OpenAI And Samsung Move From Memory Supply To Building Chips Together

OpenAI is working with Samsung Electronics on the research and production of next-generation AI chips, widening a relationship that already covers memory supply and enterprise use.
- Companies: OpenAI and Samsung Electronics
- Scope: joint research and production of next-generation AI chips
- Disclosed: 9 September, at a press conference in Seoul
- Existing ties: memory for the Stargate data centre project, and ChatGPT across Samsung's workforce
What Has Been Confirmed
Harrison Kim, general manager of OpenAI Korea, said joint research and production of next-generation chips was among the areas where the two companies had progressed furthest. He did not detail the chips or set out Samsung's precise role, and the company has not said whether Samsung will manufacture, supply components or memory, or take some other part.
Why OpenAI Keeps Moving Down The Stack
In June the company unveiled Jalapeno, its first custom AI chip, built with Broadcom and designed by OpenAI's engineers for inference, the stage at which a model processes data to produce a response. Taiwan Semiconductor Manufacturing Company was named as manufacturer. A Samsung collaboration would give OpenAI a position in a different part of the supply chain.
The pattern is consistent: as models get more capable, the chips and memory needed to train and run them become a competitive constraint rather than a procurement line. Securing manufacturing capacity and advanced memory is becoming as decisive as designing the processor.
What Came Before
Last year Samsung Electronics and SK Hynix signed letters of intent to supply memory chips for Stargate, OpenAI's large-scale computing infrastructure programme. Kim said demand for advanced memory would keep rising as AI systems need faster and more complex computing, making cooperation with South Korean semiconductor companies a priority for OpenAI's local operation.
Why This Reaches Marketing
Because this is the layer underneath the tools the industry now runs on. Generation, editing, planning and increasingly media buying all sit on inference capacity, and inference capacity is decided by exactly these agreements. Memory has become a bottleneck for AI infrastructure, with high-performance systems needing large quantities of it alongside the processors. More platform news.
