OpenAI has begun building an exclusive collaboration line with Samsung Electronics to secure high-bandwidth memory (HBM), a core component of artificial intelligence (AI) infrastructure. As the shortage of memory semiconductors deepens, it is seen as an active move to lock in Samsung Electronics' production capacity.
According to the industry on the 16th, OpenAI Chief Financial Officer Sarah Friar met privately with Samsung Electronics executives in the middle of last month during her visit to Korea to focus on discussing supply and rollout plans for sixth-generation HBM (HBM4). The two sides are said to have exchange views on the HBM4 mass production timeline, supply stability, and a medium- to long-term cooperation framework.
Observers say this contact goes beyond a simple technical consultation and amounts to a strategic gambit over control of a key component in the AI era. During this trip, Friar held external meetings, including with executives at SK Networks and Upstage, but is said to have focused on talks with Samsung Electronics without a separate meeting with SK hynix, another pillar of the HBM market.
The market sees this as a strategic choice by OpenAI to reduce reliance on a specific supply chain while designating Samsung Electronics, which can handle large-scale demand, as a key partner to increase its bargaining power.
In fact, OpenAI is said to be facing both astronomical funding burdens and a memory supply crunch as it advances its AI models. An industry official said, "OpenAI's supply-demand situation is so urgent that it is asking memory manufacturers for active cooperation," adding, "Samsung Electronics, with its vertically integrated supply chain, has emerged as the most attractive partner for OpenAI's infrastructure buildout."
Friar's latest move is directly tied to Stargate, OpenAI's project to build a mega-scale AI data center, and the development of its own AI Semiconductor "Titan." Samsung Electronics Chairman Lee Jae-yong met OpenAI Chief Executive Officer Sam Altman in Oct. last year and signed a letter of intent (LOI) on memory supply, and the latest talks are seen as a step to flesh out plans to equip the Titan chip with Samsung HBM4 based on that agreement.
However, some speculate the talks between the two companies may have focused more on developing the "Titan chip." OpenAI recently decided to withdraw from the Stargate data center project it had been pursuing in Norway, following Texas in the United States and the United Kingdom, and executives in charge of the project have also left, prompting changes. OpenAI is reducing the weight of the Stargate project and is focusing on developing its own AI chips to reduce reliance on Nvidia.
For that reason, there is also an interpretation that it prioritized cooperation with Samsung Electronics over SK hynix. The idea is that OpenAI aims to procure better products for Titan development. A semiconductor industry official said, "Up to fifth-generation HBM (HBM3E), SK hynix held market leadership, but as we move to HBM4, Samsung Electronics is rising," adding, "The fact that Samsung Electronics mass-produced and shipped products without design change issues likely influenced Friar's latest move."
As AI models evolve into a "self-generation" structure that creates and learns data on their own, demand for memory semiconductors is surging even more. A senior semiconductor industry official said, "I understand OpenAI recently overcame the performance degradation issue that occurs when AI-generated data is used as AI training data," adding, "As AI self-generation has become possible, the rapid increase in demand for memory to store the torrent of data is one of the main reasons for this visit to Korea."
Starting with HBM4, the competition for leadership in the AI memory market is shifting beyond technical prowess into a battle over "stable large-scale supply capacity." The industry is watching whether Samsung Electronics can use its cooperation with OpenAI as leverage to reshape the HBM market landscape and secure leadership.