South Korea’s science ministry pursues a strategic shift towards open AI computing infrastructure, aiming to challenge GPU dominance with local NPU development and international collaborations.
South Korea’s science ministry has stepped up its push to build an open AI computing ecosystem, holding talks with domestic chipmakers and software firms on how to strengthen the country’s position in a market increasingly shaped by inference workloads, energy efficiency and system-level integration. At the centre of the discussion was the challenge of turning local neural processing units into commercially useful products in data centres and AI services, rather than leaving the field dominated by GPU-led platforms.
According to the ministry, the meeting in central Seoul followed a memorandum of understanding signed in July with AMD, after Deputy Prime Minister and science minister Bae Kyunghoon met AMD chair Lisa Su. The ministry said the aim was to examine how South Korea can broaden participation in the global AI compute stack and build out an open infrastructure that combines CPUs, GPUs, NPUs, memory, networking and software more flexibly.
FuriosaAI outlined global trends in open AI computing and the response required from the domestic chip sector. AMD Korea then presented the main provisions of the ministry partnership and the planned workstreams. MangoBoost and Moreh described their cooperation with AMD on software and performance optimisation, underscoring a broader industry view that the market is moving away from single-chip rivalry towards platform competition across hardware and software layers.
Participants also argued that next-generation technologies such as CXL and DPU will be important in shaping open AI infrastructure, alongside longer-term investments in PIM and neuromorphic chips. They said South Korea will need open interfaces, international standards and open-source compatible technologies, backed by national test environments that can prove those systems work in practice. The ministry said AMD’s Ryzen AI Halo computer, donated to mark the July agreement, was formally handed over to KAIST, following a proposal from Bae that the equipment should be used by promising researchers. In comments relayed by the ministry, Bae said the shift from training to inference is widening the contest beyond chip performance to power and cost efficiency, and that support for full-stack demonstrations would be expanded so local NPU makers can secure real-world deployment cases as quickly as possible.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
