Hard disk drives

getty

I did an analysis recently to try and determine just how much the AI data center buildout has impacted the demand for hard disk drive (HDD) storage capacity. This is tricky because just before the surge in demand for storage for AI there was a surge in demand for storage in 2021 due to Covid and then a big slump in demand in 2022 and 2023 as data centers used up the inventory that had accumulated during Covid.

Capacity demand rebounding in 2024 and through 2025, when excess inventory was consumed and there was both a return to ordinary storage demand but also additional demand due to the buildout to support AI training.

To make this estimate I tried to get the best fit to HDD capacity shipments from 2011 through 2019. The result of this best fit plotted over historical data from 2011 through 2025 is shown below as well as Coughlin Associates (I am the owner of Coughlin Associates) projections for continued capacity growth in 2026 through 2030.

The chart also shows the estimates of additional demand above the historical growth rate that we largely attribute to AI driven demand. In 2026 we estimate that storage capacity demand without the AI buildout will be about 1,654 Exabytes (EB) with about 363EB of additional storage demand due to the AI buildout, or about 18% of the total expected capacity shipments in 2026. If we continue the projected growth, by 2028 AI additional HDD capacity demand will be about 43% of total shipments and in 2030 AI additional HDD capacity demand will be about 58%.

In AI workflows, SSDs using NAND flash are used for primary storage that supports the HBM memory with data for immediate AI training or inference. HDDs have lower read and write performance but are considerably less expensive in $/terabyte (TB) and are used as secondary storage to keep older training data, training logs and other content created during the AI training or inference. As a consequence, HDD AI storage demand tracks SSD storage demand but with a higher multiplier as a repository for older data, which grows over time. Currently more than 85% of data center data lives on HDDs.

Much of the growth in HDD capacity will be met by shipping larger HDDs rather than by shipping many more HDDs. We project that the average shipping HDD capacity will increase by about 2.7 times between 2025 and 2030. We estimate that the number of HDDs shipped in 2030 will be about 13% higher than in 2025, which is 19% less than the number of HDDs I estimate shipped in 2022.

Seagate CEO CEO Dave Mosley recently said that the company’s focus is on increasing storage density (using HAMR technology to increase terabytes per platter) rather than increasing the physical number of drives produced, as diverting resources to build new factories would slow down technological innovation. There are projections that HDD capacity will triple by 2030, which is close to our 2.7 times increase for the average shipping HDD capacity.

If current AI driven growth trends continue AI driven demand for HDD capacity will exceed shipping non-AI demand by 2030. This will be driven mostly by increased HDD capacities rather than shipping more HDDs.