54.6% year-on-year to RMB9.9 million, primarily due to the reversal of non-refundable long-term advances from discontinued projects within the industry solutions to other income in 2024, which did not occur in 2025.

Finance Costs

For the year ended 31 December 2025, finance costs decreased by 32.4% year-on-year to RMB7.9 million, primarily due to a reduction in interest-bearing bank and other borrowings.

Other Income/(Losses), Net

For the year ended 31 December 2025, we recorded other income, net of RMB25.6 million, compared with other losses, net of RMB34.3 million for the year ended 31 December 2024. This was primarily attributable to foreign exchange gains, net of RMB24.4 million for the year ended 31 December 2025, compared with foreign exchange losses, net of RMB16.8 million for the year ended 31 December 2024. This is driven by the translation of RMB denominated loans provided by our Company to our subsidiaries.

Fair Value Changes of Preferred Shares, Warrants and Convertible Notes

For the year ended 31 December 2025, a loss of RMB6,414.0 million was recorded in the fair value changes of preference shares, compared with a gain of RMB186.0 million recorded in the fair value changes of preference shares, warrants and convertible notes for the year ended 31 December 2024. This change was primarily attributable to an increase in the fair value of the preference share liability resulting from a rise in the actual conversion price (i.e. the offer price) compared to the previous value of the preference shares.

(Loss)/Profit for the Year

In 2025, the loss for the year amounted to RMB6,412.7 million, compared with a profit for the year of RMB7.9 million in 2024, primarily due to the fair value changes of preference shares, warrants and convertible notes.

Non-HKFRS Measure

To supplement our consolidated financial statements that are presented in accordance with HKFRS, we also use adjusted operating profit (non-HKFRS measure) and adjusted net profit/(loss) (non-HKFRS measure) as additional financial measures, which are not required by, or presented in accordance with, HKFRS. We believe that these non-HKFRS measures facilitate comparisons of operating performance from period to period and company to company by eliminating potential impact of items. We believe that these measures provide useful information to investors and others in understanding and evaluating our consolidated results of operations in the same manner as they help our management. However, our presentation of adjusted operating profit (non-HKFRS measure) and adjusted net profit/(loss) (non-HKFRS measure) may not be comparable to similarly titled measures presented by other companies. The use of such non-HKFRS measures has limitations as an analytical tool, and you should not consider them in isolation from, or as substitute for analysis of, our results of operations or financial condition as reported under HKFRS.

The following table set forth a reconciliation of the operating loss/adjusted operating profit and the (loss)/profit for the year/adjusted net profit/(loss) for the years indicated:

	Year ended 31 December
	2025	2024
	RMB'000	RMB'000
Operating loss	(15,721)	(132,347)
Add:
Share-based payment expenses	16,208	106,577
Listing expenses	24,497	26,350
Adjusted operating profit (non-HKFRS measure)	24,984	580
(Loss)/Profit for the year	(6,412,674)	7,949
Add:
Share-based payment expenses	16,208	106,577
Listing expenses	24,497	26,350
Fair value changes of preferred shares, warrants and convertible notes	6,414,012	(185,989)
Adjusted net profit/(loss) (non-HKFRS measure)	42,043	(45,113)

29 -

LIQUIDITY AND FINANCIAL RESOURCES

Previously, we primarily met our cash requirements through funds generated from business operations, bank loans and capital injections from shareholders. The Company was listed on the Main Board of the Stock Exchange on 3 November 2025, issuing 7,219,000 new shares at an offer price of HK$141.00 per share, with net proceeds of approximately HK$900.8 million after deducting underwriting commissions, fees and other expenses related to the Global Offering. In addition, following the exercise of the over-allotment option, the Company received the additional net proceeds of approximately HK$148.1 million (after deducting the underwriting commissions, fees and other expenses). As of 31 December 2025, our liquid funds amounted to RMB1,540.6 million, comprising cash and cash equivalents, pledged deposits, and restricted cash and time deposits. We believe that, given our business development and expansion plans, this level of liquidity is sufficient to fund our operations. As at 31 December 2025 and 31 December 2024, the Group's cash position was as follows:

	As at 31 December
	2025	2024
	RMB'000	RMB'000
Time deposits	18,615	13,570
Pledged deposits and restricted cash	139,884	147,677
Cash and cash equivalents	1,382,135	400,370
Total	1,540,634	561,617

The following table sets forth a summary of our cash flows for the years indicated:

	Year ended 31 December
	2025	2024
	RMB'000	RMB'000
Net cash generated from/(used in) operating activities	18,030	(27,917)
Net cash (used in)/generated from investing activities	(16,014)	20,639
Net cash generated from financing activities	992,855	87,245
Net increase in cash and cash equivalents	994,871	79,967
Cash and cash equivalents at the beginning of the year	400,370	294,915
Effect of foreign exchange rate changes, net	(13,106)	25,488
Cash and cash equivalents at the end of the year	1,382,135	400,370

Net cash generated from/(used in) operating activities

During the Reporting Period, net cash generated from operating activities was RMB18.0 million. The improvement in cash flow from operating activities was primarily attributable to the control of costs and expenses, and a further enhancement in profitability.

Net cash (used in)/generated from investing activities

During the Reporting Period, net cash used in investing activities was RMB16.0 million, primarily attributable to the purchase of property and equipment and the settlement of amounts payable for the acquisition of subsidiaries in prior years during the current year.

