[Edaily Reporter Choi Oh-hyun ] SamsungElectronics and SK hynix have unveiled different strategies targeting the next-generation artificial intelligence (AI) memory market. While SamsungElectronics is focusing on securing a technological edge by leveraging its 3D stacking technology, SK hynix has adopted a strategy of expanding the market itself by establishing standard specifications and a collaborative ecosystem.

According to industry sources on the 6th, SamsungElectronics and SK hynix each unveiled their next-generation AI memory technologies at “FMS (Future of Memory and Storage) 2026,” the world’s largest memory and storage technology conference, held in the U.S. from the 4th to the 6th (local time). These announcements held particular significance as they went beyond simply showcasing new technologies to offer insights into each company’s next-generation business direction for the AI era.

SamsungElectronics highlighted “zHBM,” which vertically stacks High-Bandwidth Memory (HBM) on top of an AI accelerator. Unlike conventional designs where HBM is placed beside the accelerator, this structure stacks it on top to reduce data transfer distances. While several companies are developing related technologies, SamsungElectronics is the first to unveil a prototype.

This move is interpreted as reflecting Samsung’s determination to reclaim “technological leadership” by securing a head start on next-generation 3D memory structures, especially given that it has ceded market leadership in HBM to SK hynix. In particular, since zHBM requires advanced packaging technology, it also highlighted the strengths of an integrated device manufacturer (IDM) that encompasses not only memory but also system-on-chip (SoC) design, foundry services, and packaging.

The key variables for zHBM are its actual mass-production feasibility and customer adoption. The ability to stably resolve issues such as heat generation, power efficiency, and packaging—which arise from zHBM’s structural characteristics—and achieve mass production, as well as whether this leads to securing actual AI accelerator customers, are expected to be crucial in proving its technological superiority.

Meanwhile, SK hynix has partnered with SanDisk to establish standard specifications and build an ecosystem for High-Bandwidth Flash (HBF). The two companies agreed to develop products independently but to jointly establish specifications that customers can use universally. Google and Tenstorrent have also joined the consortium.

An industry insider stated, “While each company can supply its own products, it will be difficult to establish a single market standard without common specifications,” adding, “Even as competitors, they have no choice but to cooperate as needed to grow the next-generation NAND storage market.”

The shift in the focus of AI memory competition from training to inference is also cited as a key factor behind this strategy. In an inference environment, AI accelerators, software, and storage must be seamlessly integrated; therefore, securing standard interfaces and an ecosystem has become just as important as the performance of individual products. Once standard specifications are established, customers will be able to adopt products without being locked into a specific vendor, and manufacturers will also find it easier to expand their customer base. Going forward, it will be crucial to see whether early adopters like Google implement this technology in actual products and whether the ecosystem expands to include AI accelerator vendors.

HBF is a next-generation NAND-based memory that is gaining attention as storage for AI inference. While HBM—which processes large volumes of data per second—was key for AI training, the role of storage that can store vast amounts of data and retrieve it quickly becomes more significant for inference. HBF is based on the concept of placing existing enterprise solid-state drives (eSSDs) close to AI accelerators to reduce data transfer distances and lower latency.