Core Insights - Alibaba's international cross-border e-commerce AI solution, Marco, won the SAIL Star Award at the 2025 World Artificial Intelligence Conference, marking it as the first recipient in this field [1] - Marco, developed by Alibaba's AI Business team, supports over 30 languages and covers more than 60 scenarios across the e-commerce value chain, providing solutions for marketing, compliance, and after-sales at a lower cost than mainstream large models [1] - The daily usage of Marco has reached 1 billion calls, a 1000-fold increase compared to 2023, and all capabilities are available for trial on Alibaba's AI open platform, Aidge [1] Technology and Application - In April 2023, Alibaba International Digital Commerce Group established a dedicated AI large model department, AI Business, to explore AI technology in global e-commerce scenarios [3] - All e-commerce platforms under Alibaba International have integrated AI solutions, with the most frequently used AI features being image translation, image recognition for product information enhancement, and marketing copy generation [3] - The translation feature transitioned from small models to large models by the end of last year, resulting in a 30% increase in consumer satisfaction in top European languages due to improved translation quality [3] - AI-driven SEO now accounts for nearly 40% of overall SEO efforts, with potential to reach 50% or more in the future [3] - External partners have seen a 23-fold increase in AI usage, with major cross-border e-commerce service providers integrating Alibaba's AI services [3] - The call volume for Alibaba's AI services doubles every two months, with an average daily call volume exceeding 1 billion as of July 2025, making cross-border e-commerce the first industry to achieve large-scale AI application [3] Technical Architecture - Alibaba International AI Business has developed a comprehensive AI solution for cross-border e-commerce, consisting of a five-layer technical architecture [4] - The infrastructure layer includes the MarsEngine inference engine, which optimizes token generation scheduling, KV cache, operator fusion, and computation, achieving a 116% improvement in extreme throughput compared to industry-leading engines [5] - The data layer has established a large-scale data collection, processing, and screening platform for model training, along with an evaluation framework for industrial applications [5] - The model layer offers multi-language enhanced Marco-LLM, e-commerce optimized translation model Marco-MT, self-developed multi-modal understanding model Ovis, and image generation model [5] - The product layer provides standardized capabilities for various business scenarios [5] - The AI solution layer delivers end-to-end AI capabilities or product collections for high-value scenarios [5]