OpenAI Says It Is Building Next-Generation Chips With Samsung Electronics

What makes the remark new is that the relationship between the two companies had so far been known only on the memory-supply side. METAL reported OpenAI's statement that it will deploy its in-house inference chip Halapeno this year; that chip was designed with Broadcom and is manufactured by TSMC in Taiwan. The new remark means Samsung Electronics is involved in producing and researching the following generation of chips, but which process node is involved and what Samsung is responsible for were not disclosed.

Here is what the two companies have made public so far. On October 1, 2025, OpenAI announced that Samsung Electronics and SK hynix were joining Stargate. The two companies agreed to expand advanced memory production with a target of 900,000 DRAM wafers a month. In the same announcement, OpenAI signed a memorandum of understanding with Korea's Ministry of Science and ICT to review candidate sites for AI data centers outside the Seoul metropolitan area, agreed to explore building domestic data centers with SK Telecom, and agreed to look at additional data center capacity with Samsung C&T, Samsung Heavy Industries, and Samsung SDS. Both groups committed to adopting ChatGPT Enterprise and the API internally, and the deployment-scale comment at this briefing is confirmation that the commitment has actually been carried out.

Some detail on how Samsung Electronics uses the product also came out. Kim explained that Samsung Electronics employees in Korea and abroad use ChatGPT in research and development, marketing, and sales. Samsung Electronics would not confirm specifics, citing customer confidentiality. The side saying the company uses it is the seller; the buyer kept quiet.

The one figure offered was a growth multiple. ChatGPT Enterprise users at Korean companies grew roughly 28-fold between August 2025 and August 2026. OpenAI did not disclose absolute user numbers, adding only that as of late 2025, Korea had the most paid subscribers of any country outside the United States. The multiple is large, but without the starting point it is the kind of number that tells you nothing about scale.

On the chip side, the weight of the remark comes from where it sits in the supply chain. The compute OpenAI uses has so far consisted of NVIDIA GPUs, its own in-house inference chip, and the memory attached to them. Supplying memory and co-building a chip are different kinds of contracts. The first is a matter of volume and price; the second is a relationship that requires sharing design schedules and process choices. That said, in the remarks METAL reviewed, there was no mention of a foundry contract and no specification of which chip or which generation. Neither company issued a separate official statement.

Readers should watch for three things. First, whether the next-generation chip production is an actual foundry order or still a joint research stage. Second, whether the Stargate memory volume has moved from a memorandum of understanding to a binding contract. Last year's announcement was a target, not a committed volume. Third, whether the review of domestic data centers has progressed to site and power contracts. Until those three are confirmed, this remark is a signal about the direction of the relationship, not the contents of a contract.

To sum up: OpenAI said it is building next-generation chips with Samsung Electronics and named Samsung Electronics as one of ChatGPT's largest deployments. Neither side disclosed volumes, process nodes, or the form of the contract. Two things to watch from here: whether the next-generation chip collaboration is documented as a foundry order, and whether the Stargate memory target turns into a firm contract.