From Memory Buyer to Co-Designer: OpenAI Confirms It Is Building Next-Generation Chips With Samsung

For most of the past year, OpenAI’s relationship with Samsung has been a customer relationship: the ChatGPT maker buys enormous quantities of DRAM and HBM, and Samsung (alongside SK hynix) feeds the memory-hungry Stargate data center buildout. On September 9, that framing changed. Speaking at a press conference in Seoul, Harrison Kim, General Manager of OpenAI Korea, said the two companies are now working together on the research and production of OpenAI’s next-generation chips — and that chip development is one of the areas where OpenAI has made the most progress in Korea.

Reuters, which first reported the remarks, framed it plainly: OpenAI is deepening cooperation with Samsung Electronics as it develops its own chips, expanding ties with South Korea’s semiconductor industry beyond memory procurement into joint silicon development. That is a meaningful step up in ambition, and it lands at a moment when every hyperscaler and frontier lab is racing to secure customized compute.

What was actually said

According to multiple reports from the Seoul briefing, Kim made three points worth separating carefully:

1. Joint research and production, not just supply. Kim said OpenAI and Samsung have made “meaningful progress” (some outlets quote “significant progress”) in working together on both the production and the research side of OpenAI’s next-generation chips. This goes beyond the October 2025 Stargate letters of intent, which were about memory supply and data center capacity. The new language is about co-developing logic silicon — the processors themselves.

2. Memory demand keeps compounding. Kim reiterated that OpenAI sees continued growth in memory chip demand. This is consistent with what is publicly known about inference economics in 2026: long-context models, agent workloads that hold state for hours, and retrieval-heavy pipelines are all memory-bandwidth-bound. The original Stargate framework with Samsung and SK hynix — up to 900,000 memory chips a month at its peak, by some reports — no longer looks like an upper bound.

3. Korea is becoming a full-stack partner. Kim noted that Samsung is one of the largest ChatGPT Enterprise deployments globally, and that OpenAI’s enterprise usage across South Korea has grown roughly 28x year over year. Samsung SDS, meanwhile, acts as OpenAI’s official ChatGPT Enterprise reseller in Korea, having signed on customers like Korea Zinc, iCRAFT, and TmaxSoft since late 2025. Chips, enterprise software, and distribution are now advancing together.

The double-sourcing question

The most strategically interesting detail came in the follow-on analysis. Tom’s Hardware and Yahoo Tech both noted that OpenAI’s next-generation processors could be manufactured at Samsung — double-sourcing with TSMC — a move that hints at massive volume requirements.

Context matters here. OpenAI unveiled its first custom AI chip, Jalapeño, on June 24, 2026 — an inference “Intelligence Processor” co-designed with Broadcom and manufactured by TSMC. Engineering samples have been running, and OpenAI presented the design in detail at Hot Chips 2026 in August, with early benchmark results on GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T. Jalapeño is an inference part; its job is to serve tokens cheaply at scale.

If Samsung’s fabs enter the picture for the next generation, the logic is straightforward: a single foundry cannot absorb OpenAI’s projected demand, and dual-sourcing gives OpenAI pricing leverage, supply resilience, and geographic hedging in an era when Taiwan-centric supply chains are treated as a strategic risk. For Samsung Foundry — which has trailed TSMC on leading-edge AI accelerator wins — landing even a share of OpenAI’s next-gen volume would be a marquee reference customer and a validation of its 2nm-class process roadmap.

No timeline was given for when Samsung-manufactured OpenAI silicon might tape out, and Kim’s remarks did not confirm that production is committed — only that joint research and production work is underway and progressing.

Why this matters

For OpenAI, custom silicon is existential economics. Inference costs dominate the P&L of any frontier lab, and Jalapeño’s pitch was performance-per-watt and cost-per-token improvements over merchant GPUs. A second foundry partner de-risks the scale-up and strengthens OpenAI’s negotiating position against Nvidia — whose GPUs still power the bulk of Stargate. It also deepens the vertical integration story: memory (Samsung/SK), logic (Broadcom design, TSMC and possibly Samsung fabs), and data centers (Stargate Korea and beyond).

For Samsung, this is a two-sided win. The memory division keeps shipping HBM and DRAM into the single most aggressive compute buildout in history, while the foundry business gets a shot at the kind of anchor customer that has historically gone exclusively to TSMC. And the enterprise side — SDS reselling ChatGPT, Samsung Electronics running one of the world’s largest ChatGPT Enterprise deployments — gives Korea Inc. a stake in OpenAI’s commercial success, not just its capex.

For the industry, the pattern is now unmistakable: frontier labs are becoming silicon companies. Google has TPU. Amazon has Trainium. Meta has MTIA. xAI, Anthropic, and now OpenAI are all building or specifying their own accelerators. When a lab the size of OpenAI starts double-sourcing its next-generation chips, it is not an experiment anymore — it is industrial policy for the model layer.

What to watch

• Whether OpenAI formally confirms Samsung Foundry as a manufacturing partner for a named next-gen chip (as opposed to “joint research”)
• HBM4 supply allocations between Samsung and SK hynix as Stargate Korea phases up
• Any Samsung LSI/Foundry announcements at upcoming semiconductor events referencing OpenAI volumes
• Whether ChatGPT Enterprise’s 28x Korea growth holds as SDS expands its reseller pipeline

One press conference does not seal a foundry deal. But the direction of travel is clear: OpenAI is no longer just buying Korean memory — it is designing Korean silicon.

Sources

• Reuters — OpenAI says working with Samsung on next-generation chips, deepening cooperation
• The Information — OpenAI Works With Samsung on Next-Generation Chips
• Tom’s Hardware — OpenAI says its next-generation processors could be made at Samsung
• TrendForce — OpenAI says it is working with Samsung on next-gen chips, ties could expand beyond memory
• OpenAI — Samsung and SK join OpenAI’s Stargate initiative