High-Capacity SSDs Will Enable AI Workloads But Also Drive HDD Demand

Emerging Trends in SSD Technology: High-Capacity Storage for AI Workloads

The recent FMS conference in Santa Clara showcased the latest advancements in SSD technology, with a focus on high-capacity storage solutions. Almost all SSD companies introduced high-capacity SSDs, many exceeding 200TB, with some promising large form factor SSDs with 1PB capacities in the future. These SSDs utilize higher logical density, four-bit per cell, or QLC flash memory, and multiple chips to achieve these impressive capacities.

SSD Differentiation for AI Workloads

The SanDisk keynote highlighted the differentiation of SSDs for AI workloads, categorizing them into two types: fast eSSDs for high-bandwidth DRAM memory, such as HBM, and high-capacity storage eSSDs for higher-performance data lakes. The latter offers a significant advantage over traditional HDDs, providing faster access to data and improved overall system performance.

High-Capacity SSDs: A New Era in Storage

SanDisk’s high-capacity eSSD, expected to reach capacities of up to 256TB by 2026, is a notable example of this emerging trend. The company also outlined a path to a 512TB version by 2027 and a 1PB product in the future. Other SSD companies, including Kioxia, Micron, and Samsung, are also exploring high-capacity QLC SSDs for near-storage AI applications, often utilizing the E3.S form factor to enable higher capacities.

Memory and Storage Hierarchy: Trends and Developments

The Silicon Motion keynote illustrated the traditional memory and storage hierarchy, highlighting trends in NAND flash supporting GPUs directly. This includes announcements from SK Hynix and SanDisk at the FMS, as well as the introduction of an ultra-high-capacity SSD layer for warm storage in AI applications. As a supplier of SSD controllers, Silicon Motion is well-positioned to address the growing demand for high-capacity storage solutions.

Cost and Performance Considerations

While high-capacity SSDs offer significant advantages in terms of performance and rack space, they currently come at a higher cost per storage capacity than HDDs. According to projections from Western Digital’s investor day, flash memory is expected to remain approximately 6X more expensive per storage capacity than HDDs for some time. However, the growth of AI workflows is driving demand for both high-capacity SSDs and HDDs, with the latter expected to continue providing lower-cost storage for longer-term data retention.

Market Projections and Growth Opportunities

Coughlin Associates has updated its projections for storage capacity shipped, anticipating increased growth in SSD storage and a reduction in HDD capacity shipments. The firm’s projection for HDD storage capacity prices out to 2030 suggests a continued decline in costs, with SSD prices expected to follow suit. Assuming a 6X cost differential between SSDs and HDDs, the projected revenue for SSDs and HDDs in 2030 is $93B and $55B, respectively, driven by the growing demand for storage solutions to support AI workflows.