OpenAI chief executive Sam Altman has travelled to South Korea to strengthen ties with Samsung and SK Hynix as worsening memory shortages threaten the company’s enormous data centre ambitions.
According to Tom’s Hardware, Seoul Economic Daily and Seoul Economic Daily, Altman met approximately 50 senior Samsung executives at the company’s training centre in Yongin on 31 October. He held separate discussions with SK Group chairman Chey Tae-won and executives from SK Hynix and SK Telecom during the visit.
The meetings come as memory manufacturers enjoy unprecedented bargaining power. Samsung and SK Hynix dominate supplies of high-bandwidth memory (HBM), which is essential for the accelerators powering increasingly expensive AI data centres.
The two Korean manufacturers controlled about 83 per cent of global HBM revenue in the second quarter of 2026. They accounted for around 63 per cent of conventional DRAM revenue, giving them considerable influence across the memory industry.
Samsung said its discussions with Altman covered AI development, business opportunities and the companies’ existing cooperation. The relationship includes a letter of intent signed in October 2025 to supply memory for OpenAI’s Stargate infrastructure project. No new supply agreement was announced.
Altman’s meeting with SK focused on existing agreements covering HBM supplies and data centre construction. SK Hynix previously outlined plans to establish production capacity capable of meeting demand equivalent to as many as 900,000 DRAM wafers per month for the Stargate project.
The timing is particularly awkward for OpenAI. The company is trying to secure enormous quantities of memory while manufacturers are committing production capacity years in advance. Samsung reportedly has its 2027 HBM output sold out, while SK Hynix says much of its production through 2029 is effectively committed.
Market research firm TrendForce expects HBM average selling prices to increase by 121 per cent in 2027, partly because of the growing proportion of expensive HBM4 products. Such increases threaten to make OpenAI’s already eye-watering infrastructure spending considerably more painful.
The shortage extends beyond specialised memory. Conventional DRAM supplies are increasingly constrained as manufacturers devote resources to more profitable data centre products. TrendForce estimates that DRAM supply will fall between one and two per cent short of demand in 2026, with conditions expected to deteriorate further next year.
Altman recently suggested that conventional PCs use scarce memory inefficiently because much of their installed DRAM spends its time idle. He argued that data centres could use resources better by sharing computing power across workloads. PC owners will doubtless be delighted to discover he thinks that their memory belongs somewhere else.
The comments attracted criticism because the rush to build centralised AI infrastructure is helping push up prices for consumer memory. PC manufacturers, smartphone makers and other electronics companies are competing for increasingly expensive components while large technology companies secure long-term supply agreements.
AMD chief executive Lisa Su recently visited South Korea to discuss additional HBM supplies with Samsung. She emphasised the importance of planning semiconductor requirements three to five years ahead, reflecting how difficult it has become to secure sufficient production capacity.
Altman’s Korean visit comes amid growing investor concerns about the cost of AI infrastructure. OpenAI reportedly reduced its latest fundraising target from $30 billion to $20 billion, while the cost of memory, accelerators, electricity and data centre construction continues climbing. Samsung and SK have not confirmed whether the latest meetings produced any new investment commitments or agreements on memory pricing.
