OpenAI CEO Sam Altman returned to South Korea over the weekend for a whirlwind round of meetings with Samsung and SK Group executives, according to The Korea Herald, in a visit that underscored how dramatically the balance of power has shifted in the AI hardware supply chain. Tight memory supplies have strengthened chipmakers' bargaining power at precisely the moment OpenAI needs more of everything — more high-bandwidth memory, more wafers, more factories — while its own fundraising targets are shrinking under investor scrutiny.
Altman arrived Friday night and met with about 50 Samsung executives in Yongin, Gyeonggi Province, on Saturday, before a lunch with SK Group Chairman Chey Tae-won and executives from SK hynix and SK Telecom, the newspaper reported. At Samsung, discussions focused on using AI to improve business competitiveness, with Altman telling his hosts that "simply using AI will no longer be enough to secure a sustainable competitive advantage" and adding that he hoped to deepen cooperation with the Korean electronics giant. For more context on this story, see our ongoing latest AI developments.
Why Memory Is Suddenly the Bottleneck
The stakes for the visit are visible in the numbers coming out of the memory industry. TrendForce forecasts that the blended average selling price of high-bandwidth memory will surge 121 percent in 2027 from 2026, reflecting both tight supplies and a growing share of expensive HBM4 chips in the mix. HBM is the memory stacked directly against AI accelerators, and it has become the single most contested component in the AI buildout.
Samsung said last week that its 2027 HBM supply was sold out. SK hynix went further, saying that production through 2029 was effectively accounted for once firm contracts and volumes under customer negotiation are combined.
"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, according to The Korea Herald.
A Reversal of Leverage
For two years, the AI hardware story has been about GPU scarcity and Nvidia's pricing power. The memory crunch inverts that dynamic. "When memory chips are in short supply, buyers have less flexibility," an industry source told the newspaper. "Maintaining long-term relationships with Samsung and SK hynix is increasingly important."
That is precisely why Altman boarded a plane to Seoul. For OpenAI, securing sufficient memory at manageable cost is becoming critical as the company expands its computing infrastructure and develops custom AI accelerators with Broadcom — chips that will need guaranteed HBM supply of their own if they are to ship at scale.
The Stargate Shadow Over the Meetings
There is history behind the handshakes. In October 2025, OpenAI signed preliminary Stargate agreements with Samsung and SK outlining projected demand equivalent to 900,000 DRAM wafer starts per month — a figure The Korea Herald notes was a production target, not a binding order. A year later, with suppliers sold out years into the future, the question is how much of that aspirational demand OpenAI can actually lock in, and at what price.
Rising memory costs cut directly against OpenAI's infrastructure economics. Every dollar of HBM inflation flows into the cost of the data centers the company is racing to build, just as the investors financing those data centers are demanding clearer returns on AI spending.
Financing Pressure Raises the Stakes
The visit also comes amid signs that OpenAI's own fundraising is getting harder. According to industry sources cited by The Korea Herald, OpenAI has secured about $11 billion in commitments against an initial $30 billion fundraising target. The same sources indicated that a proposed valuation of $1.25 trillion fell short of the $1.4 trillion the company sought, and that the fundraising target may ultimately be reduced to around $20 billion. The figures have not been independently confirmed, and OpenAI has not publicly commented on them.
The financing pressure adds uncertainty to OpenAI's planned AI data center projects in Korea, which were part of the attraction for Korean partners when the Stargate agreements were signed. Cheaper memory is not on offer anywhere; the question for Altman was whether relationship-building can at least secure allocation priority when every hyperscaler is chasing the same wafers.
What It Means for Korea
For Samsung and SK hynix, the visit is confirmation that their HBM franchises have become strategic assets in geopolitics as much as commerce. Korea's two memory giants now sit at the center of the AI supply chain, with pricing power to match, and both are using the moment to lock in multi-year commitments from the biggest AI spenders.
For OpenAI, the weekend's meetings were about insurance. The company's roadmap — custom Broadcom accelerators, Stargate-scale data centers, and consumer products that depend on all of it — requires HBM supply that does not yet exist. Persuading Samsung and SK hynix to build it faster, and to give OpenAI its place in line, is now one of the most important jobs in the company. Altman's surprise trip to Yongin and Seoul shows he intends to do that job in person.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
