TL;DR
SK Group’s chairman warns that AI memory demand will outpace supply significantly by 2027, risking a global shortage. The capacity gap and geopolitical tensions are key concerns for the industry.
South Korea’s SK Group has publicly warned that the global AI memory shortage is imminent, with demand expected to increase by 60-100% by 2027 and no significant new capacity coming online. This announcement, made by SK hynix chairman Chey Tae-won during a press briefing at the Korea Chamber of Commerce and Industry’s Jeju Forum, highlights a critical supply-demand imbalance that could impact AI development and geopolitical stability.
Chey Tae-won stated that AI now accounts for more than half of semiconductor consumption, and industry demand for high-bandwidth memory (HBM) is projected to grow at a minimum of 50-60% annually. Despite this, no meaningful new capacity is expected to be available in 2027, leading to what he described as a near-chaotic lobbying environment involving both corporate and government actors.
SK hynix, which holds approximately 58% of the global HBM revenue as of Q1 2026, has announced investments to boost capacity, including the acceleration of the Yongin mega-cluster’s first clean room to February 2027 and a 21.6 trillion won (~$14.5B) investment. Still, these projects will not address the capacity shortfall until 2028 at the earliest, creating a supply gap that could influence global AI deployment and geopolitical dynamics.
Chey warned that high memory prices, driven by supply constraints, are causing chipflation and attracting new entrants into the semiconductor industry, including Elon Musk’s interest in manufacturing. He also cautioned that the current pricing environment is unsustainable and poses a strategic risk, especially as governments begin viewing memory access as a matter of economic security.
Implications of Memory Shortage for AI Development
The warning from SK hynix’s chairman underscores a looming critical bottleneck in AI advancement: the availability of high-bandwidth memory. As demand outstrips supply, costs are likely to rise, potentially slowing AI innovation and deployment. Additionally, the concentration of HBM capacity among a few companies raises concerns about geopolitical influence and security. The industry’s inability to meet demand could lead to increased government intervention and strategic competition, affecting global AI progress and economic stability.
High Bandwidth Memory (HBM) modules
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Memory Capacity and Industry Concentration in 2026
SK hynix’s dominance in the HBM market, with 58% of global revenue, highlights a highly concentrated supply chain. This oligopoly, combined with demand growth that has outpaced guidance for two consecutive years, has exacerbated supply constraints. While TSMC and other chipmakers face geopolitical scrutiny, the HBM market’s tight capacity and demand surge pose a more immediate and concentrated challenge—one that could influence both industry pricing and national security policies.
SK hynix’s investments aim to expand capacity, but the earliest significant capacity additions will not arrive until 2028, leaving a short-term supply gap that could impact AI hardware availability and costs.
“No company has meaningful new capacity coming online next year.”
— Chey Tae-won, SK Group Chairman
As an affiliate, we earn on qualifying purchases.
Uncertainties About Capacity and Geopolitical Impact
It remains unclear how quickly SK hynix and other suppliers can scale capacity to meet the projected demand, or how governments will respond to the increasing geopolitical tensions over memory access. The exact timeline for capacity additions beyond 2028 and the potential for new entrants or alternative solutions are still developing.
As an affiliate, we earn on qualifying purchases.
Next Steps for Industry and Policy Responses
Industry players are expected to continue ramping up investments, with capacity additions targeted for 2027 and beyond. Governments may increase scrutiny and intervention over memory supply chains, especially as demand grows and geopolitical tensions rise. Monitoring these developments will be crucial for assessing how the industry addresses the looming shortage and geopolitical challenges.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why is memory considered the main bottleneck in AI development?
High-bandwidth memory (HBM) is essential for AI training and inference. Demand is growing rapidly, but supply is limited, creating a bottleneck that can slow down AI progress and increase costs.
What are the geopolitical implications of the memory shortage?
Concentration of memory capacity among a few companies and countries raises concerns about supply security and strategic influence. Governments are increasingly viewing memory access as vital to economic security, which could lead to intervention and conflict.
When will new memory capacity be available to meet demand?
SK hynix’s current investments aim to expand capacity by 2027-2028, but the earliest significant additions are not expected until 2028, leaving a short-term shortfall.
How might this shortage affect AI hardware costs?
Limited supply and high demand are likely to keep memory prices elevated, contributing to ‘chipflation’ and increasing costs for AI hardware and consumer devices.
Could alternative memory solutions alleviate the shortage?
While some inference tasks can use unified memory like LPDDR, high-performance training relies heavily on HBM. Developing alternative solutions or increasing capacity is crucial but remains challenging in the short term.
Source: ThorstenMeyerAI.com
