Seagate announced shipments of Exos M hard disk drives, HDD, samples to customers with up to 36TB storage capacity. These HDDs use the company’s Mozaic 3+ heat assisted magnetic recording, HAMR, technology, where lasers are used to reduce the magnetic fields from the write head required to write a high-densities on stable high coercivity magnetic recording media. Seagate also said that it is ramping up production of Exos M HDDs with up to 32TB to a leading cloud service provider.
These 36TB HDDs have 10-disks and so they provide 3.6TB per disk. This likely results in about 1.8TB per square inch areal density, a 13% advance over the prior 32TB HDDs. The chart below shows Coughlin Associates history of shipping HDD areal densities. This chart shows a significant increase in HDD areal densities in 2023 and 2024 and if this trend continues, it indicates that HAMR allows HDD areal density growth, and thus the increased capacity per HDD, could grow as much as 15-20% annually. Seagate projects that 10TB per disk is attainable, enabling a 100TB 10-disk HDD.
Seagate says that this product will provide significant advantages for data center storage, in terms of scale, total cost of ownership and sustainability. The company says that these drives can provide 300% more storage capacity within the same data center footprint, a 25% cost reduction per TB and 60% power reduction per TB compared to prior generation products, for a 30TB Exos M Mozaic drive compared to a 10TB Exos X10 HDD. Seagate says that 10TB is a common drive capacity currently used in data centers and needing upgrades today.
A Recon Analytics survey, commissioned by Seagate and conducted in November 2024 with 1,062 participants, showed some interesting trends on how AI is creating increased digital storage demand, and thus the need for cost effective storage, such as HDDs. 61% of infrastructure buyers who predominately use cloud storage for AI data management said they expect storage requirements to at least double by 2028 due to retention times increasing from 6 months to forever, 73% using daily or weekly LLM checkpointing, and 80% deeming data replication for AI very or moderately important.
95% of storage buyers, using AI or planning to, say they are taking measures to accommodate the growing storage requirements, including 61% adopting more scalable storage, 56% implementing data management software, 49% using compression techniques and 55% upgrading existing storage infrastructure. Recon Analytics’ research finds that wherever AI is adopted, existing storage practices will need to be upgraded to realize the full potential of AI.
Cloud storage is expected to remain the main storage vehicle for AI with 65% of data stored in the cloud versus in-house in 2024 and increasing to 69% by 2028. 90% of respondents who have adopted AI believe longer data retention improves the quality of AI outcomes. Of which 93% claim data retention requirements have changed due to the implementation of AI and the ability to refine models including checkpoints. The importance of data replication to a company’s AI data management strategy also increases the amount of storage a company uses.
Saving interim modelling results is also driving storage demand. 73% of respondents say AI training is driving increased data storage as they are backing up their previously saved checkpointing data on a daily to weekly basis. Compounding the storage impact of saving AI checkpoints, infrastructure buyers also need to factor in how long they will save each checkpoint as part of the LLM training. Of those respondents saving checkpoints daily (28% of respondents), 32% are retaining data for more than 12 months while 29% are retaining for six to 12 months.
Seagate announced that they are shipping 36TB HAMR evaluation drives to customers. These drives offer significant scale, power savings and storage cost advantages over prior HDD generations. A study shows AI will drive demand for storage solutions, such as these.
