SK hynix has unveiled a new memory strategy aimed at addressing the rapidly growing data processing demands of the artificial intelligence (AI) era. The company is looking to strengthen its leadership in next-generation AI infrastructure by building an integrated memory ecosystem that combines High Bandwidth Memory (HBM), DRAM, NAND flash and SSDs into a unified architecture.
The announcement was made at FMS 2026, one of the world's largest memory and storage conferences, held in Santa Clara, California. During the event, SK hynix presented its vision for future AI memory architectures alongside its latest memory technologies designed for AI data centers.
According to the company, the AI industry is evolving beyond generative AI toward the era of AI agents, where increasingly complex inference workloads require far more efficient memory systems. In this environment, relying on a single high-performance memory solution is no longer sufficient. Instead, SK hynix believes that a hierarchical memory architecture, in which different types of memory are optimized for specific workloads, will become the foundation of future AI computing.
At the center of this strategy is High Bandwidth Flash (HBF), a next-generation memory technology designed to bridge the performance gap between HBM and conventional SSDs. By applying Through-Silicon Via (TSV) technology to NAND flash, HBF is expected to deliver significantly higher bandwidth than existing storage products while maintaining the large-capacity advantages of NAND.
The company expects HBF to serve as an intermediate memory layer, enabling faster data movement between ultra-high-speed HBM and high-capacity SSDs, thereby improving the efficiency of AI inference workloads.
SK hynix and Sandisk are currently leading efforts to establish industry standards for HBF, while Google and AI chip developer Tenstorrent have joined the ecosystem to accelerate technology development. Industry observers expect demand for HBF to increase as AI inference applications continue to expand.
The company also showcased its latest AI memory portfolio during the exhibition. Featured products included next-generation HBM4, HBM4E and HBM3E, along with DDR5 server memory modules and enterprise SSD solutions. Its high-capacity QLC-based enterprise SSD is scheduled to begin sampling with major global cloud customers this month.
In addition, SK hynix introduced its latest 375-layer 4D NAND technology, which incorporates wafer-bonding technology for the first time. The company said the new NAND significantly improves power efficiency compared with the previous generation and plans to begin mass production of enterprise SSDs based on the technology in the first half of next year.
Another highlight was the demonstration of Compute Express Link (CXL)-based memory expansion technologies. SK hynix showcased an AI agent serving system that enables servers and GPUs to share memory resources, as well as a hybrid memory solution integrating DRAM and SSDs to improve scalability and efficiency in AI data centers.
Beyond introducing individual products, the company emphasized a broader architectural approach to AI infrastructure. As AI models continue to grow in size and real-time inference becomes increasingly important, minimizing data movement across different memory layers will be critical for maximizing both performance and energy efficiency.
Industry analysts believe SK hynix is expanding its competitive edge beyond HBM by investing in emerging technologies such as HBF, advanced NAND and CXL. Rather than positioning memory products independently, the company is increasingly offering integrated AI memory solutions, a strategy expected to strengthen its position in the rapidly evolving AI infrastructure market.
