OpenAI ropes in Samsung, SK Hynix to source memory chips for Stargate (techcrunch.com)

OpenAI announced letters of intent with Samsung Electronics and SK Hynix to produce and supply up to 900,000 high-bandwidth memory (HBM) DRAM chips per month for its Stargate AI infrastructure and to help build AI data centers in South Korea. The deals followed a Seoul meeting between OpenAI CEO Sam Altman, South Korea’s president and executives from both chipmakers; SK Group said the output would be more than double current industry HBM capacity. The agreements also include integrating ChatGPT Enterprise and OpenAI APIs into Samsung and SK operations and a separate pact with SK Telecom to explore local data-center development. Stargate is a large-scale partnership with Oracle and SoftBank targeting roughly $500 billion in AI-dedicated data-center investment. For the AI/ML community this addresses a critical hardware bottleneck: HBM capacity is essential for training and running large transformer models, and a sustained increase in supply could accelerate model scaling, reduce training bottlenecks, and lower costs for massive GPU clusters. The move complements recent compute commitments — Nvidia’s up-to-$100B investment and multi-gigawatt capacity access, Oracle’s large compute sale, and OpenAI’s plans for multi-gigawatt datacenters — and signals strategic geographic diversification of AI compute. Increased HBM production in South Korea could shorten supply chains, boost regional AI ecosystems, and materially affect how quickly organizations can build and iterate on next-generation large models.