OpenAI CEO Sam Altman returned to South Korea over the weekend facing two mounting challenges: securing enough memory chips and raising the billions needed to build the infrastructure behind them.

His surprise visit highlighted the growing importance of Samsung and SK to the ChatGPT maker, as tight memory supplies strengthen chipmakers' bargaining power while investors demand clearer returns on massive AI spending.

Altman arrived Friday night and met about 50 Samsung executives in Yongin, Gyeonggi Province, on Saturday before lunch with SK Group Chair Chey Tae-won and executives from SK hynix and SK Telecom.

At Samsung, discussions focused on using AI to improve business competitiveness.

"Simply using AI will no longer be enough to secure a sustainable competitive advantage," Altman said, adding that he hoped to deepen cooperation with Samsung.

The stakes are rising as demand for high-bandwidth memory, a critical component of AI accelerators, continues to outstrip supply.

TrendForce forecasts HBM's blended average selling price will surge 121 percent in 2027 from 2026, reflecting tight supplies and a greater share of expensive HBM4 chips rather than uniform price increases.

Samsung said last week its 2027 HBM supply was sold out. SK hynix said production through 2029 was effectively accounted for, combining firm contracts with volumes under customer negotiations.

"Customers are requesting more than twice the volume we can supply," Kim Ki-tae, SK hynix's head of HBM sales and marketing, said during an earnings call.

For OpenAI, securing sufficient memory at manageable costs is becoming increasingly important as it expands computing infrastructure and develops custom AI accelerators with Broadcom.

In October 2025, OpenAI signed preliminary Stargate agreements with Samsung and SK outlining projected demand equivalent to 900,000 DRAM wafer starts per month. The figure was a production target, not a binding order.

"When memory chips are in short supply, buyers have less flexibility," an industry source said. "Maintaining long-term relationships with Samsung and SK hynix is increasingly important."

Higher memory costs could inflate OpenAI's AI infrastructure budgets just as investors become more selective about funding the industry's expansion.

According to industry sources, OpenAI has secured about $11 billion in commitments against an initial $30 billion fundraising target. Its proposed valuation of $1.25 trillion reportedly fell short of the $1.4 trillion sought, while the fundraising target may be reduced to $20 billion. The figures have not been independently confirmed.

The financing pressure adds uncertainty to OpenAI's planned AI data center projects in Korea. Samsung affiliates and SK Telecom agreed last year to explore joint developments, but investment budgets and construction schedules remain unsettled.

"Building a data center involves more considerations than we initially expected," OpenAI Korea chief Kim Kyoung-hoon said in September, citing evolving technical requirements.

For Samsung, deeper cooperation could create opportunities in enterprise AI and data center services alongside chip sales. SK could combine SK hynix's memory business with SK Telecom's infrastructure capabilities.

No new agreements were announced after Altman's visit, and neither side confirmed discussions of chip pricing or investment commitments.

With memory supplies tightening and investors demanding clearer returns, the real test will be whether OpenAI can turn its Korean partnerships into firm supply contracts and funded projects.

By Jie Ye-eun (yeeun@heraldcorp.com)