Korean Startup That Caught AMD's Lisa Su's Eye Takes on NVIDIA's Dominance

MangoBoost launches Mango Inference, an AI inference service powered by its own hardware
CEO Kim Jangwoo: "No problem securing AMD GPUs"... Roadmap in place to obtain tens of thousands of units
Open-weight model service set to compete with neoclouds focused on GPU supply
Expanding use of Korean AI models and NPUs... Plans to build 500-MW infrastructure

MangoBoost CEO Kim Jangwoo explains “Mango Inference” at a press conference held on the 12th. Photo by Paek Jongmin, Tech Specialist.

South Korean AI infrastructure company MangoBoost is joining forces with AMD to challenge Nvidia's dominance. Its edge is technology that turns AMD chips into AI services customers can use right away. The company has an ambitious plan to use a stable supply of AMD semiconductors as a springboard to expand its offerings to include Korean AI models and reduce the Korean AI industry's dependence on Nvidia.

Kim Jangwoo, a Seoul National University professor and CEO of MangoBoost, unveiled its AI inference service, Mango Inference, at a press conference on October 12, declaring, "MangoBoost is no longer a component company." He said the company would move beyond data processing units (DPUs), which handle network and storage-related tasks, and become a provider of AI inference services to end customers.

Mango Inference is an AI inference service that jointly optimizes hardware and software. It is designed to let customers use AI models without building their own servers, while increasing response speeds and throughput and reducing operating costs.

As neocloud companies such as CoreWeave and Lambda grow, MangoBoost is distinguishing itself by combining AMD graphics processing units (GPUs) with its own server and network designs and inference software.

Customers can choose from open-weight models such as DeepSeek, GLM and MiniMax without building their own infrastructure. Businesses pay based on the number of tokens they use, without having to purchase GPUs or install models.

AMD is chasing Nvidia in the AI GPU market, but closing the gap is proving difficult. To use AI, networks that connect GPUs, storage devices and software must all work together, and this is an area where AMD is weak.

MangoBoost is focusing on designing and optimizing this entire system. AMD even introduced its joint technology achievements and Mango Inference in an official blog post on October 9. "We have to succeed to break Nvidia's dominance," Kim said confidently.

According to Kim, AMD Chair and CEO Lisa Su, who visited South Korea on October 7, cited MangoBoost as a key example of expanding AI infrastructure in the country. Kim underscored their close relationship by recounting how he stood right beside Su for a commemorative photo at the time.

Kim seemed particularly disappointed that a one-on-one meeting with Su, which had been planned, fell through after her schedule changed in response to a call from U.S. President Donald Trump.

Kim said the benefits of the partnership are also evident in GPU procurement. GPUs are difficult even for major operators to secure, but Kim stressed, "We can get supplies when other companies can't."

He said MangoBoost can receive a stable supply of AMD's latest GPUs and has also mapped out a plan to secure tens of thousands of units. The company's technology, which expands the range of applications for AMD GPUs, in turn gives it leverage in negotiations to secure semiconductors.

AMD Chair and CEO Lisa Su poses for a commemorative photo after meeting with Korean semiconductor startups in Seoul on the 7th. From left: Ryu Jemyeong, Second Vice Minister of Science and ICT; Lisa Su, Chair and CEO of AMD; and Kim Jangwoo, CEO of MangoBoost. Photo courtesy of Kim Jangwoo.

Kim said MangoBoost plans to expand support for Mango Inference to include Korean AI models. He also said the company would use Korean neural processing units (NPUs).

Even a convenient service loses its competitive edge if it is too expensive. Kim said he was confident that joint hardware and software optimization would give the company room to lower service fees.

MangoBoost plans to complete a 1-megawatt (MW) infrastructure facility it is building in Gangnam, Seoul, in the first quarter of next year. It is also considering plans to expand capacity to 500 MW in partnership with companies in the United States, Japan and Southeast Asia.

Kim also outlined a plan to fund the initial expansion with the company's own resources, then pursue joint ventures with large corporations and collaborative infrastructure development with overseas companies.

This content was produced with the assistance of AI translation services.