By Protopian

Samsung and OpenAI are clearly getting closer. The less clear part is exactly how close.

On September 9, Harrison Kim, general manager of OpenAI Korea, told reporters in Seoul that one of the areas where the companies had made the most progress was “joint production and research” on next-generation chips. Reuters reported the remarks as evidence that OpenAI was deepening its semiconductor work with Samsung while developing its own processors.

That sounded like a significant expansion of the relationship. OpenAI already has its own custom inference processor, Jalapeño, designed with Broadcom and manufactured by TSMC, and has described the chip as the first step in a multi-generation hardware platform.

Two days later, however, OpenAI Korea narrowed the interpretation considerably.

In a statement reported by Yonhap on September 11, the company said Kim’s remarks referred to the existing relationship between OpenAI and Samsung, including Samsung’s adoption of ChatGPT and the companies’ earlier infrastructure agreement. It added that OpenAI had “not announced any plans on developing chips through Samsung”.

That does not mean there is no semiconductor work between the companies. It means the stronger claim — that Samsung is already a confirmed manufacturing or co-development partner for a future OpenAI processor — is not presently supported by an announced agreement.

The distinction matters because the existing partnership is already substantial enough without it.

In October 2025, Samsung and OpenAI announced a broad infrastructure partnership around Stargate. Samsung Electronics and SK hynix were to expand production of advanced memory for OpenAI’s data-centre build-out, with the companies targeting as much as 900,000 DRAM wafer starts per month as capacity expanded. Samsung’s own announcement described a wider collaboration spanning semiconductors, data centres, cloud services, shipbuilding and associated infrastructure.

Memory is not a peripheral part of modern AI hardware. High-bandwidth memory has become one of the critical constraints in building large accelerator systems, and Samsung is one of the few companies capable of supplying it at global scale. OpenAI can therefore have a strategically important chip relationship with Samsung without Samsung fabricating the logic processor itself.

The other half of the partnership runs in the opposite direction: OpenAI is now supplying software to Samsung.

In June, OpenAI announced that ChatGPT Enterprise and Codex would be made available to all Samsung Electronics employees in Korea and to all employees worldwide in Samsung’s Device eXperience division. OpenAI described it as one of its largest enterprise deployments, covering work from software development and manufacturing to marketing and product development.

But even here, the simple version of the story loses something.

Samsung is not turning itself into an OpenAI-only company. Its own June announcement said employees in the Device eXperience division would be given access to three external generative-AI systems: ChatGPT, Google Gemini and Anthropic Claude. Samsung said it deliberately chose not to restrict workers to a single provider, allowing them to select tools according to the task.

So two different forms of dependence are developing at once.

OpenAI increasingly depends on companies such as Samsung for the physical infrastructure required to build and run AI at enormous scale. Samsung, meanwhile, is increasingly embedding generative AI into ordinary corporate work — but is spreading that dependence across competing model providers.

That is a more interesting relationship than a conventional supplier agreement. Samsung can simultaneously be a customer, infrastructure partner, memory supplier and potential future manufacturing partner to OpenAI, while also buying technology from OpenAI’s competitors.

OpenAI is moving in the opposite direction through the stack. Jalapeño gives it a proprietary inference architecture rather than relying entirely on merchant accelerators, while its infrastructure partnerships secure the memory, networking, systems and data-centre capacity required to deploy such hardware. OpenAI says Jalapeño was developed with Broadcom in nine months and is intended for deployment at gigawatt scale as part of a multi-generation platform.

Samsung’s relevance to that programme is therefore obvious. The unanswered question is whether it will extend from memory and infrastructure into fabrication or deeper processor development.

Kim’s September 9 comments suggest there is at least some activity that can reasonably be described internally as chip-related collaboration. OpenAI Korea’s clarification shows that outsiders should not yet convert that into a Samsung foundry contract or a confirmed successor to Jalapeño.

There is a useful caution here about AI infrastructure news more broadly. The industry now contains overlapping agreements for chips, memory, networking, cloud capacity, data centres and enterprise software, often between companies that compete in one layer and depend on each other in another. A phrase such as “working together on chips” can describe several materially different arrangements.

For now, the verified picture is substantial but narrower than the first headlines implied: Samsung is helping supply the memory and infrastructure behind OpenAI’s expansion, while rolling OpenAI software out across a large part of its workforce. OpenAI has its own processor programme. Whether Samsung becomes part of the next processor itself remains an open question.

Sources

OpenAI — Samsung Electronics brings ChatGPT and Codex to employees OpenAI deployment announcement

Samsung Newsroom — Samsung Electronics brings external generative AI into its workplace Samsung workplace AI announcement

OpenAI — Samsung and SK join OpenAI’s Stargate initiative OpenAI Stargate partnership announcement

Samsung Global Newsroom — Samsung and OpenAI announce strategic partnership to accelerate advancements in global AI infrastructure Samsung infrastructure announcement

OpenAI — OpenAI and Broadcom unveil LLM-optimized inference chip OpenAI Jalapeño announcement

Reuters — OpenAI says working with Samsung on next-generation chips, deepening cooperation

Yonhap — ‘Nothing new to announce’ on possible chip collaboration with Samsung: OpenAI Yonhap clarification