ILUVATAR COREX (9903) 2025 Performance Report: Six highlights outline the new paradigm of high-quality growth of universal GPU.

On March 30th, ILUVATAR COREX (9903) officially released its first performance report after listing on the Hong Kong stock market, submitting an outstanding performance report: annual revenue of 1.034 billion yuan, a year-on-year increase of 91.6%; gross profit of 558 million yuan, a year-on-year surge of 110.5%; adjusted net loss significantly narrowed by 32.1%. This report not only shows impressive growth figures, but also demonstrates a high-quality growth system with coordinated efforts in technology, products, market, and finance. With the hardcore commercialization demonstration of "usable, user-friendly, large-scale use," it has led the AI industry from a parameter competition to an efficiency-driven key stage, achieving a comprehensive advancement from catching up to leading in the domestic general-purpose GPU field. First Recognition: Overall improvement in financial quality, highlighting high-quality growth prospects By 2025, ILUVATAR COREX's revenue scale and profit quality have both soared, with annual revenue of 1.034 billion yuan, a year-on-year increase of 91.6%, achieving a doubling growth; gross profit of 558 million yuan, a year-on-year increase of 110.5%, with gross profit growth rate significantly higher than revenue growth rate, gross profit margin increasing from 49.1% to 54.0%, fully demonstrating that core products have technological premium and pricing power, and profit efficiency continues to improve. The company adheres to high-intensity research and development investment, with R&D expenditure reaching 974 million yuan in the year, continuously intensifying chip architecture, software ecosystem, and cutting-edge technology breakthroughs; while ramping up R&D investment, adjusted net loss narrowed by 321 million yuan year-on-year to 4.38 billion yuan, highlighting operational efficiency. Additionally, the company's asset-liability ratio decreased from 59.1% to 39.8%, cash and cash equivalents increased to 1.505 billion yuan, financial structure continuously optimized, and risk resistance significantly strengthened. The high revenue growth, improved gross profit, narrowed losses, and stable financials collectively underpin high-quality growth prospects. Second Recognition: Training and Inference dual-track breakout, inference revenue surges by 238.2% The company has built a "training + inference" dual-wheel driving pattern, with both core product lines achieving rapid breakthroughs. The Tiange Training Series revenue reached 584 million yuan, a year-on-year growth of 116.7%, leading in the large-model training computing market; the Zhikai Inference Series revenue reached 339 million yuan, a staggering year-on-year surge of 238.2%, becoming the primary growth engine. With the surge in inference computing demand in areas such as AIGC, intelligent interaction, and large-scale model deployment, Zhikai series leverages deep software and hardware synergies to penetrate deeply into scenarios such as real-time financial risk control, medical image diagnosis, Internet AIGC content generation, and intelligent retail recognition, transforming technological advantages into tangible business value for customers. ILUVATAR COREX strategically positions itself in the inference field, leveraging its full-stack software ecosystem and scenario-adaptation capabilities to establish a first-mover advantage in the rapidly growing and commercially certain inference market, opening up long-term growth potential. Third Recognition: Comprehensive layout of cloud-edge-end, building a full-scenario computing fortress Facing the industry trend of AI computing accelerating from the cloud to the edge and terminal, ILUVATAR COREX has built core barriers through a comprehensive product layout. In 2025, the company's general GPU main business income reached 923 million yuan, a significant year-on-year increase of 149.6%, accounting for a high proportion of 89.3% of total revenue, showcasing strong growth and ongoing optimization of the core business structure. Among them, the Tiange series focuses on cloud-based large-model training to support enterprise and research institute's large-scale cluster demands; the Zhikai series covers cloud and edge inference, facilitating efficient deployment in multiple scenarios; during the reporting period, the Tongyang series was launched to provide low-power, high real-time computing power for scenarios such as Siasun Robot&Automation, smart terminals, and industrial equipment, leading to a complete stack product matrix layout embracing "cloud-edge-end." The full-spectrum coverage capability penetrates multiple layers of markets including cloud computing, edge computing, and smart terminals, forming an irreplicable comprehensive competitive barrier and laying a solid foundation for long-term high-quality development. Fourth Recognition: Over 1000 projects implemented, entering the mature "harvest period" of commercialization The value of hard technology is ultimately validated by the market. By the end of 2025, ILUVATAR COREX had served more than 340 industry clients, implementing over 1000 projects in core sectors of the national economy such as the internet, AI large models, research, finance, medical, education, and transportation, transitioning from single-point validation to commercial scale. Each implementation validates the stability, compatibility, and service capabilities of the products, with a thousand scenarios fine-tuning the products to be "usable, easy to use, and stable to use," building a high customer stickiness and trust barrier. The more than 340 clients form a solid "community of trust," with continuous optimization of customer structure and expanding coverage in various industries, marking the company's commercialization entering a stage of scalable harvest, providing a reliable foundation for sustained growth. Fifth Recognition: Breakthroughs in core technology, leading the new era of AI inference The AI industry is entering a new stage of efficiency-driven development, with cost reduction and efficiency enhancement becoming the core demands. ILUVATAR COREX focuses on tackling industry pain points with a software-hardware synergistic full-stack optimization approach, focusing on key technology developments such as large-model inference PD separation, mixed-precision quantization, MoE training, etc., to comprehensively improve resource utilization and reduce customer TCO. The new generation software development platform has achieved over 80% efficiency improvement in code migration, with operators smoothly migrating in minutes; the self-developed core acceleration library has improved performance by over 20% compared to the previous version; the large-model special acceleration toolkit, while ensuring model accuracy, has improved computing efficiency by 60% and reduced cluster communication overhead by 30%. Core technology breakthroughs effectively address the efficiency bottleneck of large-model inference, supporting stable deployment of customer production clusters, with chip computing efficiency reaching industry-leading levels, leading AI inference towards efficient development with technological strength. Sixth Recognition: Full-stack ecosystem formation, business model advancing towards high-level growth Software ecosystem is crucial for unlocking hardware value and building long-term barriers. Relying on the DeepSpark open-source community, ILUVATAR COREX has adapted over 610 mainstream algorithm models, deeply compatible with global mainstream AI frameworks and inference engines, possessing forward-looking adaptation capabilities that support the release of new models and operators on Day0, seamlessly matching the rapid iteration pace of large models. The open collaborative ecosystem significantly reduces customer migration thresholds, attracting a wide range of ISVs, research institutions, and industry partners to participate, forming a positive cycle of ecosystem synergy. Leveraging its ecological advantages, the company is evolving from a single chip supplier to a full-stack AI computing service provider integrating "chip + software + solutions," expanding towards high-margin and sustainable directions such as software licensing, technical services, and continuously optimizing its revenue structure, breaking through traditional hardware valuation constraints, and unlocking long-term value growth space. Conclusion ILUVATAR COREX's 2025 performance report is a benchmark for high-quality growth. The company does not rely on single hit products or capital narratives, but rather on a tri-dimensional linkage of product layout, technological innovation, and commercial landing to build a robust, sustainable, and highly scalable growth flywheel. This performance, while not final, clearly marks a key turning point in the development of domestic general-purpose GPUs a hard-tech company is confirming that technological innovation can be efficiently transformed into commercial value and long-term competitiveness through solid financial and operational performance. In the era of AI computing boom, ILUVATAR COREX's high-quality growth paradigm sets a new benchmark for the industry, worth the focused attention and long-term optimism of the capital market.