South Korea’s AI chip giants are fighting for a bigger role in America’s AI boom

South Korea’s two memory-chip giants are pushing harder into America’s AI boom, and the fight between them is getting sharper. SK Hynix has long been ahead in high-bandwidth memory, the fast specialized chips that AI systems need to move data quickly. Samsung is closing the gap. Over the past year, it has more than doubled its market share and taken business from its rival while signing fresh ties with Nvidia and OpenAI.

This is not just a corporate beauty contest. For U.S. AI firms, the winner matters because access to these chips can affect how quickly they can build the next wave of AI infrastructure, and how much it costs them. Christopher Miller of Tufts said the industry now cares about capacity as much as technical leadership, because memory production is tight. Ben Reynolds of Rhodium Group put it more bluntly: Nvidia’s allocation decisions still drive the market.

SK Hynix has been leaning into the U.S. side of the business with a $4 billion advanced packaging facility in Indiana, where it broke ground in August. Samsung has been building its own U.S. ties more slowly, but the list of moves is getting longer: work with Nvidia on AI-RAN in March 2025, entry into OpenAI’s Stargate project in October 2025, a joint AI factory plan with Nvidia, a June 2026 rollout of ChatGPT Enterprise and Codex for Samsung employees, and then September 2026 talks with OpenAI on next-generation chips.

The pressure is coming from China too. Both companies still have legacy plants there and still test chipmaking tools there, but the trade-war math is getting uglier. The U.S. ended the old Validated End User Program in 2025 and moved to annual licenses for export-controlled tools, while China’s domestic chip suppliers are getting better and harder to dislodge later. That leaves Samsung and SK Hynix trying to please Washington without giving up too much in China. Easy job, obviously.