OpenAI deepens next-generation chip cooperation with Samsung: 'One of the areas where we have made the most progress'

Let's first map OpenAI's custom-chip roadmap. In June, OpenAI officially unveiled its first custom AI chip, Jalapeno: designed by OpenAI's own engineers and developed with Broadcom, optimized for AI inference — the stage where applications like ChatGPT crunch data and generate answers, a different track from training chips. At the time OpenAI confirmed TSMC would manufacture Jalapeno. Now Kim's remarks in Seoul suggest the next-generation map is expanding: he did not say explicitly whether Samsung would be 'manufacturer' or 'co-developer,' but the phrase 'joint production and research' clearly goes beyond mere purchasing. Tom's Hardware's analysis offers an important lens: OpenAI's next-generation processors may be dual-sourced between Samsung and TSMC — a supply-chain arrangement that usually implies enormous volume requirements, because customers only pursue dual-sourcing when demand is too large for a single fab to absorb or single-source risk has become unacceptable.

Samsung's role in OpenAI's chessboard goes far beyond being a mere 'foundry.' Another set of details Kim revealed at the press conference sketches a complex symbiosis between the two companies — Samsung Electronics is 'one of the largest-scale deployments of ChatGPT globally': employees in Korea and overseas use ChatGPT in daily work across R&D, marketing, sales and more, making the tech giant itself one of OpenAI's heaviest enterprise customers. Earlier moves were already paving this relationship: last year Samsung Electronics and SK Hynix signed letters of intent to supply memory chips for OpenAI's Stargate data center project; in June, Samsung announced employees in its Device eXperience (DX) division — which oversees phones, TVs and home appliances — would be allowed to use generative AI tools including ChatGPT, Google's Gemini and Anthropic's Claude internally. Kim also highlighted explosive growth in the Korean market: by the end of August, ChatGPT Enterprise users at South Korean companies and institutions had surged about 28-fold year over year, as businesses expanded AI from individual tasks to company-wide functions spanning legal, sales, finance and marketing.

Stretching the timeline further, the Samsung-OpenAI 'marriage' has been long in the making. Last year, Samsung Electronics and SK Hynix signed letters of intent to supply memory chips for OpenAI's Stargate data center project — the hundred-billion-dollar AI infrastructure initiative OpenAI is driving with SoftBank and others, whose appetite for advanced memory like HBM is virtually bottomless. Kim said plainly at the press conference that as more advanced chips enable faster and more complex computing, memory chip demand will keep growing, making cooperation with Korean semiconductor suppliers a priority for OpenAI's local office. The industry backdrop: the global AI chip landscape is undergoing a quiet restructuring — Nvidia remains the undisputed king of training chips, but OpenAI, Google (TPU), Amazon (Trainium) and Meta (MTIA) are all accelerating custom silicon to reduce dependence on a single supplier and optimize inference costs. By pulling Samsung into the next-generation chip circle, OpenAI is both securing production capacity outside TSMC for its post-Jalapeno lineup and staking an early claim in Korea — the home base of memory semiconductors — for the massive storage demand of the Stargate era.

What makes this story truly worth savoring is that it simultaneously touches three fronts: chip manufacturing, AI inference costs and enterprise AI adoption. For the chip industry, the expectation that 'OpenAI's next-gen processors could be dual-sourced from Samsung and TSMC' signals custom AI chip orders are moving from pilot to volume — if even a player like OpenAI needs dual-source insurance, the demand curve for AI inference silicon is far steeper than outsiders assume. For OpenAI, the strategic value of custom silicon lies in cost and bargaining power: every ChatGPT answer burns inference compute, and holding the inference chip in its own hands is a key step toward lowering costs and shedding the 'working for Nvidia' narrative. For Korea, the '28-fold growth in ChatGPT Enterprise users in a year' figure Kim cited is a snapshot of how fast enterprise AI is penetrating Asia — from individual experimentation to company-wide deployment, Korea's path is being replicated across more and more markets. Three forces converging in one news story is exactly why AI competition is no longer a single-model arms race but an all-front struggle spanning chips, energy, enterprise services and geopolitical supply chains.

As I finish, I am reminded of a simple observation: a company's circle of partners often says more than its keynotes. OpenAI's moves over the past half year — engaging Broadcom to design chips, TSMC to manufacture them, and now pulling Samsung into joint development of next-generation silicon — are essentially answering one question: when your product is used by hundreds of millions of people every day and every conversation burns compute, to whom are you willing to entrust your lifeline? The answer is clearly: to no single party. From Jalapeno to next-gen chips, from TSMC to Samsung, OpenAI is laying the foundation of its AI empire with an ever-denser supply chain network. For ordinary people, the most practical way to read this story may be: the next time you use ChatGPT, behind every answer there may well be a chip stamped with both Silicon Valley code and Korean process technology working away — AI competition is no longer just what happens on your screen.