At the recent FMS conference in Santa Clara, almost all SSD companies have introduced high -capacity SSD companies, with many more than 200TB promises, with high -formula factories with 1PB capacity in the future. These SSDs utilize higher logical density, four bits per cell or Flash QLC and many chips to achieve these abilities.
Keynote Sandisk has differentiated some different uses for SSDs to support AI workload. A guy is fast essds to support high -range Dram memory, HBM. The other type is Essds High-Capacity Storage for a higher performance lake than HDDs can offer. These two types of SSDs are shown below.
Sandisk FMS Keynote about branching SSDS for AI
Ron Denison
The following slide shows the Sandisk’s high capacity ESD. This is a QLC BICS8 NAND FLASH U.2 and EDSFF FORCT SSDN that is expected to have up to 256TB by 2026.
Sandisk 256TB SSD
Ron Denison
Keynote Sandisk showed a route in a 512TB version by 2027 and 1PB product at some point in the future. The complex image below shows the Kioxia, Micron and Samsung announcements of high capacity QLC SSDS. All SSD companies are investigating the product for AI applications near storage. Many of the higher capacity products use the E3.S format factor, which can keep more Nand chips to allow higher capabilities.
Complex Kioxia, Micron and Samsung high capacity SSD
Tom Coughlin
Keynote Motion Silicon gave a representation of a traditional representation of the Memory and Storage Hierarchy that show trends for GPU supporting NAND directly supporting, such as HBF announcements from SK Hynix and Sandisk in FMS. It also shows an extremely high -capacity SSD layer to support hot storage for AI applications. Provides silicon motion controllers for SSD.
The hierarchy of silicon memory and storage
Tom Coughlin
The highest SSDS capacity can take less rack than HDDs and offer higher performance than HDD offers. This may be an advantage for AI’s training and conclusion with RAG, but Flash memory is now about 6 times more accurate per HDD storage capacity and is expected to remain so for some time in the future, as shown in the picture below from WDC investors day last February. Seagate has similar trends. For example, by 2026 44TB HDD should be in production, an increase of 38% of the largest HDDs available today. This is due to the fact that the expected increase in HDDS storage capacity has been accelerated with the introduction of HDD HAMR to match about increasing SSD capabilities.
WDC Investor Day Views for HDD trends and SSD $/TB
Wdc
Consequently, we believe that these larger SSDs will be used for data lakes that feed directly to GPUs for AI work flows. However, HDDs will continue to provide lower cost storage for long -term data retention and, therefore, these higher capacity SSDs will result in greater development of HDDs.
Coughlin Associates has updated its views on storage capacity sent for HDD, SSD and magnetic movieshown below. This new promotion increases our expectations for the development of SSD storage from previous versions in 2030 with some reduction in HDD capacity missions.
History and predicted annual storage capacity for HDD, SSD and magnetic movie
Coughlin Associates
The promotion of the Coughlin Associates for HDD storage prices by 2030 is presented below.
History and Projected $/GB for HDD
Coughlin Associates
Assuming SSDs remain at 6 times the cost per storage capacity by 2030 and the HDD price per GB of 0.0051 dollars in 2026, the Flash Nand price would be about $ 0.031 per GB. With views for the SSD and HDD shipping capacity of about 3.0ZB and 10.7ZB, SSD and HDD revenue in 2030 is $ 93b and $ 55B respectively. The rising storage vessel, driven by AI, is expected to lead to a significant increase in revenue for both HDDs and SSDs.
FMS 2025 shows high capacity SSD development, up to 1PB as well as high capacity HDD to support AI work flow development.