Net cash generated from financing activities

During the Reporting Period, net cash generated from financing activities was RMB992.9 million, primarily comprising proceeds received from the Global Offering (as defined in the Prospectus).

DEBT

As at 31 December 2025 and 31 December 2024, the Group’s debt position was as follows:

	As at 31 December
	2025	2024
	RMB’000	RMB’000
Interest-bearing bank borrowings	219,189	231,200
Lease liabilities	36,965	47,431
Preferred shares	–	7,816,400
Total	256,154	8,095,031

The Group maintains a prudent approach to cash management. As at 31 December 2025, among interest-bearing bank borrowings, RMB199.5 million was at a floating rate and RMB19.7 million was at a fixed rate. The Group has not entered into any interest rate swap contracts or other financial instruments to hedge against interest rate risk. The Group will continue to monitor interest rate risk and will consider hedging significant interest rate risk where necessary.

GEARING RATIOS

As at 31 December 2025, the Company recorded net cash and as such, the gearing ratio, which is calculated as net debt divided by the capital and net debt and expressed as a percentage, was not applicable (31 December 2024: 13.6%).

31 -

32 -

PLEDGED ASSETS

As at 31 December 2025, we did not pledge or charge any other assets except for the pledged deposits amounting to RMB131.1 million for bank borrowings.

SIGNIFICANT INVESTMENT, ACQUISITION AND DISPOSAL

As at 31 December 2025, the Group did not have any significant investments required to be disclosed pursuant to paragraph 32(4A) of Appendix D2 to the Listing Rules. During the Reporting Period, the Group did not have any significant acquisitions or disposals of subsidiaries, associates or joint ventures.

FUTURE PLANS FOR SIGNIFICANT INVESTMENTS AND CAPITAL ASSETS

As of 31 December 2025, the Group has no specific plans for any significant investments or acquisitions of capital assets.

FOREIGN CURRENCY RISK

The Group mainly operates in Mainland China with most of the Group's monetary assets, liabilities and transactions principally denominated in RMB, USD and HK$. In 2025, we have not used any derivative to hedge its exposure to foreign currency risk. Moreover, we will continue to monitor fluctuations in exchange rates and take the necessary measures to mitigate their impact.

CONTINGENT LIABILITIES

As at 31 December 2025, the Group had no material contingent liabilities.

CAPITAL COMMITMENT

As at 31 December 2025, commitments of the Group in respect of associates, joint ventures and financial assets at fair value through profit or loss amounted to RMB36.9 million (31 December 2024: RMB38.9 million).

USE OF PROCEEDS

On 3 November 2025, the Class A Shares have listed on the Main Board of the Stock Exchange. The net proceeds received by the Company from the Global Offering were approximately HK$900.8 million (after deducting the listing expenses). In addition, following the exercise of the over-allotment option, the Company received the additional net proceeds of approximately HK$148.1 million (after deducting the underwriting fees, commissions and expenses). As of the date of this announcement, there had been no change in the intended use of net proceeds as previously disclosed in the section headed "Future Plans and Use of Proceeds" in the Prospectus.

As of 31 December 2025, the Group had utilized the net proceeds as set out in the table below:

Intended use of the net proceeds	Net proceeds from the Global Offering (including the proceeds from the exercise of the over-allotment option) RMB'000	Percentage of net proceeds as stated in the Prospectus	Utilized net proceeds during the year ended 31 December 2025 RMB'000	Net proceeds unutilized as of 31 December 2025 RMB'000	Expected timeline of full utilization of the net proceeds
Technology R&D to enhance our technology R&D capabilities	335,207	35%	10,951	324,256	Before 31 December 2027
Product development to enrich our product portfolio	383,093	40%	9,175	373,918	Before 31 December 2027
Marketing, brand promotion and sales team expansion	143,660	15%	1,366	142,294	Before 31 December 2027
Working capital and general corporate purposes	95,773	10%	-	95,773	Before 31 December 2027
Total	957,733	100%	21,492	936,241

PURCHASE, SALE OR REDEMPTION OF THE COMPANY'S LISTED SECURITIES OR SALE OF TREASURY SHARES

From the Listing Date to 31 December 2025, neither the Company nor any of its subsidiaries purchased, sold or redeemed any of the Company's securities or sold any treasury shares (as defined under the Listing Rules). The Company did not hold any treasury shares as of 31 December 2025.

SIGNIFICANT EVENT AFTER THE REPORTING PERIOD

Save as disclosed elsewhere in this announcement, no other significant events occurred after the Reporting Period and up to the date of this announcement.

ANNUAL GENERAL MEETING

The annual general meeting (the "AGM") of the Company is scheduled to be held on Friday, 12 June 2026. A notice convening the AGM will be published and dispatched to the Shareholders in the manner required by the Listing Rules in due course.

CLOSURE OF THE REGISTER OF MEMBERS

The register of members of the Company will be closed from Tuesday, 9 June 2026 to Friday, 12 June 2026, both days inclusive, in order to determine the identity of the Shareholders who are entitled to attend and vote at the AGM, during which period no share transfers will be registered. To be eligible to attend and vote at the AGM, unregistered holders of shares must lodge all properly completed transfer forms accompanied by the relevant share certificates with the Company’s branch share registrar in Hong Kong, Tricor Investor Services Limited, at 17th Floor, Far East Finance Centre, 16 Harcourt Road, Hong Kong for registration not later than 4:30 p.m. on Monday, 8 June 2026. The record date for determining the Shareholders’ entitlement to attend and vote at the AGM is Friday, 12 June 2026.

