Sam Altman Visits Samsung: Stargate Memory Supply Moves from Letter of Intent Toward Concrete Lock-In

On October 31, OpenAI CEO Sam Altman visited Yongin, Gyeonggi Province, South Korea, holding a discussion with about 50 Samsung Group executives at the Samsung Human Resources Development Center—nearly a year after his last visit to Samsung. The theme was "AI Changes the Next Decade: Corporate Choices," with attendees spanning Samsung Electronics, finance, and biotech business lines.

Samsung said in an official announcement: "Samsung executives and CEO Sam Altman discussed the direction of AI technology development and its potential applications, as well as the new markets and business opportunities this will create." Altman said during the exchange that a company's competitive advantage does not depend on simply adopting AI, but on how it uses AI to act faster, build better products, and outperform competitors. He also presented a Codex Micro keyboard—a device designed for use with OpenAI's coding agent—to the head of Samsung's AI Center.

Notably, Samsung Electronics Chairman Lee Jae-yong did not attend. Those present included Jeon Young-hyun, head of Samsung Electronics' Device Solutions (DS) division; Roh Tae-moon, head of the Device eXperience (DX) division; and executives from affiliates such as Samsung SDI and Samsung Electro-Mechanics. Although global DRAM and high-bandwidth memory (HBM) supply is extremely tight, Samsung did not announce new memory supply contracts or disclose additional purchase commitments at this meeting.

From Letter of Intent to Processors: A Year of Expanding Cooperation

The meeting continues existing cooperation between the two sides. On October 1, 2025, OpenAI announced a strategic partnership and signed a letter of intent with Samsung and SK Group for the "Stargate" project, covering memory semiconductors and AI infrastructure. According to public information, Samsung Electronics plans to supply memory chips for Stargate; Samsung SDS is responsible for data center construction and enterprise AI services; and Samsung C&T and Samsung Heavy Industries are jointly developing offshore floating data centers.

At the time, Samsung identified potential demand from OpenAI of up to 900,000 DRAM wafers per month—a figure representing potential demand, not confirmed monthly supply orders. For reference, Stargate is a hyperscale AI infrastructure initiative launched by OpenAI with Oracle, SoftBank, and others, planning up to US$500 billion in investment over four years.

Cooperation has continued to expand since then. In July, Lee Jae-yong made a special trip to OpenAI headquarters in San Francisco to meet Altman and discuss cooperation on HBM, DRAM, and other memory products. In September, foreign media reported that OpenAI and Samsung were exploring next-generation processor development, with cooperation potentially extending from memory to wafer foundry, though no manufacturing contract was disclosed at the time. Sources said that before Altman's arrival in Korea, Stargate project staff had already traveled to Korea to hold multiple rounds of preliminary consultations with local companies on contract scale and chip supply terms.

Memory Super Cycle: AI Pushes DRAM Prices Toward a 270% Increase

The backdrop to the meeting is white-hot competition for memory chips driven by AI infrastructure. AI accelerators need HBM for high-bandwidth data access, while the servers surrounding accelerators also consume large amounts of conventional DRAM—AI data centers, enterprise servers, and traditional electronics makers are competing head-on in the same supply chain. The supply imbalance is already quantified: TrendForce estimates DRAM supply will fall 1% to 2% short of demand this year, with the gap expected to widen further in 2027; Micron warns that new capacity takes time to come online and shortages could last until 2028. Market research firm SemiAnalysis calculates that memory's share of hyperscale cloud providers' capital expenditure has jumped from about 8% in 2023–2024 to about 30% in 2026.

Chart interpretation: TrendForce expects general-purpose DRAM contract prices to rise 10% to 15% quarter-on-quarter in Q4 2026, with NAND Flash up 15% to 20%. Memory price pressure will persist throughout the year, with cumulative 2026 gains expected to reach about 270%.

Chart interpretation: Memory's share of spending jumped from about 8% to about 30% in two years. The AI computing race is making memory one of the biggest variables in data center capital expenditure—and the core reason Altman traveled to Korea personally to lock in supply.

The pricing power comes from a highly concentrated supply landscape: Samsung and SK Hynix together hold about 83% of global HBM revenue and about 70% of the global DRAM market. To meet demand, Samsung announced commercial shipments of HBM4 in February this year, with 12-layer devices delivering 11.7Gbit/s per pin and capacities ranging from 24GB to 36GB. Sources said Altman also had lunch with SK Group Chairman Chey Tae-won the same day, attended by SK Hynix and SK Telecom executives and OpenAI hardware vice president Richard Ho. The industry widely believes HBM supply and Korean AI data center construction were core topics.

No Contract Yet, AI Training First

Samsung appears to value long-term positioning more. The company said it has provided AI training to about 2,300 executives and plans to extend it to all employees by February 2027. The discussion with Altman is part of Samsung's internal AI transformation plan, rather than a public expansion of memory supply agreements.

However, the relationship is not without discord. Altman recently said on a podcast that DRAM in personal computers is often underutilized and that this situation is "pretty bad," suggesting data centers should share resources across AI workloads and use memory more intensively—remarks that drew widespread criticism given OpenAI's own services rely heavily on centralized computing. TrendForce also noted that consumer customers' tolerance for memory prices is nearing its limit, PC makers are beginning to cut SSD capacity in mainstream models, and PC DRAM supply in 2027 may tighten further as capacity shifts to servers.

From a letter of intent to a binding long-term supply contract, the two sides still have a way to go. As of publication, neither Samsung nor OpenAI has disclosed substantive terms such as new investment, memory pricing arrangements, or data center construction cost sharing. But the industry widely interprets the meeting as a signal that Stargate memory supply is moving from the intent stage toward substantive lock-in.