OpenAI has struck letters of intent with Samsung Electronics and SK Hynix to supply undiced DRAM wafers for its ambitious“Stargate” AI infrastructure initiative, with projected demand reaching as high as 900,000 wafers per month-potentially equivalent to 40 percent of global DRAM output.

The agreements were announced during meetings in Seoul among OpenAI CEO Sam Altman, South Korean President Lee Jae-myung, Samsung Chairman Jay Y. Lee and SK Chairman Chey Tae-won. The memory chips are intended to support the massive compute and storage demands of next-generation AI systems.

Samsung and SK Hynix together control nearly 70 percent of the global DRAM market and roughly 80 percent of the high-bandwidth memory segment-technology critical for AI workloads. Under the terms of the deal, they will supply undiced wafers rather than finished chips, leaving the downstream processing and integration to third parties or future internal capacity.

Already, market reaction to the announcement has been sharp: SK Hynix shares surged by nearly 10 percent while Samsung's stock gained around 3–4 percent, hitting levels not seen since 2021. The deal signals mounting confidence in the scale of AI-induced semiconductor demand.

Analysts warn those volume targets shift supply dynamics. Global DRAM production is forecast to reach about 2.25 million wafer starts per month in 2025; if Stargate does indeed absorb 900,000 wafers, it would account for close to 40 percent of that output. This represents a scale more than double existing global HBM capacity.

To meet that demand, both chipmakers are working to reconfigure their production mix. Samsung has indicated it will funnel support via multiple arms-including Samsung SDS on data centre operations and Samsung C&T and Heavy Industries to explore floating data-centre designs to address cooling and land constraints. SK Hynix is advancing expansion of its Cheongju M15X plant and accelerating its Yongin cluster deployments.

Alongside the memory supply pact, SK Telecom has signed a memorandum of understanding with OpenAI to co-develop a“Stargate Korea” data centre, focused on the southwest region of the country. The strategic intent is to build regional compute hubs that reduce latency for Asian workloads and bolster South Korea's position in the AI infrastructure supply chain.

Floating data centres-being explored by Samsung in partnership with maritime engineering units-offer the promise of lower land costs and better thermal management via seawater cooling. But such designs also face challenges around maintenance, energy provisioning and regulatory jurisdiction over offshore facilities.

OpenAI has already committed up to $100 billion in its U. S.“Stargate” deployment schedule, with regional expansions planned across Asia, Europe and the Middle East. The South Korean deals extend that geographic reach and diversify supply chain risk as geopolitical tensions and export restrictions complicate semiconductor flows.

Memory shortage is already tightening margins in traditional DRAM markets. Inventory levels for DRAM and NAND have plunged to multi-year lows, with global memory firms reporting inventory coverage of just a few weeks-levels last seen during supply crunches earlier in this decade. The new allocations toward AI could exacerbate that squeeze.

Critics note the“up to 900,000 wafers” target is ambitious and may not be met in the near term. Scaling such volume across process nodes, yields, packaging and thermal design remains nontrivial. Yet the deal underscores OpenAI's hunger for control over compute supply chains rather than reliance on external procurement.

MENAFN12102025000152002308ID1110183700