TL;DR
SK hynix’s chairman warns that AI memory demand is set to outpace supply by 2027, with no significant new capacity coming online. This could create major bottlenecks and geopolitical tensions affecting AI progress.
SK hynix’s chairman, Chey Tae-won, warned last week that AI memory demand is expected to increase by 60 to 100 percent in 2027, with no meaningful new capacity coming online next year. This shortage could create critical bottlenecks in AI development, impacting both industry and geopolitical stability, according to his remarks at the Korea Chamber of Commerce and Industry’s Jeju Forum.
The demand for high-bandwidth memory (HBM) used in AI accelerators is growing rapidly, driven by AI now accounting for over half of total semiconductor consumption. Chey Tae-won stated that, despite this surge, no significant new capacity is expected in 2027, creating a supply-demand imbalance.
He highlighted that SK hynix’s market share of global HBM revenue was approximately 58% in Q1 2026, with the rest split evenly between Micron and Samsung. The current capacity constraints have led to high memory prices, which Chey described as ‘abnormal’ and a potential driver of ‘chipflation,’ risking broader economic and geopolitical repercussions.
In response, SK hynix has announced plans to accelerate capacity expansion, including moving the Yongin mega-cluster’s first clean room to February 2027 and investing over $14 billion in new facilities. However, these developments will not address the capacity shortfall until at least 2027, leaving a gap year where demand outstrips supply.
Implications of Memory Shortage for AI and Geopolitics
The projected memory shortage poses a risk to the future of AI development, especially for training large models that rely heavily on high-bandwidth memory. As demand grows faster than supply, costs for AI infrastructure could increase, making advanced AI less accessible and slowing innovation.
Additionally, the concentration of HBM production among three companies, with SK hynix holding a majority share, raises concerns about supply security. Chey Tae-won warned that governments may soon intervene, treating memory access as a matter of national economic security, which could lead to geopolitical tensions and export restrictions.
For industry players and policymakers, these developments highlight the importance of diversifying supply chains and investing in capacity expansion to avoid bottlenecks that could hinder AI progress and exacerbate geopolitical conflicts.
High Bandwidth Memory (HBM) for AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Current State of AI Memory Capacity and Industry Trends
Memory bottlenecks have been an ongoing concern, with demand for high-bandwidth memory increasing sharply as AI models grow larger and more complex. SK hynix, Micron, and Samsung dominate the HBM market, with SK hynix holding a majority share.
Recent industry reports indicate that demand for AI memory is expected to grow at a compound annual rate of around 33% through 2030, but capacity expansion plans are lagging. SK hynix’s announced investments aim to address this, yet the new capacity will not be operational until at least 2027, creating a significant short-term gap.
The situation is compounded by geopolitical factors, as countries view memory access as critical to national security, leading to increased lobbying and potential export controls. This concentration of supply among few firms increases vulnerability to supply shocks.
“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 Capacity Expansion and Geopolitical Responses
It remains unclear how quickly SK hynix and other manufacturers can accelerate capacity expansion beyond announced plans, or how governments might intervene to secure supply chains. The timeline for new capacity coming online is also subject to delays due to technical, regulatory, or geopolitical factors.
Additionally, the extent to which geopolitical tensions will escalate or lead to export restrictions remains uncertain, which could further complicate supply security for AI memory.
As an affiliate, we earn on qualifying purchases.
Next Steps in Capacity Building and Policy Developments
Industry players are expected to accelerate investment in new manufacturing facilities, with SK hynix moving forward on its planned expansions. However, the capacity shortfall is likely to persist through 2026, with the first significant supply increases expected in 2027.
On the geopolitical front, governments may increase intervention to secure memory supplies, potentially leading to export controls or strategic stockpiling. Monitoring policy responses and capacity developments will be critical over the coming months.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why is memory capacity so critical for AI development?
Memory capacity, especially high-bandwidth memory (HBM), is essential for training and deploying large AI models. Insufficient memory can create bottlenecks, slowing down AI progress and increasing costs.
What are the main risks of a memory shortage?
The risks include slowed AI innovation, increased infrastructure costs, and potential geopolitical conflicts as countries vie for control over critical supply chains.
How are companies responding to the capacity shortfall?
Companies like SK hynix are investing heavily in new manufacturing facilities, with plans to accelerate capacity expansion, though these will not fully address the shortage until 2027.
Could this shortage impact consumer technology?
Yes, high memory prices and supply constraints could lead to increased costs for consumer electronics and workstation hardware, contributing to broader inflation in tech markets.
What might governments do to address this issue?
Governments may intervene by restricting exports, providing subsidies for capacity expansion, or establishing strategic reserves to secure memory supplies for critical industries.
Source: ThorstenMeyerAI.com
