TECHNOLOGY · AI · CYBERSECURITY · THE SYSTEMS SHAPING BUSINESS
AMD CEO Lisa Su met South Korean officials and executives from 13 Korean AI companies in Seoul Wednesday to discuss pairing AMD processors with Korean-designed AI chips, The Korea Herald reported. AMD also plans to open an AI Center of Excellence in the country by the end of the year.
What Happened
Su met Ryu Je-myung, the second vice minister of science and ICT, along with leaders from FuriosaAI, HyperAccel, DeepX, Mobilint, MangoBoost, Panmnesia, Dnotitia, Moreh, Nota AI, Lablup, Fadu, Elice Group and Upstage.
The talks centered on combining AMD's CPUs and GPUs with neural processing units, or NPUs, made by Korean companies to build systems for AI inference, the stage where trained models answer user requests. The two sides discussed demonstration projects meant to produce deployment blueprints that could be sold in overseas markets, and the Korean firms were encouraged to join AMD's open-source ROCm software ecosystem, its alternative to Nvidia's CUDA.
RelatedAMD will fund the planned Center of Excellence, and its location has not been decided. Su said the center would bring in hundreds of researchers; Seoul Economic Daily, which first reported the plan, quoted her as promising "hundreds of top-tier researchers."
AMD's first cloud infrastructure project in Korea will be with MangoBoost, and the company has also signed a cooperation agreement with Seoul National University's AI Institute. Both AMD and the ministry agreed to keep supporting graduate-level training in AI and AI transformation.
Su told the meeting she is a big believer in an open ecosystem and called Korea an important AI partner. Ryu said the global race to build AI infrastructure is moving beyond individual chips toward full systems and platforms, according to Seoul Economic Daily.
Why It Matters
Korea has a cluster of NPU startups, including FuriosaAI and DeepX, trying to sell inference chips as cheaper or more efficient alternatives to GPUs. Most need host processors, software and system partners to reach data center buyers. AMD offers all three, and a mixed-hardware approach lets it compete with Nvidia's tightly integrated systems by positioning itself as the open option that works with other vendors' accelerators.
For Korea, the arrangement fits a government push to build domestic AI computing capacity without relying entirely on imported GPU systems. The ministry's involvement and the demonstration projects aimed at foreign markets suggest Seoul sees AMD as a route for local chip designers to reach customers abroad.
Context
The Seoul stop follows Su's visit to Taiwan, where she said AMD would raise its $10 billion commitment to Taiwan's supply chain as demand outruns supply, as The Primary Wire reported Tuesday. AMD and Korea's science ministry signed a cooperation agreement in July during President Lee Jae Myung's visit to San Francisco, and Wednesday's meeting builds on that pact.
Su was also expected to meet Samsung Electronics Vice Chairman Jun Young-hyun, who heads the company's semiconductor business, during the trip, The Korea Herald reported.
What to Watch
The key details still missing are where the Center of Excellence will be based and how many researchers it will employ in its first phase. Any announced demonstration projects pairing AMD hardware with a specific Korean NPU would show whether the talks produce products. A meeting with Samsung could touch on memory supply and manufacturing ties, given AMD's need for high-bandwidth memory for its AI accelerators.