CORPORATE GOVERNANCE

The Company believes that effective corporate governance is fundamental in safeguarding the interests of shareholders and other stakeholders and enhancing shareholder value, and is therefore committed to achieving and maintaining a high standard of corporate governance that best meets the needs and interests of the Group.

The Company has adopted the Corporate Governance Code as its own corporate governance code since the Listing Date. The Board considers that, during the period from the Listing Date to 31 December 2025, the Company has complied with all provisions of the Corporate Governance Code, with the exception of code provision C.2.1. Pursuant to code provision C.2.1, companies listed on the Stock Exchange are required to separate the roles of chairperson and chief executive officer, and such roles should not be held by the same individual. We do not have a separate chairperson and chief executive officer and Mr. Minghui Wu currently performs these two roles. The Board believes that vesting the roles of both chairperson and chief executive officer in the same person has the benefit of ensuring consistent leadership within the Group and enables more effective and efficient overall strategic planning for our Group. The Board considers that the balance of power and authority for the present arrangement will not be impaired, and this structure will enable our Company to make and implement decisions promptly and effectively.

The Company will continue to review its corporate governance practices to ensure ongoing compliance with the Corporate Governance Code, enhance its corporate governance standards, comply with increasingly stringent regulatory requirements and meet the rising expectations of shareholders and investors.

MODEL CODE FOR SECURITIES TRANSACTIONS BY DIRECTORS

The Company has adopted the Model Code as its code of conduct regarding securities transactions by the Company’s Directors and relevant employees. Having made specific enquiries with all the Directors, each of them has confirmed that they have complied with the required standards set out in the Model Code from the Listing Date to 31 December 2025. No incident of non-compliance with the Model Code was noted by the Company from the Listing Date to 31 December 2025.

The amended Corporate Governance Code effective from 1 July 2025 will apply to corporate governance reports and annual reports for financial years commencing on or after 1 July 2025. For the purpose of this announcement, the Company will refer to the Corporate Governance Code effective at that time.
34 -

35 -

AUDIT COMMITTEE

The Audit Committee comprises three independent non-executive Directors, namely Mr. Yunan Ren (chairperson), Mr. Hing Yuen Ho, and Mr. John Fei Zeng. The annual results of the Group for the year ended 31 December 2025 have been reviewed by the Audit Committee. The Audit Committee is of the opinion that the preparation of the financial information complies with the applicable accounting standards, the requirements of the Listing Rules and any other applicable laws, and that adequate disclosures have been made.

SCOPE OF WORK OF AUDITOR

The figures set out in this announcement relating to the consolidated statement of financial position, the consolidated statement of profit or loss and consolidated statement of comprehensive income, and the related notes of the Group for the year ended 31 December 2025 have been agreed by the Group's auditors, Ernst & Young, as being consistent with the amounts set out in the audited consolidated financial statements of the Group for the year. The work performed by Ernst & Young in this regard did not constitute an assurance engagement and consequently, no opinion or assurance conclusion has been expressed by Ernst & Young on this announcement.

SUFFICIENCY OF PUBLIC FLOAT

Based on the information that is publicly available to the Company and to the best knowledge of the Directors, the Company has satisfied the public float required under the Listing Rules from the Listing Date to the date of this results announcement.

PUBLICATION OF RESULTS ANNOUNCEMENT AND ANNUAL REPORT

This results announcement is published on the website of the Company (www.mininglamp.com) and the website of the Stock Exchange (www.hkexnews.hk). The annual report of the Company for the year ended 31 December 2025 will be dispatched to the shareholders in the manner in which the shareholders have selected to receive corporate communications and will be published on the websites of the Company and the Stock Exchange in due course.

FINAL DIVIDEND

The Board resolved not to recommend the payment of any final dividend for the year ended 31 December 2025.

36 -

CONSOLIDATED STATEMENT OF PROFIT OR LOSS

For the year ended 31 December 2025

| | Notes | 2025
RMB'000 | 2024
RMB'000 |
| --- | --- | --- | --- |
| Revenue | 4 | 1,425,775 | 1,381,382 |
| Cost of sales | | (636,154) | (668,688) |
| Gross profit | | 789,621 | 712,694 |
| Research and development expenses | | (360,555) | (353,047) |
| Administrative expenses | | (244,756) | (362,263) |
| Selling and marketing expenses | | (175,087) | (127,299) |
| Impairment losses on financial assets and contract assets, net | | (34,893) | (24,342) |
| Other operating income, net | | 9,949 | 21,910 |
| Operating loss | | (15,721) | (132,347) |
| Finance costs | | (7,917) | (11,703) |
| Other income/(losses), net | | 25,613 | (34,349) |
| Share of profits of joint ventures | | 274 | 384 |
| Share of losses of associates | | (953) | (104) |
| Fair value changes of preferred shares, warrants and convertible notes | | (6,414,012) | 185,989 |
| (LOSS)/PROFIT BEFORE TAX | 5 | (6,412,716) | 7,870 |
| Income tax credit | 6 | 42 | 79 |
| (LOSS)/PROFIT FOR THE YEAR | | (6,412,674) | 7,949 |
| Attributable to: | | | |
| Owners of the parent | | (6,414,969) | 4,735 |
| Non-controlling interests | | 2,295 | 3,214 |
| | | (6,412,674) | 7,949 |
| (LOSS)/EARNINGS PER SHARE
ATTRIBUTABLE TO ORDINARY
EQUITY HOLDERS OF THE PARENT | 8 | | |
| Basic (RMB) | | (137.41) | 0.18 |
| Diluted (RMB) | | (137.41) | (2.47) |

