OpenAI's massive memory deal with Samsung and SK Hynix signals shift in AI chip supply chain

OpenAI has signed a sweeping cooperation agreement with Samsung Electronics and SK Hynix, securing high-performance memory supplies for the massive Stargate Project. According to South Korean outlets ZDNet Korea and Money Today, OpenAI CEO Sam Altman personally visited Seoul, meeting Samsung Group chairman Jae-yong Lee and SK Group chairman Tae-won Chey before signing a letter of intent (LOI) to formalize the deal.

The agreement calls for Samsung and SK Hynix to supply DRAM and HBMs for Stargate, a hyperscale data center initiative OpenAI is developing alongside Oracle and SoftBank Group. The project is expected to require up to US$500 billion in investment by 2029. Industry attention quickly focused on the scale of Altman's demand: a long-term supply of as many as 900,000 DRAM wafers per month. Both Samsung and SK Hynix confirmed the figure in their press releases, sparking intense debate about the feasibility.

Citing industry estimates, ZDNet Korea noted that Samsung's current DRAM production capacity is around 600,000-650,000 wafers per month, while SK Hynix produces about 500,000. Together, their combined monthly output barely matches OpenAI's request. In global terms, TechInsights analysts told Tom's Hardware that total DRAM production in 2025 is expected to reach 2.25 million wafers per month—meaning Stargate alone could consume up to 40% of the worldwide supply.

Analysts in Seoul stress that while the volume seems unrealistic in the near term, the request carries strong symbolic weight. One South Korean industry source told ZDNet Korea that the figure should be seen in the context of OpenAI's long-term outlook for the next five to ten years, reflecting its confidence that AI demand will continue to surge.

Redefining the AI supply chain

The deal represents more than just a bulk procurement agreement. According to Ddaily, the direct tie-up between OpenAI and the memory makers could reshape the balance of power in the AI semiconductor supply chain. Traditionally, companies like Samsung and SK Hynix sell HBM to chip suppliers such as Nvidia or AMD, which integrate the memory into GPUs later sold to hyperscalers like OpenAI or Meta.

This time, the model is reversed: memory firms will supply directly to OpenAI, bypassing Nvidia and AMD. Some observers interpret this as a signal that OpenAI is seeking to reduce its dependency on Nvidia GPUs, whose prices have surged amid tight supply. By securing memory capacity at the source, OpenAI gains more control over its critical hardware base.

OpenAI's partnership with Samsung and SK Hynix is not a routine procurement deal—it could be a turning point that disrupts Nvidia's dominance in the AI ecosystem, Ddaily noted.

Preparing for an Nvidia alternative

The supply pact also aligns with reports that OpenAI is developing its own AI chips. Industry sources told Reuters earlier this year that OpenAI is working with Broadcom on custom accelerators slated for production in 2026. The South Korean memory deal could therefore provide the foundation for those in-house chips, ensuring secure access to scarce HBM at a time when demand is skyrocketing.

Analysts argue that if OpenAI follows through, the current supply chain structure—where Nvidia sits at the center of AI hardware—could begin to fracture. "If OpenAI starts mass-producing its own AI semiconductors, Nvidia's monopoly position will inevitably weaken," one South Korean semiconductor executive told Businesspost.

From a memory industry perspective, the timing is strategic. HBM prices have been volatile due to intense competition, but long-term contracts with OpenAI could provide stability and sustained revenue growth. KB Securities analyst forecasts cited by Businesspost suggest that Samsung and SK Hynix's alignment with OpenAI may also strengthen South Korea's hand in trade and tariff negotiations with the US, as memory becomes a central pillar of AI infrastructure.

A new AI era, a new industrial order

The Stargate Project underscores how the AI race is transforming global semiconductor dynamics. As model training and inference workloads grow ever larger, the need for high-efficiency DRAM and HBM has become as critical as GPUs themselves. For memory suppliers, the partnership offers a rare chance to step out of Nvidia's shadow and establish direct relationships with AI service providers.

For OpenAI, the strategy is about future-proofing. Securing unprecedented levels of DRAM supply not only mitigates risks from GPU shortages but also lays the groundwork for building an independent hardware ecosystem. With AI demand showing no signs of slowing, the agreement highlights how semiconductor firms and AI developers are beginning to redraw the lines of cooperation.

As ZDNet Korea put it, the monthly request of 900,000 wafers may not be achievable today, but in the long arc of AI growth, it signals "the scale of ambition OpenAI sees as essential to lead in the AI era."

Source: DIGITIMES, October 2025

Article translated by Jingyue Hsiao and edited by Charlene Chen