With the rapid growth of the AI inference market, the storage industry is gradually entering the "post-HBM era" - not only are storage giants such as Samsung and SK Hynix promoting the sixth generation of HBM, but new technologies such as HBF are also "eyeing" the market, trying to participate in this wave of AI storage power competition.
According to reports from Digitimes, South Korean Financial News, and other media outlets, Samsung, SK Hynix, and SanDisk... Storage manufacturers are increasingly investing in the research and development of HBF technology. At the recent 2025 OCP Global Summit, SK Hynix launched a new product line called the "AIN series," which includes HBF. Prior to this, the company signed a memorandum of understanding with SanDisk , stating that the two parties would jointly develop HBF technical specifications.
At the same time, Samsung has also started early concept design work for its own HBF products.
HBF, short for High Bandwidth Flash, is a product made by stacking NAND flash memory, similar in structure to HBM (Hyperband Memory Block), which stacks DRAM chips. Kim Jung-ho, a professor at the Korea Advanced Institute of Science and Technology (KAIST) known as the "father of HBM," points out that while NAND flash memory is slower than DRAM, it offers approximately 10 times the capacity , which could be crucial for supporting next-generation AI.
In February of this year, SanDisk first proposed the HBF concept, positioning it as an innovative product combining the capacity of 3D NAND and the bandwidth of HBM. Although there is still a distance between concept and actual implementation, in the aforementioned collaboration between SK Hynix and SanDisk, the two parties have established a general direction—to launch the first batch of HBF memory samples in the second half of 2026, and expect the first batch of AI inference device samples using HBF to be available in early 2027.
Shin Young Securities, South Korea Data shows that the HBF market is expected to reach $12 billion by 2030. Although this is only 10% of the HBM market size (approximately $117 billion) in the same year, it is expected to complement HBM and accelerate its growth.
Due to its relatively "light speed, heavy capacity" characteristics, HBF has also been evaluated as "auxiliary memory" to compensate for the insufficient capacity of HBM. The logic behind its demand can be traced back to the rise of AI inference. As large language models such as ChatGPT and Gemini have evolved into multimodal models, they can generate not only text but also images and videos, which requires much more data than in the past when only text was generated.
GF Securities It is pointed out that with the rapid growth in demand for AI inference, the deployment of lightweight models is driving a rapid increase in storage capacity demand, and the overall demand is expected to surge to the hundreds of exabytes in the future.
Against this backdrop, capacity has become a bottleneck limiting the full potential of computing power. At SK Hynix's recent Q3 earnings call, the company identified its hyperscale data centers, which are highly dependent on HDDs, as a key factor. Customers have begun to shift towards enterprise-grade SSDs based on high-capacity QLC. According to a TrendForce report, the focus of AI infrastructure deployment in the next two years will be more on supporting high-performance inference services, and the supply of enterprise-grade SSDs will be tight in 2026, with this demand surge expected to continue into 2027.
Currently, the storage industry is experiencing a market boom described as a "supercycle." Shanghai Securities stated that the rapid application of AI inference is driving real-time access and high-speed processing of massive amounts of data. Demand is driving HDD and SSD suppliers to actively expand the supply of high-capacity storage products. Furthermore, with the recovery of CSP (Consumer Service Provider) deployment momentum, demand for DDR5 products continues to strengthen, and DRAM procurement demand from CSPs is expected to grow significantly in 2026. Due to capacity constraints from overseas manufacturers, the upward trend in storage prices is expected to continue in Q4 2025, and the sustainability of this cycle is viewed favorably.
(Article source: CLS)