37 -

CONSOLIDATED STATEMENT OF COMPREHENSIVE INCOME

For the year ended 31 December 2025

| | 2025
RMB'000 | 2024
RMB'000 |
| --- | --- | --- |
| (LOSS)/PROFIT FOR THE YEAR | (6,412,674) | 7,949 |
| OTHER COMPREHENSIVE INCOME/(LOSS) | | |
| Other comprehensive income/(loss) that may be
reclassified to profit or loss in subsequent periods: | | |
| Exchange differences on translation of the Group’s subsidiaries | 76,623 | (43,511) |
| Net other comprehensive income/(loss) that may be
reclassified to profit or loss in subsequent periods | 76,623 | (43,511) |
| Other comprehensive income/(loss) that will not be
reclassified to profit or loss in subsequent periods: | | |
| Exchange differences on translation of the Company | 18,142 | (46,272) |
| Equity investments designated at fair value through other
comprehensive income: | | |
| Changes in fair value | 3,736 | (2,301) |
| Income tax effect | (588) | – |
| | 3,148 | (2,301) |
| Net other comprehensive income/(loss) that will not be
reclassified to profit or loss in subsequent periods | 21,290 | (48,573) |
| OTHER COMPREHENSIVE INCOME/(LOSS)
FOR THE YEAR, NET OF TAX | 97,913 | (92,084) |
| TOTAL COMPREHENSIVE LOSS FOR THE YEAR | (6,314,761) | (84,135) |
| Attributable to: | | |
| Owners of the parent | (6,317,056) | (87,349) |
| Non-controlling interests | 2,295 | 3,214 |

38 -

CONSOLIDATED STATEMENT OF FINANCIAL POSITION

As at 31 December 2025

	Notes	As at 31 December
		2025
RMB'000	2024
RMB'000
NON-CURRENT ASSETS
Property and equipment		24,127	26,483
Right-of-use assets		36,929	48,117
Goodwill		754,823	754,823
Other intangible assets		35,103	45,676
Investments in joint ventures		4,137	3,863
Investments in associates		2,773	1,583
Equity investments designated at fair value through other comprehensive income		14,883	11,147
Financial assets at fair value through profit or loss		116,731	127,224
Trade and bills receivables	9	20,362	-
Contract assets		1,363	2,985
Prepayments, other receivables and other assets		9,978	13,523
Deferred tax assets		83	85
Total non-current assets		1,021,292	1,035,509
CURRENT ASSETS
Inventories		87,055	141,574
Trade and bills receivables	9	637,354	547,354
Contract assets		3,287	854
Prepayments, other receivables and other assets		88,620	94,457
Financial assets at fair value through profit or loss		12,293	-
Time deposits		18,615	13,570
Pledged deposits and restricted cash		139,884	147,677
Cash and cash equivalents		1,382,135	400,370
Total current assets		2,369,243	1,345,856

CONSOLIDATED STATEMENT OF FINANCIAL POSITION (continued)

As at 31 December 2025

	Note	As at 31 December
		2025	2024
		RMB'000	RMB'000
CURRENT LIABILITIES
Trade and bills payables	10	273,822	193,749
Other payables and accruals		263,985	271,459
Contract liabilities		132,783	171,617
Interest-bearing bank borrowings		219,189	231,200
Lease liabilities		17,866	22,456
Tax payable		1,047	268
Preferred shares, warrants and convertible notes		-	7,816,400
Other liabilities		23,239	23,846
Total current liabilities		931,931	8,730,995
NET CURRENT ASSETS/(LIABILITIES)		1,437,312	(7,385,139)
TOTAL ASSETS LESS CURRENT LIABILITIES		2,458,604	(6,349,630)
NON-CURRENT LIABILITIES
Lease liabilities		19,099	24,975
Deferred tax liabilities		4,525	5,515
Other payables and accruals		13,300	19,844
Total non-current liabilities		36,924	50,334
Net assets/(liabilities)		2,421,680	(6,399,964)
EQUITY/(DEFICITS)
Share capital		1,019	178
Reserves		2,384,409	(6,434,083)
		2,385,428	(6,433,905)
Non-controlling interests		36,252	33,941
Total equity/(deficits)		2,421,680	(6,399,964)

39 -

NOTES TO THE CONSOLIDATED FINANCIAL STATEMENTS
For the year ended 31 December 2025

CORPORATE INFORMATION

The Company is a limited liability company incorporated in the Cayman Islands. The registered office of the Company is located at PO BOX 309, Ugland House, Grand Cayman, KY1-1104, Cayman Islands. The Company’s shares have been listed on the Stock Exchange since 3 November 2025.

The Company is an investment holding company. During the year, the Company’s subsidiaries registered in the PRC were principally engaged in the provision of data intelligence services, agentic services and other services.

2 ACCOUNTING POLICIES

2.1 BASIS OF PREPARATION

These financial statements have been prepared in accordance with HKFRS Accounting Standards (which include all Hong Kong Financial Reporting Standards, Hong Kong Accounting Standards (“HKASs”) and Interpretations) as issued by the Hong Kong Institute of Certified Public Accountants (“HKICPA”) and the disclosure requirements of the Hong Kong Companies Ordinance. They have been prepared under the historical cost convention, except for financial instruments at fair value through profit or loss (“FVPL”), equity investments designated at fair value through other comprehensive income (“FVOCI”), other liabilities and preferred shares, warrants and convertible notes, which have been measured at fair value.

