[The Future of K-Semiconductors ③] MangoBoost, AMD's Key Partner Challenging NVIDIA

Kim Jangwoo: "NVIDIA's True Weapon Is Not a Single Chip, but the Connection System"
Korea's Path Lies in 'AI Infrastructure Independence'
Must Become a Platform Company That Dominates the System
Solving Data Input/Output Bottlenecks to Unlock the

Kim Jangwoo, CEO of MangoBoost, poses for an interview with The Asia Business Daily at the office in Sadang-dong, Seoul. Photo by Paek Jongmin, Tech Specialist

In the race to win the artificial intelligence (AI) war, governments and corporations are pouring astronomical budgets into securing Nvidia’s graphics processing units (GPUs) or, as an alternative, developing neural processing units (NPUs).

However, there is one figure who asserts that such strategies alone are not enough to secure victory in the AI war. That is Kim Jangwoo, a professor at the Department of Electrical and Computer Engineering at Seoul National University and CEO of the data processing unit (DPU) development startup MangoBoost.

Throughout his interview with The Asia Business Daily, Kim repeatedly emphasized that “Nvidia’s true weapon is not just a single chip, but the system that connects those chips,” highlighting the need for Korea to pursue a new path toward ‘AI infrastructure independence.’

After meeting Kim at the Supercomputing 2025 (SC25) event in St. Louis, USA, in November last year, we met again in early December at the MangoBoost office in Sadang-dong, Seoul. His conviction that Korea must become a platform company that dominates the entire system, rather than just selling components, had only grown stronger as the new year began.

According to Kim, the biggest weakness of Korea’s system semiconductor industry is “fragmentation.” While world-class fabless companies designing NPUs have emerged, there is a lack of ‘system architecture’ to smoothly control data when those chips are actually installed in data centers. Kim defined this as the role MangoBoost aims to fulfill.

“No matter how good an engine (NPU) you build, you can’t achieve speed if the road (network) is unpaved. MangoBoost is a company that lays the ‘expressway’ to resolve data input/output and network bottlenecks, allowing domestic NPUs to perform at 100% of their potential.”

Kim has recently been strengthening strategic collaborations with leading domestic accelerator companies such as FuriosaAI and HyperAccel. Rather than a simple memorandum of understanding (MOU), the strategy is to physically integrate MangoBoost’s DPU technology with domestic NPUs to build a ‘K-AI infrastructure full stack’ that can be immediately deployed in data centers without relying on foreign equipment.

“Nvidia’s strength is not the GPU chip itself, but the powerful network connectivity it secured through the acquisition of Mellanox. While our NPU companies focus on chip performance, MangoBoost will support system-level optimization so that even when dozens or hundreds of those chips are grouped together, there is no performance degradation. Only by moving as ‘One Team’ can we overcome the massive fortress that is Nvidia.”

◇ Nvidia’s Next-Generation ‘Vera Rubin’ Storage DPU Mirrors MangoBoost = Ironically, MangoBoost’s technological foresight was validated by Nvidia’s latest moves. Kim was surprised when he reviewed the detailed specifications of Nvidia’s next-generation ‘Vera Rubin’ architecture, unveiled at the world’s largest electronics and IT exhibition, CES 2026. Nvidia introduced a DPU concept with significantly enhanced storage functions to alleviate data bottlenecks handled by GPUs, and the structure and functions were similar to MangoBoost’s products.

MangoBoost is currently supplying this technology to AMD, which is chasing Nvidia. In a twist, Nvidia has adopted a secret weapon that was originally meant to help AMD catch up. Despite having powerful GPUs (the MI300 series), AMD had struggled to catch up with Nvidia due to a lack of software optimization and networking technology. MangoBoost precisely targeted this gap and became AMD’s partner.

“AMD’s GPUs are excellent on paper, but when connected to actual data centers, software issues often prevented them from reaching their full performance. However, when we attached our ‘Mango LLMBoost’ software and DPU to AMD systems, the performance rose to a level comparable to Nvidia. Since the price is much more competitive than Nvidia’s systems while performance is equal, customers naturally welcomed it.” Currently, AMD features MangoBoost as a core partner on its official blog and technical white papers.

Currently, out of MangoBoost’s total staff of about 140, more than 110 are based in Korea. The core team consists of Kim Jangwoo’s students from Seoul National University, as well as top Korean talents from Samsung Electronics, Google, and Nvidia. Some of the world’s best talents, who receive annual salary offers of 500 million to 1 billion won from Silicon Valley companies, have chosen to stay in Korea, united by the goal of proving themselves in the system semiconductor field where Korea has traditionally been weak. The U.S. corporation serves only as a forward base for global business and large-scale investment, while the heart of MangoBoost beats in Korea.