MangoBoost Teams Up With AMD to Challenge Nvidia in AI Inference

AMD Chair and CEO Lisa Su speaks at an AI semiconductor industry roundtable held at the Four Seasons Hotel in Jongno-gu, Seoul, on Oct. 7, 2026. South Korean AI infrastructure startup MangoBoost is expanding its collaboration with Advanced Micro Devices (AMD) as it seeks to challenge Nvidia’s dominance in the artificial intelligence market through an integrated approach to hardware, software and AI inference services. AMD Chair and CEO Lisa Su has highlighted the startup as an example of AI infrastructure development in South Korea, underscoring the growing relationship between the two companies. Kim Jang-woo, CEO of MangoBoost and a professor at Seoul National University, unveiled the company's new AI inference platform, Mango Inference, during a press conference at the Grand InterContinental Seoul Parnas in Gangnam, Seoul, on Oct. 12. “MangoBoost is no longer simply a component company,” Kim said, announcing the company's transformation into an AI factory operator that designs and integrates hardware and software to deliver AI services directly to customers. Mango Inference is a serverless inference service that enables businesses and developers to access AI models using a single API key without building their own server infrastructure. Users can select open-weight models, including DeepSeek and GLM, and pay according to usage. The company said its proprietary inference optimization software, LLMBooost, reduces bottlenecks in data transfers between storage devices and GPUs, lowering costs by at least 20% compared with existing high-performance services. A central element of MangoBoost's expansion strategy is its close collaboration with AMD, which is competing with Nvidia in the AI GPU market. While AMD has been working to strengthen its networking and software ecosystem, MangoBoost contributes system-level design and optimization capabilities. The companies' joint technical achievements have also been featured on AMD's official blog. During her visit to South Korea on Oct. 7, Su cited MangoBoost as a notable example of domestic AI infrastructure development and maintained close communication with Kim, according to the company. Kim said the relationship has also strengthened MangoBoost's ability to secure GPU supplies. He explained that the company has established a roadmap for obtaining a stable supply of AMD's latest GPUs and has drawn up plans to procure tens of thousands of units. To improve service reliability and operational efficiency, MangoBoost is building a 1-megawatt (MW) AI infrastructure facility near Dosan Park in Seoul's Gangnam district. The facility is already partially operational, with AMD's latest Instinct-series GPUs being introduced in phases. Full completion is targeted for the first quarter of 2027. By increasing the proportion of internally designed systems, MangoBoost aims to reduce its reliance on external cloud providers and improve overall operating efficiency. The company is also targeting the sovereign AI market, seeking to provide domestic infrastructure that allows organizations to process sensitive AI data within South Korea rather than relying on overseas servers. Kwon Dong-up, MangoBoost's chief technology officer, emphasized the importance of managing AI data used in South Korea within the country. He said the company plans to expand its supply of on-premises AI systems to security-sensitive sectors, including finance and defense. Building on its Seoul facility, MangoBoost plans to work with partners in the United States, Japan and Southeast Asia to expand its infrastructure capacity to as much as 500MW over the longer term. The initial expansion will be financed through the company's own resources, while subsequent projects could involve joint ventures with major corporations to achieve economies of scale. Kim said MangoBoost's ability to optimize hardware and software together would be central to its efforts to compete with Nvidia's established AI computing ecosystem.