Basis of consolidation

The consolidated financial statements include the financial statements of the Group for the year ended 31 December 2025. A subsidiary is an entity (including a structured entity), directly or indirectly, controlled by the Company. Control is achieved when the Group is exposed, or has rights, to variable returns from its involvement with the investee and has the ability to affect those returns through its power over the investee (i.e., existing rights that give the Group the current ability to direct the relevant activities of the investee).

Generally, there is a presumption that a majority of voting rights results in control. When the Company has, less than a majority of the voting or similar rights of an investee, the Group considers all relevant facts and circumstances in assessing whether it has power over an investee, including:

(a) the contractual arrangement with the other vote holders of the investee;
(b) rights arising from other contractual arrangements; and
(c) the Group’s voting rights and potential voting rights.

40 -

The financial statements of the subsidiaries are prepared for the same reporting period as the Company, using consistent accounting policies. The results of subsidiaries are consolidated from the date on which the Group obtains control, and continue to be consolidated until the date that such control ceases.

Profit or loss and each component of other comprehensive income are attributed to the owners of the parent of the Group and to the non-controlling interests, even if this results in the non-controlling interests having a deficit balance. All intra-group assets and liabilities, equity, income, expenses and cash flows relating to transactions between members of the Group are eliminated in full on consolidation.

The Group reassesses whether or not it controls an investee if facts and circumstances indicate that there are changes to one or more of the three elements of control described above. A change in the ownership interest of a subsidiary, without a loss of control, is accounted for as an equity transaction.

If the Group loses control over a subsidiary, it derecognises the related assets (including goodwill), liabilities, any non-controlling interest and exchange fluctuation reserve; and recognises the fair value of any investment retained and any resulting surplus or deficit in profit or loss. The Group's share of components previously recognised in other comprehensive income is reclassified to profit or loss or retained profits, as appropriate, on the same basis as would be required if the Group had directly disposed of the related assets or liabilities.

2.2 CHANGES IN ACCOUNTING POLICIES AND DISCLOSURES

The Group has adopted amendments to HKAS 21 Lack of Exchangeability for the first time for the current year's financial statements. The Group has not early adopted any other standard or amendment that has been issued but is not yet effective.

Amendments to HKAS 21 specify how an entity shall assess whether a currency is exchangeable into another currency and how it shall estimate a spot exchange rate at a measurement date when exchangeability is lacking. The amendments require disclosures of information that enable users of financial statements to understand the impact of a currency not being exchangeable. As the currencies that the Group had transacted in and the functional currencies of overseas subsidiaries and the Company for translation into the Group's presentation currency were exchangeable, the amendments did not have any impact on the Group's financial statements.

41 -

42 -

2.3 ISSUED BUT NOT YET EFFECTIVE HKFRS ACCOUNTING STANDARDS

The Group has not applied the following new and amended HKFRS Accounting Standards, that have been issued but are not yet effective, in these financial statements. The Group intends to apply these new and amended HKFRS Accounting Standards, if applicable, when they become effective.

HKFRS 18	Presentation and Disclosure in Financial Statements²
HKFRS 19 and its amendments	Subsidiaries without Public Accountability: Disclosures²
Amendments to HKFRS 9 and HKFRS 7	Amendments to the Classification and Measurement of Financial Instruments¹
Amendments to HKFRS 9 and HKFRS 7	Contracts Referencing Nature -dependent Electricity¹
Amendments to HKFRS 10 and HKAS 28	Sale or Contribution of Assets between an Investor and its Associate or Joint Venture³
Amendments to HKFRS 21	Translation to a Hyperinflationary Presentation Currency²
Annual Improvements to HKFRS Accounting Standards – Volume 11	Amendments to HKFRS 1, HKFRS 7, HKFRS 9, HKFRS 10 and HKAS 7¹

Effective for annual periods beginning on or after 1 January 2026
Effective for annual/reporting periods beginning on or after 1 January 2027
No mandatory effective date yet determined but available for adoption

The Group is in the process of making an assessment of the impact of these new and amended standards upon initial application. HKFRS 18 introduces new requirements for presentation within the statement of profit or loss, including specified totals and subtotals. Entities are required to classify all income and expenses within the statement of profit or loss into one of the five categories: operating, investing, financing, income taxes and discontinued operations and to present two new defined subtotals. It also requires disclosure of management-defined performance measures in a note and introduces new requirements for aggregation and disaggregation of financial information. The new requirements are expected to impact the Group’s presentation of the statement of profit or loss and disclosures of the Group’s financial performance. Except for IFRS 18, the directors of the Company anticipate that the application of these new and amended HKFRS Accounting Standards will have no material impact on the Group’s financial performance and financial position in the foreseeable future.

43 -

3. OPERATING SEGMENT INFORMATION

For management purposes, during the year, the Group has only one reportable operating segment, which is the provision of data intelligence services, agentic services and other services, because the Group's chief operating decision maker, who has been identified as the Chief Executive Officer ("CEO"), regularly reviews the consolidated results when making decisions about allocating resources and assessing performance of the Group as a whole. Since this is the only reportable operating segment of the Group, no further operating segment analysis thereof is presented.

