MangoBoost, a startup on the verge of achieving unicorn status, is entering the artificial intelligence (AI) data center infrastructure business. The company has set out to solve the data bottleneck problem plaguing the AI industry, with its data processing unit (DPU) at the core of its strategy.
According to industry sources on the 29th, MangoBoost recently signed a lease agreement for a portion of a colocation data center in Gangnam-gu, Seoul. The company plans to install 10 racks in the 42U standard equipped with AMD's latest graphics processing unit (GPU), the MI355X. A 42U rack is a large-format rack measuring 2 meters in height and 1 meter in depth. The racks MangoBoost installs will house not only AMD's GPUs but also MangoBoost's proprietary DPU, "BoostX." Companies using these servers will also have access to MangoBoost's data processing optimization software, "LLM Boost."
MangoBoost's package combining GPU, DPU and optimization software will be offered to companies developing AI services. A range of companies from large conglomerates to startups are currently in discussions with MangoBoost over server usage. These companies are particularly interested in the computational efficiency gains when running large language models (LLMs) using MangoBoost's package. MangoBoost plans to complete the facility build-out by May and launch full-scale operations.
Since its founding in 2022, MangoBoost has focused on selling DPU products. A DPU is a type of system semiconductor — an accelerator used to resolve data bottleneck issues by optimizing the operation of equipment such as central processing units (CPUs), GPUs and solid-state drives (SSDs). AMD, the world's second-largest GPU company, has also shown interest in MangoBoost's DPU, and the two companies have conducted a proof-of-concept (PoC) test on AI server efficiency.
Starting with the Gangnam data center, MangoBoost plans to expand its business into data center infrastructure. The AI industry is currently struggling with data bottleneck problems. MangoBoost's strategy is to leverage the industry's demand for solving these issues as a growth catalyst. The company also recognizes that relying solely on standalone DPU sales would make it difficult to secure a leading position in the AI server market.
"MangoBoost's goal is to establish a standard computer system that fundamentally changes the efficiency of AI data centers," MangoBoost CEO Kim Jang-woo said. "The servers going into the Gangnam data center will be the first milestone proving how much MangoBoost's integrated solution can maximize the potential of GPUs and storage."
