OpenAI just locked down the memory supply chain for its ambitious Stargate project, striking deals with Samsung and SK Hynix to produce 900,000 high-bandwidth DRAM chips monthly. The agreements, signed during a high-level Seoul summit with South Korea's president, more than double current industry capacity and cement OpenAI's hardware supply for its $500 billion AI infrastructure push.
OpenAI isn't just building the future of AI - it's securing the hardware supply chain to power it. The ChatGPT maker announced Wednesday it has struck agreements with memory giants Samsung Electronics and SK Hynix to manufacture DRAM wafers for the massive Stargate AI infrastructure project.
The deals came together during a Seoul summit between OpenAI CEO Sam Altman, South Korean President Lee Jae-myung, Samsung's executive chairman Jay Y. Lee, and SK chairman Chey Tae-won. It's the kind of government-level meeting that signals just how crucial memory chips have become in the AI arms race.
Under the agreements, Samsung and SK Hynix will scale production to churn out up to 900,000 high-bandwidth memory DRAM chips monthly specifically for Stargate and AI data centers. According to SK Group's statement, this represents more than double the current global industry capacity for high-bandwidth memory chips.
The timing couldn't be more critical. Stargate is OpenAI's $500 billion joint venture with Oracle and SoftBank to build AI-dedicated data centers across the United States. But even the most powerful GPUs are useless without the high-speed memory to feed them data.
"OpenAI is leaving few stones unturned in the race to build compute capacity for its AI efforts," the company noted in announcing the Korean partnerships. The statement reveals just how aggressive OpenAI has become in securing its hardware supply chain ahead of competitors.
Wednesday's memory chip deals cap off what can only be described as a frenetic month of infrastructure investment. Just two weeks ago, Nvidia announced plans to invest up to $100 billion in OpenAI, giving the AI company access to over 10 gigawatts of compute capacity through Nvidia's training systems. The following day, OpenAI revealed it would build five new Stargate data centers with SoftBank and Oracle, targeting 7 gigawatts of total compute capacity.
Earlier in September, Oracle agreed to sell $300 billion worth of compute capacity to OpenAI over five years - a deal that initially caught Wall Street by surprise.
But the Korean agreements go beyond just chip supply. OpenAI is also working with South Korea's Ministry of Science and ICT to identify locations for AI data centers outside Seoul. The company has struck a separate deal with SK Telecom to build an AI data center, plus additional agreements with Samsung subsidiaries to explore more data center opportunities across the country.
The partnerships include integration components too - both Samsung and SK Group will integrate ChatGPT Enterprise and OpenAI APIs into their operations as part of the broader deal.
For the Korean memory giants, the OpenAI partnership represents a massive revenue opportunity at a time when the memory market has been volatile. High-bandwidth memory chips command premium pricing, and guaranteed volume from OpenAI's infrastructure buildout provides revenue visibility that investors crave.
The geopolitical implications are significant too. By securing memory supply from Korean manufacturers, OpenAI reduces dependence on other regions while strengthening ties with a key technology ally. South Korea has been positioning itself as a critical player in the global AI hardware supply chain, and these deals cement that strategy.
What makes these agreements particularly strategic is their timing. As AI models continue growing in size and complexity, memory bandwidth has become just as critical as raw compute power. The 900,000 monthly chip target isn't just about meeting current needs - it's about ensuring OpenAI has the memory capacity to train next-generation models that could dwarf today's systems.
OpenAI's Korean memory chip partnerships represent more than just supply chain security - they're a bet on AI's explosive growth requiring unprecedented hardware capacity. By locking in 900,000 monthly DRAM chips and doubling industry capacity, OpenAI isn't just building for today's models but preparing for AI systems we can barely imagine. The real question isn't whether this massive infrastructure investment will pay off, but whether it will be enough to maintain OpenAI's lead as every tech giant races to secure their own AI hardware supply chains.