Geographical information

(a) Revenue from external customers

During the year ended 31 December 2025, substantially all (2024: substantially all) of the Group's revenue derived from external customers were located in Chinese mainland.

(b) Non-current assets

As at 31 December 2025, all (2024: all) of the Group's non-current assets were located in Chinese mainland.

Information about major customers

During the year ended 31 December 2025, revenues from transactions with single external customers (including entities under common control with those customers) amounting to 10% or more of the Group's revenues are as follows:

| | 2025
RMB'000 | 2024
RMB'000 |
| --- | --- | --- |
| Customer A | 249,967 | 267,038 |

44 -

4. REVENUE

An analysis of revenue from contracts with customers and other sources is as follows:

| | 2025
RMB'000 | 2024
RMB'000 |
| --- | --- | --- |
| Revenue from contracts with customers: | | |
| Marketing intelligence services | 718,156 | 730,853 |
| Operational intelligence services | 528,909 | 506,948 |
| Data intelligence services | 1,247,065 | 1,237,801 |
| Agentic marketing services | 95,403 | – |
| Others | 4,821 | – |
| Agentic services | 100,224 | – |
| Others | 65,196 | 127,716 |
| Subtotal | 1,412,485 | 1,365,517 |
| Revenue from other sources: | | |
| Operational intelligence services-rental income | 13,290 | 15,865 |
| Total | 1,425,775 | 1,381,382 |

Disaggregation of the Group’s revenue from contracts with customers by the timing of revenue recognition is set out below:

| | 2025
RMB’000 | 2024
RMB’000 |
| --- | --- | --- |
| Transfer over time: | | |
| Marketing intelligence services | 428,064 | 434,075 |
| Operational intelligence services | 123,139 | 123,207 |
| Data intelligence services | 551,203 | 557,282 |
| Agentic marketing services | 91,476 | – |
| Others | 565 | – |
| Agentic services | 92,041 | – |
| Others | 118 | – |
| Subtotal | 643,362 | 557,282 |
| Transfer at a point in time: | | |
| Marketing intelligence services | 290,092 | 296,778 |
| Operational intelligence services | 405,770 | 383,741 |
| Data intelligence services | 695,862 | 680,519 |
| Agentic marketing services | 3,927 | – |
| Others | 4,256 | – |
| Agentic services | 8,183 | – |
| Others | 65,078 | 127,716 |
| Subtotal | 769,123 | 808,235 |
| Total | 1,412,485 | 1,365,517 |

The following table shows the amounts of revenue recognised during the years ended 31 December 2025 and 2024 that were included in the contract liabilities at the beginning of those periods:

| | 2025
RMB’000 | 2024
RMB’000 |
| --- | --- | --- |
| Data intelligence services | 90,771 | 103,994 |
| Marketing intelligence services | 49,507 | 42,910 |
| Operational intelligence services | 41,264 | 61,084 |
| Others | 43,417 | 72,745 |
| Total | 134,188 | 176,739 |

5. (LOSS)/PROFIT BEFORE TAX

The Group’s (loss)/profit before tax is arrived at after charging/(crediting):

| | 2025
RMB’000 | 2024
RMB’000 |
| --- | --- | --- |
| Cost of services provided | 535,097 | 505,266 |
| Depreciation of property and equipment | 11,634 | 26,834 |
| Depreciation of right-of-use assets | 23,524 | 29,849 |
| Amortisation of other intangible assets | 10,695 | 11,412 |
| Lease payments not included in the measurement of lease liabilities | 9,739 | 8,431 |
| Listing expense | 24,497 | 26,350 |
| Auditor’s remuneration | 3,350 | – |
| Employee benefit expense (excluding directors’ and chief executive’s remuneration): | | |
| Wages and salaries | 404,733 | 440,584 |
| Pension scheme contributions (defined contribution scheme) | 47,452 | 51,031 |
| Share-based payment expenses | 9,168 | 102,387 |
| Termination benefits | 15,254 | 28,695 |
| Total | 476,607 | 622,697 |
| Impairment losses/(reversal of impairment losses) on financial and contract assets, net | | |
| Trade and bills receivables | 30,634 | 26,967 |
| Financial assets included in prepayments, other receivables and other assets | 4,235 | (2,283) |
| Contract assets | 24 | (342) |
| Total | 34,893 | 24,342 |
| (Reversal of impairment)/impairment of inventories | (2,474) | 3,684 |
| Impairment of investments in associates and joint ventures | – | 1,811 |
| (Gain)/loss on disposal of property and equipment* | (394) | 1,205 |
| Loss on disposal of other intangible assets | 7 | – |
| Fair value losses on financial assets at fair value through profit or loss | 10,420 | 14,206 |
| Fair value losses/(gains) on financial liabilities at fair value through profit or loss | 6,413,405 | (173,492) |
| Loss/(gain) on termination of leases* | 33 | (1,745) |
| Government grant *** | (6,955) | (8,516) |
| Bank interest income | (10,227) | (10,649) |
| Foreign exchange (gains)/losses, net** | (24,435) | 16,818 |

46 -

This item is included in “Cost of sales” in the consolidated statements of profit or loss.
** These items are included in “Other operating income, net” in the consolidated statements of profit or loss.
*** These items are included in “Other income/(losses), net” in the consolidated statements of profit or loss.

Various government grants during the year were mainly attributable to the Group’s development in advanced technology. There are no unfulfilled conditions or contingencies relating to these government grants.

