MangoBoost, a South Korean AI data center company, is evolving its business strategy from designing data processing unit (DPU) semiconductors to offering full-stack AI integrated systems. CEO Jangwoo Kim emphasizes the company's ability to build sovereign AI data centers more cost-effectively than NVIDIA, positioning MangoBoost as a direct competitor in this space. The company collaborates with major Korean firms like Samsung Electronics and SK hynix and utilizes AMD's CPUs and GPUs, having proven the cost-effective performance of AMD products in its solutions.
MangoBoost's core technology includes its DPU chip, BoostX, which offloads network and data movement tasks from the CPU, and an AI inference software platform called LLM Boost. The company has introduced GPU server Alphonso and storage server Kesar, integrating these technologies. By building its own data centers with AMD MI355X GPUs, MangoBoost has validated its full-system performance, achieving 2.8x higher cost-per-throughput compared to NVIDIA H100 in MLPerf Inference 5.0, based on internal cost calculations.
Expanding its offerings, MangoBoost has commercially launched "Mango Inference," a serverless AI inference platform, allowing developers to use AI models via API without needing to manage GPUs and servers directly. The company anticipates significant growth, expecting $7.5 million (10 billion won) in revenue this year from complete server rack system sales, with a potential tenfold increase next year. MangoBoost has secured cumulative funding exceeding $100 million and employs over 125 engineers, with 75% dedicated to R&D across its global operations.
