THE SIGNAL IN ONE SENTENCE
Samsung is helping supply and research the hardware behind OpenAI's systems while its own employees use ChatGPT across everyday business functions. The chip work and user totals remain light on public detail.
01
WHAT ACTUALLY CHANGED
OpenAI Korea General Manager Harrison Kim said at a Seoul press conference on September 9 that OpenAI has made substantial progress with Samsung Electronics on joint research and production for next-generation chips. He did not identify the chip design, manufacturing step, technical target, schedule, or commercial terms.
That carefully vague description sits beside a more concrete hardware program. OpenAI unveiled its first custom inference chip, Jalapeno, in June. Reuters reported that OpenAI engineers designed it with Broadcom and that TSMC would manufacture it. Samsung's role in the newly discussed next-generation work has not been publicly separated into memory, fabrication, packaging, design, or some combination.
The companies already have an infrastructure relationship. Samsung Electronics and SK Hynix signed letters of intent in October 2025 to supply memory chips for OpenAI's Stargate data centers. The September remarks suggest the relationship has expanded beyond supplying a known component, but neither company released a new contract or technical announcement with the press conference.
The flow also runs in the other direction. Kim described Samsung as one of the largest ChatGPT deployments in the world and said employees in South Korea and overseas use it across research and development, marketing, and sales. Samsung said it could not confirm customer-related information. The company had separately said in June that employees in its Device eXperience division could use ChatGPT, Gemini, and Claude internally.
OpenAI said ChatGPT Enterprise users across South Korean companies and institutions increased about 28 times between the end of August 2025 and the end of August 2026. It declined to provide the absolute user count. OpenAI also said South Korea had more paying ChatGPT subscribers than any country outside the United States as of late 2025. Both scale claims come from OpenAI rather than an independent usage audit.
02
WHY THIS MATTERS
Samsung sits in a rare position. It can help make the memory and other semiconductor systems that AI companies need, then use the resulting models throughout its own enormous operation. That creates a practical feedback loop between chip engineers, product teams, business users, and the people buying computing capacity.
South Korea's role in AI is therefore larger than being the world's very sophisticated parts cabinet. If local companies learn how models change design work, sales operations, manufacturing, and customer support, the country can accumulate deployment knowledge alongside semiconductor revenue. The useful asset is not only the chip. It is knowing what the chip becomes inside a working company.
Custom chips matter because general-purpose accelerators are expensive and scarce. A processor tuned for inference can target the calculations, memory movement, power use, and latency of serving models. Samsung's memory and manufacturing expertise could be valuable at several points in that system, although the public evidence does not yet tell us which points it owns.
A deployment inside Samsung would also be a serious enterprise test. Marketing drafts and sales summaries are one thing. Research files, chip plans, internal forecasts, and product road maps are quite another. Large-scale use requires data boundaries, access controls, evaluation, record keeping, and clear decisions about which work should never leave a protected environment.
The missing numbers are part of the story. A 28-fold growth rate can describe a move from a tiny base or a national shift. Calling Samsung one of the largest deployments does not reveal active users, departments covered, tasks completed, or business value. The next useful evidence will be outcome data, not another impressive multiplier floating without a denominator.
03
WHERE IT COULD HELP
- Use AI to support semiconductor research without exposing protected designs
- Deploy assistants across sales, marketing, finance, and operations with separate data rules
- Pair custom inference chips with memory designed for high-throughput model serving
- Measure adoption through completed work and verified outcomes rather than seat counts alone
- Build feedback loops between hardware teams and the employees using AI products
KEEP A HAND ON THE WHEEL
OpenAI supplied the new chip-cooperation and workplace-deployment claims at its own press conference. Samsung declined to confirm customer information, and neither company published a new technical agreement, contract value, delivery schedule, chip specification, employee count, or measured productivity result. The reported 28-fold increase covers ChatGPT Enterprise users across South Korean companies and institutions, not Samsung alone, and OpenAI withheld the underlying total. Letters of intent for memory supply are preliminary agreements rather than proof that every planned chip shipment or data center will occur.
04
TERMS WORTH KEEPING
OPEN GLOSSARY CARD
Custom AI chip
A processor designed around the particular calculations, memory patterns, and performance needs of AI workloads.
OPEN GLOSSARY CARD
High-bandwidth memory
Memory built to move very large amounts of data quickly between tightly stacked components and an AI processor.
OPEN GLOSSARY CARD
Enterprise deployment
The organized use of a technology across a company or institution with managed access, data rules, support, and oversight.
SOURCES AND VERIFICATION STATUS
This article was written from the materials below. Product claims and dates were checked against those sources on September 9, 2026.
PUBLICATION RECEIPT: Revision 1. Published September 9, 2026.
THE PUBLICATION ENGINE
WANT A SIGNAL OF YOUR OWN?
We build source-grounded publications, private briefings, and editorial systems for organizations with something useful to say.
WORK WITH US