While most of the tech industry spends its time obsessing over GPU clusters and high-bandwidth memory shortages, hard drive makers are quietly running their factories at full capacity. AI isn't just hungry for raw compute power. It generates mountains of persistent data that someone has to store somewhere—and fast-spinning magnetic platters remain the most cost-effective way to do it at scale.
Seagate chairman and CEO Dave Mosley revealed during the company's Q4 earnings call that datacenter demand now represents roughly 90 percent of its total exabyte shipments. Even more striking, long-term supply agreements with major cloud operators have effectively locked up the vast majority of Seagate's nearline hard drive capacity all the way into calendar 2028.
To keep up with this appetite, Seagate is actively accelerating production of its Mozaic 4 storage platform. Built on heat-assisted magnetic recording (HAMR) technology, Mozaic 4 pushes individual hard drive capacities up to 44TB per drive. Cloud providers aren't waiting around, either. Back in February 2026, Seagate warned that its nearline production capacity was completely booked out for the year and started taking orders for 2027. Now, hyperscale customers are pushing their commitments out through 2028 without shrinking their planning horizons.
The Financial Surge Behind Spinning Disks
The financial numbers paint a clear picture of how aggressive cloud purchasing has become. Seagate reported full-year revenue of $12.2 billion, up 34 percent from $9.1 billion in the previous fiscal year. For its fourth quarter, ended July 3, 2026, the company generated $3.6 billion in revenue—a 50 percent jump compared to the $2.4 billion reported in the same period a year earlier.
Looking ahead to the current quarter, Seagate expects revenue to hit approximately $4.1 billion, give or take $100 million. That guidance represents a 56 percent year-over-year growth rate.
Yet, this revenue surge isn't coming from selling millions of extra physical drives. CFO Gianluca Romano pointed out a crucial operational reality on the earnings call: unit shipments remained essentially flat year-over-year. Instead, capacity growth came almost entirely from packing more disks and recording heads into each drive enclosure.
According to Romano, the number of internal disks and heads per drive box grew by 15 to 20 percent over the past year. Seagate is generating higher revenue not by building more drive chassis, but by engineering higher density into every slot in the server rack.
Why Agentic AI Shifts Storage Requirements
Why are cloud providers locking in hard drive capacity four years ahead of time? The answer lies in how enterprise AI workloads are evolving.
As AI applications shift from static model training to active inference and autonomous agentic workflows, data retention requirements change dramatically. Autonomous AI agents don't just process a query and forget it. They perform tasks across multi-step workflows, requiring context retention, regulatory compliance tracking, and historical audit logs.
Mosley emphasized that these agentic systems generate massive amounts of intermediate data that must be saved for future reuse. When datacenter operators build out these massive environments, they run straight into cost, power, and thermal limits. Running everything on ultra-fast solid-state storage or expensive RAM simply doesn't scale economically.
To address this, cloud architectures rely heavily on tiered storage strategies. A joint white paper published by Seagate and memory manufacturer SK hynix outlines how modern systems manage data across high-performance memory, enterprise SSDs, and high-capacity hard drives as explored further here.
A key factor here is key-value (KV) caching. In agentic AI systems, KV cache holds state context across repeated user interactions and agent execution steps. Recomputing that context every time an agent takes an action drains precious GPU clock cycles. By offloading KV cache data across a tiered storage architecture—placing persistent context onto cost-effective hard drive tiers much like the strategies discussed in our exploration of Optane alternatives—cloud providers avoid wasting expensive GPU power on redundant recomputation. That leaves high-cost compute hardware free to focus on revenue-generating inference work.
Roadmaps and the Road to Mozaic 5
Seagate's technology roadmap shows no signs of slowing down. While Mozaic 4 is currently being deployed to meet cloud demand at capacities up to 44TB, the company is already laying the groundwork for its next platform generation.
The upcoming Mozaic 5 platform remains on track for initial qualification shipments in late 2027. Mozaic 5 is targeted to deliver areal densities exceeding 5TB per platter. Achieving densities higher than 5TB per platter requires continued advances in HAMR technology, media grain size, and read-write head precision.
Because Seagate's solid-state drive (SSD) business represents a relatively small portion of its overall revenue, the company's enterprise trajectory rests heavily on HAMR innovation and nearline hard drive dominance. For hyperscalers building out data infrastructure for the next decade, multi-terabyte mechanical drives remain an irreplaceable layer of the storage hierarchy.
With nearline exabytes locked up through 2028, cloud infrastructure builders are signaling that mechanical storage isn't a bridge to the past—it's a critical component of the AI buildout.
