TAIPEI (Taiwan News) — Taiwanese companies are set to benefit from High Bandwidth Flash, a next-generation memory technology for artificial intelligence servers, CTEE reported Thursday.
AI servers are handling increasingly large models and heavier workloads. Existing memory systems are starting to reach their limits in speed and storage capacity.
High Bandwidth Flash, first proposed by SanDisk and now being developed by SK Hynix and Samsung, is emerging as a potential solution. It aims to deliver faster data access and larger storage for advanced AI applications.
Although Taiwan is not setting standards for HBF, analysts say mass adoption could create opportunities for local technology companies that design controllers and provide advanced packaging services.
Phison Electronics Corp. is considered a leading Taiwanese contender for early HBF design wins. The company has long worked with AMD and is known for designing storage controller chips used in AI server systems.
Analyst Cheng Ting-yi (鄭廳宜) has also previously noted that Phison is co-developing next-generation storage technologies with global partners and holds a full intellectual property portfolio for NAND controllers, positioning it to expand into HBF.
Powertech Technology Inc., one of the world’s largest memory packaging and testing providers, is also seen as strategically important. Like high-bandwidth memory, HBF relies on advanced packaging.
The packaging work is more complex than that used in traditional storage products and may be outsourced to major contract packaging firms that handle large AI server projects. Taiwanese packaging houses could therefore take on additional work from US and Korean suppliers once production ramps up.
Global memory makers are investing heavily in HBF as the next phase of the AI memory race begins.
SanDisk has made HBF central to its strategy. The company introduced an AI-focused HBF design in February, signed an MOU with SK Hynix in August to support standardization, and plans to begin sampling late next year before launching HBF-based inference systems in early 2027.
Samsung has begun early concept development but has yet to finalize specifications or a production timeline. Analysts expect the company to join the competition, given its strength in NAND, HBM, and advanced packaging technologies.
