The US export restrictions reflect ongoing geopolitical tensions over technology dominance
The United States government has introduced stricter export controls on high-bandwidth memory (HBM) chips used in artificial intelligence (AI) applications, targeting both US-made and foreign-produced technologies. These advanced semiconductors are vital for cutting-edge AI systems, and the restrictions are part of ongoing efforts to limit China’s access to critical technology.
What Is High-Bandwidth Memory?
High-bandwidth memory is a stack of memory chips designed to store and transfer data faster than older technologies like dynamic random access memory (DRAM). HBM is widely used in high-performance computing, graphics cards, data centres, and autonomous vehicles.
These chips are particularly crucial for running AI applications, including generative AI models, which require both processors like Nvidia’s GPUs and efficient memory systems to handle massive datasets. As G Dan Hutcheson, vice chair of TechInsights, explains, “The processor and the memory are two essential components to AI. Without the memory, it’s like having a brain with logic but not having any memory.”
Impact On China
The restrictions, announced on December 2, build on previous export curbs by the Biden administration aimed at curtailing China’s access to military-critical technologies. In response, Beijing has imposed its own curbs on exporting materials like germanium and gallium, essential for semiconductor manufacturing.
Industry experts predict the restrictions will slow China’s development of advanced AI chips in the short term. Currently, South Korea’s SK Hynix and Samsung, along with America’s Micron, dominate HBM production, leaving Chinese manufacturers, such as Yangtze Memory Technologies and Changxin Memory Technologies, at a disadvantage.
Jeffery Chiu, CEO of Ansforce, a tech consultancy, says, “What the US export restrictions would do is cut China’s access to HBM of better quality in the short run. In the long run, though, China will still be able to produce them independently, albeit with less advanced technologies.”
Why HBM Matters
HBM chips are essential for AI due to their larger storage capacity and faster data transmission compared to conventional memory chips. These attributes enable smoother operation of AI applications, especially large language models (LLMs), by allowing them to process vast amounts of data without delays.
The increased speed of HBM can be compared to upgrading a two-lane road to a hundred-lane highway, significantly reducing data bottlenecks. This performance boost makes HBM a cornerstone for applications in AI, autonomous systems, and advanced data processing.
Who Dominates Market?
Three companies—SK Hynix, Samsung, and Micron—control nearly the entire global HBM market. As of 2022, SK Hynix held 50 per cent of the market, Samsung 40 per cent, and Micron 10 per cent. SK Hynix and Samsung are projected to maintain similar shares in the coming years, while Micron aims to increase its market share to 20–25 per cent by 2025.
HBM’s high value is driving manufacturers to prioritise its production, with the chip expected to account for over 20 per cent of the total memory chip market by value in 2024, potentially exceeding 30 per cent in 2025, according to TrendForce.
The Challenges Of Manufacturing HBM
Creating HBM chips is a complex and precise process. These chips are essentially stacks of memory layers, each ground to the thickness of half a strand of hair and interconnected through precisely drilled holes for electrical wiring. This advanced packaging process is costly and prone to failures, making the chips significantly more expensive than traditional memory components.
“You have a lot more failure points when you try to make these devices. It’s almost like building a house of cards,” Hutcheson explains.
The Bigger Picture
The US export restrictions reflect ongoing geopolitical tensions over technology dominance. While China continues to push for self-sufficiency in semiconductor production, the new rules underscore the strategic importance of memory technologies in AI and advanced computing. For now, these controls are likely to create hurdles for China’s AI ambitions, while reinforcing the dominance of established HBM manufacturers in the global market.