There are no forfeited contributions that may be used by the Group as the employer to reduce the existing level of contributions.

6. INCOME TAX

| | 2025
RMB’000 | 2024
RMB’000 |
| --- | --- | --- |
| Current: | | |
| Chinese mainland | | |
| Charge for the year | 1,477 | 1,813 |
| Over provision in prior year | - | (49) |
| Elsewhere | | |
| Charge for the year | 57 | - |
| Deferred | (1,576) | (1,843) |
| Total tax credit for the year | (42) | (79) |

The Group is subject to income tax on an entity basis on profits arising in or derived from the countries/jurisdictions in which members of the Group are domiciled and operate.

Hong Kong

No provision for Hong Kong profits tax has been made as the Group did not generate any assessable profits arising in Hong Kong during the current year (2024: nil). The Hong Kong profits tax rate during the current year was 16.5% (2024: 16.5%).

Chinese mainland

Pursuant to the Corporate Income Tax Law of the PRC and the respective regulations, the entities which operate in Chinese mainland are subject to corporate income tax at a rate of 25% on the taxable income. During the current year, several (2024: several) PRC subsidiaries were entitled to a preferential tax rate of 15% (2024: 15%) because they were regarded as a “High and New Technology Enterprise”. In addition, the Group’s certain subsidiaries operating in Chinese mainland were entitled to effective preferential tax rates of 5% for the year ended 31 December 2025 (2024: 5%), because they were regarded as “small-scaled minimal profit enterprises” with taxable income no more than RMB3,000,000.

DIVIDENDS

There was no dividend declared or paid by the Group during the current year (2024: nil).

(LOSS)/EARNINGS PER SHARE ATTRIBUTABLE TO ORDINARY EQUITY HOLDERS OF THE PARENT

The calculations of the basic (loss)/earnings per share amounts is based on the (loss)/profit for the year attributable to ordinary equity holders of the parent, and the weighted average number of ordinary shares of 46,683,468 outstanding during the years ended 31 December 2025 (2024: 26,505,990).

The calculation of the diluted (loss)/earnings per share amounts is based on the (loss)/profit for the year attributable to the ordinary equity holders of the Company, adjusted to reflect the fair value changes of the preferred shares. The weighted average number of ordinary shares used in the calculation is the number of ordinary shares outstanding during the current year, as used in the basic (loss)/earnings per share calculation, and the weighted average number of ordinary shares assumed to have been issued at no consideration on the deemed exercise or conversion of all dilutive potential ordinary shares into ordinary shares.

The calculations of basic and diluted (loss)/earnings per share are based on:

| | 2025
RMB'000 | 2024
RMB'000 |
| --- | --- | --- |
| (Loss)/Earnings | | |
| (Loss)/Profit attributable to ordinary equity holders of the Company, as used in the basic earnings per share calculation | (6,414,969) | 4,735 |
| Adjustment for fair value gains on the preferred shares and warrants | - | (290,158) |
| Loss attributable to ordinary equity holders of the Company before fair value gains on the preferred shares and warrants | (6,414,969) | (285,423) |
| | 2025 | 2024 |
| Shares | | |
| Weighted average number of ordinary shares outstanding used in the basic earnings per share calculation | 46,683,468 | 26,505,990 |
| Effect of dilution – weighted average number of ordinary shares: | | |
| Share options | - | - |
| Preferred shares and warrants | - | 89,003,838 |
| | 46,683,468 | 115,509,828 |

48 -

The share options were ignored in the calculation of diluted loss per share amounts for the years ended 31 December 2024 because they had anti-dilutive effects on the basic earnings per share amounts as evidenced by the potential decrease in diluted loss per share amounts when taking shares options into account in addition to the preferred shares. Accordingly, the diluted loss per share for the years ended 31 December 2024 only takes into account the impact of preferred shares and warrants.
The share options, preferred share, as well as fair value adjustment on the preferred shares and warrants, were ignored in the calculation of diluted loss per share amounts for the current year because they had anti-dilutive effects on the basic loss per share amounts as evidenced by the potential decrease in diluted loss per share amounts when taking shares options, preferred shares into account. Accordingly, the diluted loss per share for the current year is the same as the basic loss per share.

9. TRADE AND BILLS RECEIVABLES

	As at 31 December
	2025	2024
	RMB'000	RMB'000
Trade receivables	779,133	646,058
Impairment	(126,270)	(99,895)
	652,863	546,163
Bills receivable	4,853	1,191
Net carrying amount	657,716	547,354
Analysed into:
Current portion	637,354	547,354
Non-current portion	20,362	-

The Group's trading terms with its customers are mainly on credit. The credit period is generally one month to five months for major customers. Each customer has a maximum credit limit. The Group seeks to maintain strict control over its outstanding receivables and has a credit control system to minimise credit risk. Overdue balances are reviewed regularly by senior management. In view of the aforementioned and the fact that the Group's trade receivables relate to a large number of diversified customers, there is no significant concentration of credit risk. The Group does not hold any collateral or other credit enhancements over its trade receivable balances. Trade receivables are non-interest-bearing.

