SK Group Chief Says AI Memory Shortage Has Become an Economic Security Issue
At the Jeju Forum on July 19, 2026, SK Group Chairman Chey Tae-won said foreign governments already treat AI memory access as an economic security issue. SK Hynix customers asked for 60% to 100% more AI memory for 2027, yet no manufacturer has meaningful new capacity scheduled for next year.
The Warning at the Jeju Forum
On July 19, 2026, at the Jeju Forum, hosted by the Korea Chamber of Commerce and Industry, SK Group Chairman Chey Tae-won said the global shortage of AI memory chips has stopped being just a commercial bottleneck and is rapidly turning into a geopolitical flashpoint, with foreign governments already intervening to secure domestic chip supply, according to reporting from the Korea Herald and The Investor.
Demand Supply Can't Keep Up With
According to Chey, major SK Hynix customers asked for 60% to 100% more AI memory for 2027 compared with current levels. With AI now accounting for more than half of all global semiconductor consumption, projected total demand growth sits at a minimum of 50% to 60%. The core problem, per the executive: no manufacturer has meaningful new production capacity scheduled to come online next year.
The Sharpest Bottleneck Sits in High Bandwidth Memory
The supply and demand imbalance is most severe in high bandwidth memory (HBM), a critical component for accelerating AI model training and inference. SK Hynix held 58% of the global HBM market by revenue in the first quarter of 2026, according to Counterpoint Research data cited in the reporting, a position that puts it at the center of any geopolitical decision over access to this type of chip.
SK Hynix's Response
Facing that demand pressure, SK Hynix moved up to February 2027 the opening of the first clean room at its Yongin cluster, previously slated for May, and committed an additional 21.6 trillion won, about $14.52 billion, in investment announced in March of this year. Even so, according to Chey, that expansion is unlikely to be enough to close the gap projected for 2027.
Why It Weighs on the Cost of Running AI in Brazil
For Brazilian agencies and SMBs, high bandwidth memory is an invisible but decisive input: it sits behind the final price of any GPU used to train or run AI models in the cloud, including the services underpinning everyday customer service and marketing automations. A shortage warning with a geopolitical edge, coming from the head of one of the world's two largest memory makers, signals that AI infrastructure costs are likely to stay under pressure for longer, with a knock on effect on the price of APIs and AI tool subscriptions used by Brazilian companies.
What's Clear
When the chairman of one of South Korea's largest conglomerates describes a chip shortage as an economic security issue, the message goes beyond simple market supply and demand. Governments already courting memory makers to secure domestic supply signal that access to this input may increasingly hinge on geopolitical alliances, not just commercial contracts. For anyone building a product on top of AI infrastructure, the practical lesson is to budget with slack and avoid depending on a single computing capacity supplier.