[News] OpenAI’s Altman Visits South Korea, Taiwan Amid Chip Push, Fueling Samsung Foundry Speculation

What happened

According to Seoul Economic Daily, OpenAI CEO Sam Altman visited South Korea for the first time in about a year and met Samsung Electronics executives to discuss AI cooperation; Richard Ho, OpenAI's vice president overseeing in-house AI chip development, accompanied him, fueling speculation about potential collaboration on next-generation AI accelerators. The report says OpenAI's in-house AI accelerator, Jalapeño, is expected to use Samsung's next-generation HBM4 memory, and that Samsung signed a memorandum of understanding with Broadcom in July covering up to US$200 billion in memory and foundry cooperation through 2030; with Broadcom developing custom AI chips for OpenAI, Samsung could also gain foundry and packaging opportunities (reported/speculative, not confirmed). Per Business Korea, Altman also met SK Group Chairman Chey Tae-won, with SK hynix and SK Telecom executives reportedly attending; details were not disclosed, but the discussions are believed to have covered existing cooperation on HBM supply and AI data center development in South Korea along with potential future collaboration. Per The Asia Business Daily, OpenAI's Stargate team visited South Korea ahead of Altman to discuss semiconductor supply volumes, contract terms and AI infrastructure development with local companies, including HBM supply talks with Samsung. As background, per TrendForce, HBM's blended average selling price is projected to rise 121% YoY in 2027, driven by supply constraints and a growing share of higher-priced HBM4. In Taiwan, per Commercial Times, Foxconn Chairman Young Liu confirmed meeting with Altman, and Reuters reports Foxconn is working on a new collaboration with OpenAI; sources cited by Commercial Times say Altman met TSMC Chairman C.C. Wei and engaged AI server suppliers including Quanta and Wistron. In June 2026, OpenAI and Broadcom unveiled their first custom AI inference chip, Jalapeño, which took nine months from initial design to tape-out; per OpenAI, the chip has undergone performance testing, with initial deployment planned by the end of 2026 and expansion to gigawatt-scale capacity in the coming years. Per Commercial Times citing supply chain sources, OpenAI's first-generation ASIC uses TSMC's 3nm process, while its second-generation chip is reportedly planned to adopt A16 with backside power delivery to improve performance and energy efficiency.

Impact and signal

Affected components and suppliers

Source and time

Source: TrendForce · 11/1/2026 · View original source