An ageing analysis of the trade receivables as at the end of the reporting period, based on the invoice date and net of loss allowance, is as follows:

	As at 31 December
	2025	2024
	RMB'000	RMB'000
Within 1 year	559,399	471,108
1 to 2 years	72,882	63,738
2 to 3 years	20,582	11,317
Total	652,863	546,163

10. TRADE AND BILLS PAYABLES

An ageing analysis of the trade and bills payables as at the end of the reporting period, based on the date of service received, is as follows:

	As at 31 December
	2025	2024
	RMB'000	RMB'000
Within 1 year	218,849	154,734
1 to 2 years	27,992	18,037
Over 2 years	26,981	20,978
Total	273,822	193,749

The trade and bills payables are non-interest-bearing and are normally settled of not more than 3 months.

11. EVENTS AFTER THE REPORTING PERIOD

No significant events have occurred in respect of any period subsequent to 31 December 2025.

DEFINITIONS

In this announcement, unless the context otherwise requires, the following terms shall have the following meanings.

"Board"
the board of Directors of the Company

"China" or "the PRC"
the People's Republic of China, and for the purposes of this document only, except where the context requires otherwise, excluding Hong Kong, the Macao Special Administrative Region of the People's Republic of China and Taiwan

"Audit Committee"
the audit committee of the Board

"Class A Shares"
class A ordinary shares of the share capital of the Company with a par value of US$0.001 each, conferring a holder of a class A ordinary share one vote per share on any resolution tabled at the Company's general meeting

"Class B Shares"
class B ordinary shares of the share capital of the Company with a par value of US$0.001 each, conferring weighted voting rights in the Company such that a holder of a class B ordinary share is entitled to ten votes per share on any resolution tabled at the Company's general meeting, save for resolutions with respect to any Reserved Matters, in which case they shall be entitled to one vote per share

"Company", "our Company", "the Company" or "Mininglamp"
Mininglamp Technology (formerly known as Leading Smart Holdings Limited), a business company incorporated under the laws of the BVI on 1 February 2010, and registered by way of continuation in the Cayman Islands on 15 January 2019 as an exempted company with limited liability under the laws of the Cayman Islands, the Class A Shares of which are listed and traded on the Main Board of the Hong Kong Stock Exchange

"Corporate Governance Code"
the Corporate Governance Code set out in Appendix C1 to the Listing Rules

"Director(s)"
the director(s) of our Company

"Global Offering"
the Hong Kong Public Offering and the International Offering

51 -

52 -

“Group,” “our Group,” “the Group,” “we,” “us,” or “our”
the Company and its subsidiaries

“HK” or “Hong Kong”
the Hong Kong Special Administrative Region of the People’s Republic of China

“Hong Kong dollars” or “HK dollars” or “HK$”
Hong Kong dollars, the lawful currency of Hong Kong

“Listing Date”
the date, being Monday, 3 November 2025, on which the Class A Shares are to be listed and on which dealings in the Class A Shares are to be first permitted to take place on the Stock Exchange

“Listing Rules”
the Rules Governing the Listing of Securities on The Stock Exchange of Hong Kong Limited, as amended, supplemented or otherwise modified from time to time

“Listing”
the listing of the Class A Shares on the Main Board

“Main Board”
the stock exchange (excluding the option market) operated by the Stock Exchange which is independent from and operates in parallel with the Growth Enterprise Market of the Stock Exchange

“Model Code”
the Model Code for Securities Transactions by Directors of Listed Issuers as set out in Appendix C3 to the Listing Rules

“Prospectus”
the prospectus published by the Company on 23 October 2025

“Reporting Period”
1 January 2025 to 31 December 2025

“RMB” or “Renminbi”
Renminbi, the lawful currency of China

“Share(s)”
the Class A Shares and Class B Shares in the share capital of the Company

“Shareholder(s)”
holder(s) of our Share(s)

“Stock Exchange”
The Stock Exchange of Hong Kong Limited

"subsidiary" or "subsidiaries" has the meaning ascribed to it in section 15 of the Companies Ordinance

"weighted voting right" has the meaning ascribed to it in the Listing Rules

"%" per cent

"AI" artificial intelligence

"Agent" agent

"Agentic AI" agentic artificial intelligence

"Agentic Services" agentic services

"Cito" the Group's proprietary large model

"Data Intelligence" data intelligence

"DeepMiner" the Group's proprietary AI agent platform

"Foundation Agent" the Group's proprietary foundational agent framework

"GEO" generative engine optimization

"IAB" The Interactive Advertising Bureau

"Mano" the Group's proprietary large model

"ROI" the abbreviation of Return on Investment, referring to the rate of return on investment.

Unless otherwise specified, in this announcement:

Certain amounts and percentage figures have been subject to rounding adjustments; accordingly, figures shown as totals in certain tables may not be an arithmetic aggregation of the figures preceding them; and
for ease of reference, the names of Chinese laws and regulations, governmental authorities, institutions, natural persons or other entities (including certain of our subsidiaries) have been included in this announcement in both the Chinese and English languages and in the event of any inconsistency, the Chinese versions shall prevail. English translations of company names and other terms from the Chinese language are provided for identification purposes only.
53 -

By order of the Board
Mininglamp Technology
Mr. Minghui Wu
Chairman of the Board and Executive Director

Hong Kong, 26 March 2026

As at the date of this announcement, the Board comprises: (i) Mr. Minghui Wu, Mr. Ping Jiang, Ms. Jie Zhao and Mr. Qi Yu as executive Directors; (ii) Mr. Leiwen Yao as non-executive Director; and (iii) Mr. Yunan Ren, Mr. Hing Yuen Ho and Mr. Qingfei Zeng as independent non-executive Directors.

54 -