TL;DR
South Korea’s SK hynix warns that global AI memory demand will outpace supply in 2027, with no significant new capacity coming online. This shortage could impact AI development and geopolitical stability.
South Korea’s SK hynix has publicly warned that memory supply will fall significantly short of rising AI demand in 2027, with no meaningful new capacity expected to come online before then. This development highlights a critical bottleneck in AI infrastructure that could have broad industry and geopolitical implications.
During a press briefing at the Korea Chamber of Commerce and Industry’s Jeju Forum, SK hynix chairman Chey Tae-won stated that customers are requesting 60 to 100 percent more AI memory in 2027 than they are currently purchasing. With AI now accounting for over half of total semiconductor consumption, Chey estimated overall demand growth at 50–60 percent. However, he emphasized that No company has meaningful new capacity coming online next year
, creating a looming supply shortfall.
This imbalance is most acute in high-bandwidth memory (HBM), which is bonded to AI accelerators. Chey described the situation as leading to near-chaotic lobbying from corporate buyers and governments, with some nations viewing memory access as a matter of economic security. SK hynix has responded by accelerating capacity expansion plans, including moving forward the Yongin mega-cluster’s first clean room to February 2027 and committing over $14 billion to new facilities. Despite these efforts, the new capacity will not arrive until 2027, leaving a ‘gap year’ in supply.
Implications of Memory Shortage for AI and Geopolitics
This warning from SK hynix underscores a critical supply chain vulnerability that could hinder AI development and deployment globally. The concentration of memory capacity among three firms—SK hynix, Samsung, and Micron—raises concerns about geopolitical risks and market dominance. As demand outstrips supply, prices may rise, and access could become a strategic point of contention, especially amid increasing government intervention.
Moreover, the shortage impacts not just high-end training but also inference operations, with companies already owning hardware at a disadvantage if they lack sufficient memory capacity. This situation could influence AI innovation, competitive dynamics, and national security considerations.
High Bandwidth Memory (HBM) modules
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Memory Industry Concentration and Growing AI Demand
SK hynix held 58 percent of the global HBM revenue in Q1 2026, with Samsung and Micron sharing the remaining market. The industry faces a demand surge driven by AI’s rapid growth, which has outpaced supply projections for two consecutive years. While TSMC’s geopolitical focus is well-known, the HBM oligopoly is even more concentrated, making the supply chain more vulnerable.
SK hynix’s recent capacity expansion plans reflect a recognition of this bottleneck, but physical capacity additions—such as the Cheongju plant conversion and the Yongin mega-cluster—are not expected to be operational before 2027. This leaves a significant gap in supply during a period of unprecedented demand growth.
“No company has meaningful new capacity coming online next year.”
— Chey Tae-won, SK hynix Chairman
As an affiliate, we earn on qualifying purchases.
Uncertainties Surrounding Future Capacity and Demand
While SK hynix’s projections and plans are clear, it remains uncertain whether new capacity will fully meet the projected demand growth, or if geopolitical and economic factors will accelerate capacity expansion beyond current plans. Additionally, the impact on global supply chains and prices remains to be seen as the situation develops.
As an affiliate, we earn on qualifying purchases.
Next Steps in Capacity Expansion and Industry Response
SK hynix is expected to continue ramping up capacity, with targeted facility completions in 2027. Industry players and governments will likely increase focus on securing memory supply, possibly leading to new policies or collaborations to mitigate shortages. Monitoring capacity additions and demand trends will be critical in the coming months.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why is memory considered the bottleneck in AI development?
Memory, especially high-bandwidth memory like HBM, is essential for AI training and inference. The current supply cannot meet the rapidly rising demand, limiting AI progress and deployment.
What are the geopolitical implications of the memory shortage?
With a concentration of memory capacity among a few firms and countries, shortages could lead to increased geopolitical tensions, as nations vie for control over critical infrastructure and supply chains.
When will new memory capacity become available?
According to SK hynix, the first new capacity at their Yongin mega-cluster is expected to be operational by February 2027, with other expansions planned for that year.
How might this shortage affect AI innovation?
Limited memory supply could slow down AI research and deployment, especially for large-scale models, and increase costs for hardware upgrades and new deployments.
Are there alternative solutions to address this bottleneck?
Possible approaches include optimizing memory usage, developing more efficient models, or diversifying supply sources, but physical capacity constraints remain a primary challenge.
Source: ThorstenMeyerAI.com
