OpenAI is working with Samsung on next-generation AI chips, deepening a partnership between the ChatGPT maker and the world's second-largest memory manufacturer, according to a report by Quartz.

The collaboration, as described in the report, focuses on developing silicon for the next generation of OpenAI's AI systems. The engagement adds a second major Korean supplier to OpenAI's hardware footprint. Samsung joins SK hynix — the dominant supplier of high-bandwidth memory (HBM) for AI accelerators — in the ChatGPT maker's orbit, and it reinforces Samsung's push to win back share in the HBM market that its crosstown rival has largely captured during the current AI buildout.

Details of the arrangement remain sparse. The Quartz report does not specify which product categories the work covers, whether it involves high-bandwidth memory stacked onto future accelerators, custom logic, or foundry engagement through Samsung Electronics' contract manufacturing arm. Nor does it attach a dollar value, volume commitment, or timeline to the effort. What it does establish is direction: OpenAI is widening its supplier base and its design ambition at the same time, rather than relying solely on purchasing finished GPUs from NVIDIA.

That broadening carries commercial weight for both sides.

For Samsung, an OpenAI design-in would be a meaningful reference in the segment that now drives memory pricing. HBM has become the most profitable product in Samsung's memory portfolio mix, and the company has spent the past year qualifying its HBM3E stack with NVIDIA — a process that dragged on longer than investors expected and ceded early volume to SK hynix. A direct working relationship with OpenAI, the largest single AI workload owner by some measures, would give Samsung a second anchor customer as it ramps next-generation HBM capacity and, over time, positions it for HBM4.

For OpenAI, the calculus is supply security and cost. The company trains and serves models whose compute demand has repeatedly outrun available accelerator supply, and NVIDIA's GPUs remain the industry's tightest large-scale bottleneck. Working directly with a memory and foundry giant such as Samsung gives OpenAI earlier influence over component roadmaps — HBM capacity, packaging choices, and possibly custom silicon — than a pure merchant-buyer relationship would allow. The Quartz report frames the effort as part of a deepening partnership rather than a one-off procurement deal, which suggests sustained engineering engagement rather than spot purchasing.

The report surfaces amid a broader pattern of hyperscalers and AI labs moving closer to the silicon layer. OpenAI's engagement with Samsung echoes moves by Google, Amazon, and Meta to co-design accelerators and secure dedicated memory supply. For Samsung specifically, the news follows its long qualification effort at NVIDIA and its public commitments to expand HBM production; an OpenAI win would strengthen the argument that its next-generation memory will ship in volume to multiple flagship AI platforms, not one.

SK hynix sits on the other side of this equation. It has held the lead position in HBM supply to NVIDIA and counts OpenAI's compute suppliers among its end customers. If OpenAI's Samsung work extends to memory supply, the duopoly dynamics in HBM pricing — which have favored suppliers throughout the AI cycle — could gain a third competitive pole, with Samsung leveraging the relationship to press its HBM4 case.

Much depends on scope that has not yet been disclosed. If the collaboration is confined to exploratory design work, its near-term revenue impact for Samsung is minimal. If it matures into committed HBM volumes or a foundry arrangement for custom accelerators, it would mark one of the most significant supplier relationships formed during the AI infrastructure cycle.

Neither OpenAI nor Samsung has publicly detailed deliverables, timelines, or financial terms beyond what the Quartz report describes. The partnership's next measurable milestones — qualified parts, announced volumes, or a named product family — will determine whether this becomes a structural shift in AI chip supply or another design exploration.