MangoBoost, a startup developing data processing units (DPUs), has launched a platform for accessing artificial intelligence services. The company also said it will make "AI factories" its main business with backing from AMD.
Kim Jang-woo, chief executive of MangoBoost, outlined the strategy at a press briefing held on the 12th at the Grand InterContinental Seoul Parnas in Seoul's Gangnam district. The company unveiled Mango Inference, an AI inference platform that lets users call five large language models — GLM 5.2 and 5.3, MiniMax M3, Kimi K3 and DeepSeek V4 — through an application programming interface (API). MangoBoost said the platform delivers better AI performance per dollar than existing inference services because it runs on the company's own DPUs and its LLM inference optimization software, LLM Boost. "We have priced it tens of percent cheaper than competing services," Kim said.
MangoBoost also laid out a plan to reorganize its core business around AI factories. As the company defines it, an AI factory means directly operating AI computing servers it has designed and running Mango Inference on them. MangoBoost is first building a 1-megawatt data center by leasing part of a facility in Seoul's Gangnam district. After completing the Gangnam data center within the first quarter of next year, it plans to secure a combined 5 MW of data center capacity in markets including South Korea and the U.S. by the end of next year. The centers will house MangoBoost servers equipped with AMD graphics processing units (GPUs).
Kim signaled that AMD will take part directly and indirectly as a key partner in the AI factory business. "We have already set a roadmap to receive GPUs from AMD in the tens of thousands of units," he said. "Building a 5 MW data center is within MangoBoost's own financial means, and we have funding plans for projects beyond that." He added that the company is also drawing up a plan to install 500 MW of MangoBoost AI servers in the U.S., Japan and Southeast Asia.