{"id":107422,"date":"2026-08-20T12:18:10","date_gmt":"2026-08-20T12:18:10","guid":{"rendered":"https:\/\/www.europesays.com\/britain\/107422\/"},"modified":"2026-08-20T12:18:10","modified_gmt":"2026-08-20T12:18:10","slug":"bae-kyung-hoon-accelerates-open-full-stack-push-with-amd-and-south-korean-ai-firms-proposes-han-river-chicken-and-beer-gathering-biggo-finance","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/britain\/107422\/","title":{"rendered":"Bae Kyung-hoon Accelerates &#8216;Open Full-Stack&#8217; Push with AMD and South Korean AI Firms, Proposes Han River Chicken-and-Beer Gathering \u2014 BigGo Finance"},"content":{"rendered":"<p>South Korea&#8217;s Ministry of Science and ICT held a roundtable on &#8220;Building an Open AI Computing Infrastructure Ecosystem&#8221; on August 20 at the HJ Business Center in Gwanghwamun, Seoul, where it discussed cooperation measures with domestic AI semiconductor and software companies to break away from the Nvidia-centric GPU monopoly structure.<\/p>\n<p>At the meeting, Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon proposed to South Korean AI company executives, &#8220;Let&#8217;s meet by the Han River over chicken and beer in 10 years,&#8221; signaling the government&#8217;s long-term commitment to support. This is interpreted as a move to build close partnerships between the government, industry, and academia\u2014similar to how Nvidia CEO Jensen Huang communicates informally with ecosystem stakeholders during his visits to South Korea.<\/p>\n<p>The roundtable was organized as a follow-up to the memorandum of understanding (MOU) signed between the Ministry of Science and ICT and AMD in July. As the global AI market shifts from training to inference-centric workloads, building an open heterogeneous computing ecosystem that flexibly combines CPUs, GPUs, NPUs, DPUs, CXL, and storage was addressed as the core agenda item.<\/p>\n<p>AMD and South Korean NPU Integration, Demonstration Targeted by 2027<\/p>\n<p>The government announced plans to develop and demonstrate heterogeneous computing systems combining AMD processors with South Korean-made NPUs by 2027. Participating companies include FuriosaAI, HyperExcel, DeepX, Mobilint, MangoBoost, Panesia, Moreh AI, and Nota AI\u2014South Korean AI semiconductor, infrastructure, and software firms.<\/p>\n<p>Moon Kyung-sun, AMD Korea managing director, said, &#8220;AMD Chair Lisa Su has designated South Korea as a priority country for national AI collaboration,&#8221; adding that &#8220;AMD will concentrate all technical resources at the headquarters level, including establishing Asia&#8217;s first AI Center of Excellence (CoE).&#8221; He also acknowledged that while AMD Korea&#8217;s engineering headcount has more than tripled in recent years, it remains insufficient.<\/p>\n<p>Moon referenced the precedent where Intel significantly expanded pricing and supply benefits when Samsung Electronics increased AMD&#8217;s share in PC CPU adoption, emphasizing that &#8220;in an Nvidia-dominated market, we must leverage open collaboration with AMD to establish a strategic position that secures tangible benefits from both global big tech players.&#8221;<\/p>\n<p>Companies Demand Technical Support, CPU Supply, and Software Independence<\/p>\n<p>Company executives attending the roundtable made specific policy requests based on their field experience.<\/p>\n<p>Lee Joong-yeon, CEO of KTNF, pointed out the lack of AMD&#8217;s technical support infrastructure in South Korea. KTNF, the country&#8217;s only server motherboard manufacturer, stressed that &#8220;AMD Korea&#8217;s organization and Asia-Pacific (APAC) technical support infrastructure are extremely weak, making communication possible only through Taiwanese engineers,&#8221; and that &#8220;full technical support guarantees for domestic motherboard and system development must be secured at the Lisa Su level.&#8221;<\/p>\n<p>Shin Jung-kyu, CEO of Lablup, warned of a potential CPU supply crunch in the second half of the year. With the expansion of virtual machine (VM) usage for agentic AI, he said &#8220;the biggest future bottleneck will occur in CPUs, not GPUs,&#8221; and that &#8220;R&amp;D capabilities must be concentrated on domestic development of agentic AI-specific CPUs tailored to environments requiring more than 20 times the computational resources compared to humans.&#8221;<\/p>\n<p>Chae Myung-soo, CEO of Nota AI, called for guaranteed independent scope for software companies. He said, &#8220;In the second-phase Dokpamo evaluation, we lightweighted Upstage&#8217;s model to one-quarter its size and released it on Hugging Face, resulting in four times more downloads than the original model.&#8221; He emphasized that &#8220;the role of software optimization companies must be clearly guaranteed at the policy level so they can build the most affordable and efficient services without being tied to the logic of any particular camp, whether Nvidia or AMD.&#8221;<\/p>\n<p>Kim Hong-seok, Chief Science Architect (CSA) at Rebellions, assessed that &#8220;as hardware integration and AI coding adoption lower software stack development costs, vertical integration and closed ecosystems centered on global big tech are strengthening,&#8221; and that &#8220;a government-led open standards ecosystem is a survival requirement for South Korean semiconductor companies.