Hanwha Aerospace, LIG Nex1: "Domestic Chips Are Essential"

"To Validate, We Need a Platform Like CUDA"

Paek Joungho, CEO of Furiosa AI: "Standards Should Be Led by Companies, Not the Government"

On the 18th, participants are discussing at the Defense AI Innovation Network seminar held at Modu Lab Gangnam Campus in Gangnam-gu, Seoul. Photo by Paek Jongmin, Tech Specialist

On the 18th, participants are discussing at the Defense AI Innovation Network seminar held at Modu Lab Gangnam Campus in Gangnam-gu, Seoul. Photo by Paek Jongmin, Tech Specialist

View original image

Diverging opinions have emerged between buyers and suppliers regarding standards related to neural processing units (NPUs), which have rapidly risen to prominence as a core defense artificial intelligence (AI) semiconductor.


While the defense industry—which produces weapons systems—argued that a common software platform such as CUDA is necessary to utilize domestically produced AI chips, Paek Joungho, CEO of Furiosa AI and a leading figure in the NPU sector, countered that platforms should not be designed and imposed as a standard by the government, but rather led by chip companies within the ecosystem.


There was consensus on the need to foster NPUs as a domestic alternative to NVIDIA GPUs, but differences emerged over the practicality of establishing a “Korean-style CUDA” to support this effort.

Paek Joungho, CEO of Furiosa AI, is speaking at the Defense Artificial Intelligence Innovation Network seminar. Photo by Paek Jongmin, Tech Specialist

Paek Joungho, CEO of Furiosa AI, is speaking at the Defense Artificial Intelligence Innovation Network seminar. Photo by Paek Jongmin, Tech Specialist

View original image

At the “26-3rd Defense Artificial Intelligence Innovation Network Seminar” held on March 18, 2026 at Everyone's Lab Gangnam Campus in Gangnam-gu, Seoul, attendees including Lee Hyungjin, Section Chief at the Defense Acquisition Program Administration (DAPA); Paek Joungho, CEO of Furiosa AI; Shin Sungkyu, Vice President of Rebellions; Shin Dongjoo, CEO of Mobilint; Lee Jinwon, CTO of HyperAccel; Kim Jonghun, Leader at Naver Cloud; Seo Youngwoo, Senior Executive Vice President at Hanwha Aerospace; and Lee Seungyoung, CTO of LIG Nex1, shared their views on the introduction of domestic AI semiconductors in the defense sector.


There was unanimous agreement on the necessity of domestic AI semiconductors. Lee Hyungjin of DAPA explained that developing AI systems on foreign semiconductors and platforms could allow for domestic operation, but at the export stage, various controls and certification issues could arise, potentially requiring redevelopment.


DAPA stated that it began to fully recognize the importance of semiconductors in 2024, and in 2025, for the first time, secured a budget specifically for “defense semiconductors” to start research and development, and is also pursuing relevant legislation.

Lee Seungyoung, CTO of LIG Nex1, is speaking at the Defense Artificial Intelligence Innovation Network seminar panel discussion. Photo by Paek Jongmin, Tech Specialist

Lee Seungyoung, CTO of LIG Nex1, is speaking at the Defense Artificial Intelligence Innovation Network seminar panel discussion. Photo by Paek Jongmin, Tech Specialist

View original image

The defense industry also agreed on the need to reduce reliance on foreign semiconductors. Seo Youngwoo, Senior Executive Vice President at Hanwha Aerospace, said, “We want to use domestic AI software and hardware needed for defense operations, not foreign-made products,” emphasizing the need for a national effort to foster domestic NPUs and a domestic-based AI model ecosystem as alternatives to GPUs. Lee Seungyoung, CTO of LIG Nex1, the developer of the Cheongung interceptor missile, also stated that domestic defense companies are committed to introducing domestic NPUs into weapons systems to promote intelligence capabilities.


The point of contention lay in the next stage after hardware. CTO Lee argued that simply localizing the chip is not enough. He stressed the need for a CUDA-level software stack that can run universally across various vendors’ NPUs. If algorithms developed for a specific NPU must be re-verified every time they are transferred to another chip, the characteristic reliability and certification procedures of the defense industry would inevitably slow down development and application. He pointed out that even with domestic chips, without a common platform, practical application to weapons systems would be difficult.


In response, Paek Joungho, CEO of Furiosa AI and a leader in the NPU field, presented a different view. He drew a clear line against the approach of the government setting and driving platform standards. Paek stated, “It is completely unfeasible for the government to suddenly create and drive standards. Chip companies must drive this,” citing CUDA’s success as an example. He added, “There has never been a case where a third party creates and drives a platform standard. That is something that is difficult for a government to do.”


When moderator Sim Seungbae, Principal Research Fellow at the Korea Institute for Defense Analyses (KIDA), asked if this meant a “winner-takes-all” scenario, Paek replied, “The essence of platform competition is not very different from that,” expressing confidence that platform standards would naturally develop through market competition.


NPU company representatives at the discussion each presented their views. Shin Sungkyu, Vice President at Rebellions, diagnosed that in defense AI semiconductors, the core issues are not the technology itself but the supply chain and legal and regulatory matters. Shin Dongjoo, CEO of Mobilint, introduced that the company will begin mass production this year of on-device AI semiconductors and on-premises chips for drones and robots, warning that Korea does not have much time to remain competitive in the AI and defense semiconductor sectors. Lee Jinwon, CTO of HyperAccel, presented a newly developed chip and stated that as long as the requirements are clear, it is entirely possible to create optimized chips tailored to defense needs.



◆Glossary: CUDA is a parallel processing platform and development ecosystem created to run AI and high-performance computing software exclusively on NVIDIA GPUs. It is considered a decisive factor that makes it difficult for NVIDIA GPU users to switch to other GPUs.


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

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing