South Korea’s SK Hynix expects the market for high-bandwidth memory (HBM) chips used in artificial intelligence to grow 30% annually until 2030, as cloud computing giants keep increasing their demand for AI. The company noted that Amazon, Microsoft, and Google will likely spend even more on AI in the coming years, which will help boost demand for HBM chips . SK Hynix sees chip demand rising as AI use grows SK Hynix’s head of HBM business planning, Choi Joon-yong, said that demand for artificial intelligence from end users remains “firm and strong.” He says a clear and measurable connection exists between the pace of AI infrastructure growth and the volume of high-bandwidth memory (HBM) chips purchased by technology companies. AI systems will require more powerful and efficient hardware to handle increasingly complex workloads as they become more sophisticated. This creates a cycle where advances in AI technology lead to greater demand for HBM, ensuring memory manufacturing continues evolving alongside the industry’s rapid expansion. HBM is a specialized form of dynamic random access memory (DRAM) engineered for extremely high data processing speeds while maintaining lower energy consumption. Its design shortens the distance data must travel between layers to reduce latency and boost transfer speed by stacking multiple memory chips vertically within a compact package. Choi also outlined the company’s expectation that the custom HBM product market will reach tens of billions of dollars by the year 2030. These customized solutions boost speed, efficiency, and scalability because they are tailored to match specific AI models’ architecture or optimize performance for specialized workloads. He emphasized that the estimates are intended to be lowball figures, as they consider practical constraints based on the long-term sustainable energy supply. Perhaps more importantly, the company insists massive ongoing investments in AI from Amazon, Microsoft, and Google (not to mention Facebook and Apple) will ensure an upward trajectory for HBM, limitations notwithstanding. As companies seek to redefine the machine learning frontier, demand for high-performance, energy-efficient memory solutions will place HBM at the centre stage within the vanguard of the AI hardware revolution. Customization and HBM4 chips will speed up market growth Instead of producing generic, one-size-fits-all chips, the HBM industry is now developing highly customized products designed to meet the exact specifications of individual clients. This change will be most visible in the upcoming HBM4 generation being introduced by SK Hynix, Samsung, and Micron. These new HBM4 chips have a customer-specific “base die” as the central control layer for managing the memory stack. This design allows manufacturers to fine-tune performance to match the precise architecture and operational demands of a customer’s systems. However, replacing the custom chip with a competing product is extremely difficult and often costly once a company integrates a supplier’s product into its system. This dependency gives memory suppliers greater influence in the market because it strengthens long-term relationships between chipmakers and their clients. Nvidia is SK Hynix’s largest customer for HBM chips designed to match the company’s demanding AI workloads and GPU architectures. SK Hynix also provides standard designs that offers high performance, but lack the deep customization for smaller buyers who may not require the same optimization levels that offers high performance. Choi Joon-yong believes that as more companies recognize the advantages of tailoring memory to their unique performance, demand for customized chips will grow even further in the coming years. The market still faces short-term challenges, even though the long-term outlook for HBM4 remains strong. Samsung warned that prices might drop soon because the supply of its HBM3E chips could temporarily exceed demand. Still, SK Hynix is confident it can continue delivering competitive products that meet evolving customer needs. The company believes that the arrival of HBM4 and growing interest in customization will offset any short-lived imbalances between supply and demand. On the policy front, geopolitical developments are also shaping the semiconductor market. U.S. President Donald Trump has proposed a 100% tariff on imported chips from countries that do not have semiconductor manufacturing facilities in the United States. SK Hynix and Samsung will only feel a pinch of such measures because they have invested heavily in U.S. operations. If you're reading this, you’re already ahead. Stay there with our newsletter .