&#8221;<\/p>\n<p>Agentic AI Era: Resolving Infrastructure Bottlenecks Is Key<\/p>\n<p>The discussion focused intensively on infrastructure bottlenecks arising from the spread of agentic AI and corresponding strategies.<\/p>\n<p>Lee Jin-won, CTO of HyperExcel, explained that &#8220;in agentic workloads, AI-to-AI calls account for 90% of usage, and a heavy-tail phenomenon occurs where a small number of heavy users consume most resources.&#8221; He added that &#8220;since the exploding key-value (KV) cache during computation cannot be indiscriminately loaded onto high-bandwidth memory (HBM) alone, system orchestration linked with storage and networking is essential.&#8221;<\/p>\n<p>Nam Yi-hyun, CEO of FADU, said, &#8220;Current AI infrastructure discussions are excessively focused on NPU computation, but in the agentic era, bottlenecks are shifting to memory and storage.&#8221; He proposed that &#8220;open infrastructure projects should function as testbeds for domestic companies to build practical validation track records, given that Qual testing for global big tech supply takes more than three years.&#8221;<\/p>\n<p>Chung Moo-kyung, CEO of Dinotisia, argued that &#8220;in the agentic era, more than 50% of CPU workloads are concentrated on data retrieval,&#8221; and that &#8220;we must build dedicated semiconductors that directly accelerate in-memory database searches and full-stack storage systems in an AI storage environment dominated by foreign systems.&#8221;<\/p>\n<p>Cho Yong-jin, CPO of Panesia, emphasized the importance of interconnect standards between heterogeneous hardware, saying &#8220;Panesia has been developing CXL standards and UALink technology since its founding, and will collaborate with domestic companies to build a connectivity ecosystem that optimizes token costs.&#8221;<\/p>\n<p>South Korean NPUs Accelerate Global Market Entry<\/p>\n<p>Song Jun-ho, CTO of DeepX, shared the company&#8217;s global export achievements and requested government support. DeepX has achieved exports worth $13 million (approximately 18 billion won) across 77 global customers in 10 countries, with a corporate valuation of 3.5 trillion won (approximately $2.5 billion). Song recommended that the government &#8220;actively support securing mass production yield for the DX-M2 chip based on Samsung Electronics&#8217; 2-nanometer process, memory supply, and expansion of overseas export channels through diplomatic avenues.&#8221;<\/p>\n<p>Deputy Prime Minister Bae emphasized, &#8220;South Korean NPU companies must prove their capabilities on the world stage beyond the Korean market,&#8221; and that &#8220;the government will actively support K-semiconductor overseas expansion by building an AI full-stack ecosystem.&#8221;<\/p>\n<p>He also stated, &#8220;Both Dokpamo and domestic NPUs have already reached their technical target levels,&#8221; adding that &#8220;rather than being buried in benchmark metrics or simple spec comparisons, we will boldly open public and private market opportunities from a real-world usage perspective to complete the demand and supply system.&#8221;<\/p>\n<p>Concluding the roundtable, Bae said, &#8220;Amid this great transformation encompassing AI data centers and physical AI, the government and domestic companies are creating miracles as one team.&#8221; He added, &#8220;Beyond simple R&amp;D, we will spare no effort in providing full-stack, all-round support so that domestic companies can practically dominate the global market and leap forward as unicorn companies.&#8221;<\/p>\n<p>The government is expanding next year&#8217;s AI-related budget and preparing comprehensive support to enable domestic AI models, software, and South Korean-made AI semiconductor companies to execute real business. At year-end, it plans to accelerate the spread of AI adoption in South Korea through the &#8220;AI for All&#8221; project.<\/p>\n","protected":false},"excerpt":{"rendered":"South Korea&#8217;s Ministry of Science and ICT held a roundtable on &#8220;Building an Open AI Computing Infrastructure Ecosystem&#8221;&hellip;\n","protected":false},"author":2,"featured_media":107423,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_share_on_mastodon":"0"},"categories":[21224],"tags":[16917,21234,26102,8160,44591,44590,44593,44594,26103,44595,5359,44592,44596,34887],"class_list":["post-107422","post","type-post","status-publish","format-standard","has-post-thumbnail","category-bae","tag-amd","tag-bae","tag-bae-kyung-hoon","tag-bae-systems","tag-deepx","tag-furiosaai","tag-ktnf","tag-lablup","tag-ministry-of-science-and-ict","tag-nota-ai","tag-nvidia","tag-panesia","tag-rebellions","tag-samsung-electronics"],"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@UnitedKingdom\/117127800621290299","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/posts\/107422","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/comments?post=107422"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/posts\/107422\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/media\/107423"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/media?parent=107422"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/categories?post=107422"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/britain\/wp-json\/wp\/v2\/tags?post=107422"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}