AI demand is reshaping the storage industry.
Data storage has become one of the biggest winners of the AI boom. As of August 2026, shares of Seagate and Western Digital had risen approximately 470% and 600% over the preceding 12 months, respectively, while SanDisk had climbed more than 3,000%.
The rally reflects expectations that AI will require far more data storage. Solid-state drives (SSDs) provide the speed needed to access data quickly, while hard-disk drives (HDDs) offer a lower-cost way to store enormous amounts of it. Tight capacity, stronger pricing and new long-term customer agreements have also raised the possibility that an industry historically defined by boom-and-bust cycles is becoming more durable.
But is the data-storage industry truly becoming more stable, or is today’s strong demand and pricing simply another temporary period of favorable market conditions? This uncertainty has been visible in the sharp rallies and pullbacks experienced in recent months by SanDisk, Seagate and Western Digital. We interviewed multiple Third Bridge experts to understand what is driving demand, whether long-term agreements can reduce cyclicality, and which storage technologies are best positioned to benefit.
Why AI needs both SSDs and HDDs
AI creates storage demand at several stages of the data lifecycle. SSDs are used when information must be accessed quickly, including while AI models are being trained or generating responses. HDDs are slower, but provide a much cheaper way to retain the enormous quantities of data those models consume and produce.
Third Bridge experts say the growing use of inference, the process of generating answers from a trained model, is making the need for speed more important. As models are used more frequently and prompts grow longer, more data must move quickly between storage and processors. According to one expert, this has allowed SanDisk to redirect supply from consumer products toward higher-priced data-center SSDs.
At the same time, AI-generated videos, model outputs, and other large datasets can be stored and later reused to train or improve future models. Western Digital expects HDD exabyte demand to grow by more than 25% annually over the next three to five years.¹ A former strategy executive at the company said this forecast could prove conservative if enterprise AI adoption accelerates.
The cost difference between the two technologies remains one of the main reasons HDDs are used for large-scale storage, with a former storage procurement executive estimating that SSDs currently cost 10 to 15 times more per terabyte than HDDs.
Can long-term agreements make storage less cyclical?
Storage suppliers have historically built production around customer forecasts rather than firm orders. When demand fell short, excess inventory forced suppliers to reduce prices and left factories operating below capacity. Today, the market faces the opposite problem. HDD suppliers have much of their future output committed, while NAND manufacturers are operating close to full capacity. This scarcity has prompted customers to secure future supply through long-term agreements (LTAs), giving suppliers earlier visibility into demand.
A former Western Digital procurement executive said Seagate and Western Digital could raise prices by another 3-4% during the current shortage, but are instead offering more predictable pricing in exchange for customer commitments extending into 2028 and 2029. However, another former executive cautioned that HDD agreements have not traditionally required customers to purchase their full forecasted volumes, leaving them with some flexibility to reduce or defer orders.
At its August 2026 Investor Day, SanDisk said its New Business Model agreements include committed volumes, minimum financial guarantees, and structured pricing. Its eight agreements represent approximately half expected bits in FY27 and two-thirds in FY28.²
Third Bridge experts differ over how durable those protections would be for SanDisk during a downturn. One said LTAs should prevent hyperscalers from changing orders “on the fly,” while a former director argued that they remain products of scarcity. If supply becomes readily available, customers may seek to renegotiate, and suppliers may be reluctant to enforce unfavorable terms against strategic customers.
Technology will determine who captures the AI storage boom
With much of the HDD industry’s future output already committed, increasing the amount of data each drive can hold is critical to meeting additional demand. Heat-assisted magnetic recording (HAMR) is a newer technology that allows manufacturers to store more data on each drive. A former Seagate executive noted that moving from a 24TB drive to a 30TB HAMR drive provides approximately 25% more capacity within roughly the same physical footprint, without a comparable increase in power and cooling requirements.
Third Bridge experts estimate that Seagate is one to two years ahead of Western Digital in developing HAMR, while Western Digital continues to rely on its more mature energy-assisted perpendicular magnetic recording (ePMR) technology. However, one expert said Seagate’s manufacturing yields remain below those of Western Digital’s ePMR drives. This could temporarily give Western Digital higher margins. If Seagate improves its yields, HAMR’s greater density should reduce its cost per terabyte below ePMR, converting its technology lead into a manufacturing-cost advantage.
HAMR execution should also determine whether HDDs preserve their long-term cost advantage as NAND manufacturers increase SSD density and reduce cost per terabyte. Because SSDs already provide faster access and require less physical space, HDD manufacturers must continue increasing drive capacity to remain the economical choice for large-scale storage. One former procurement executive said Seagate needs to reach approximately 60TB by late 2028, with Western Digital following in 2029, before both companies progress toward 100TB drives during the 2030s.
Third Bridge experts identified quad-level cell (QLC) NAND as a nearer-term opportunity for SanDisk. QLC and TLC are types of NAND flash used in SSDs. QLC stores four bits in each memory cell, compared with three for TLC, allowing manufacturers to increase SSD capacity and reduce cost, although it carries reliability tradeoffs. One expert expects QLC to account for approximately half of enterprise SSD demand by the end of the decade. High-bandwidth flash (HBF) is a longer-term opportunity designed to place lower-cost NAND closer to processors, reducing the time and power required to move data. However, experts do not expect meaningful volume before 2029 and say customer qualification remains the critical test.
Conclusion
Taken together, Third Bridge interviews suggest that AI is creating genuine demand for both SSDs and HDDs. Long-term agreements and capacity discipline will impact whether the industry becomes less cyclical, while technology execution will help dictate how the resulting growth and profits are divided among SanDisk, Seagate and Western Digital.
Relevant transcripts
1. 2026/03/20 Seagate – Riding the AI Storage Wave – Durable Supercycle or Peak Demand?
2. 2026/04/30 Seagate Technology – Can AI Demand & HAMR Extend the HDD Upcycle?
3. 2026/07/21 Western Digital – AI Demand, HDD Scarcity & Margin Sustainability
4. 2026/07/16 SanDisk – AI Inference, High-bandwidth Flash & Durability of the NAND Upcycle
5. 2026/03/24 SanDisk – AI Infrastructure Demand & Structural Shift in the NAND Market
6. 2026/05/05 HDD Industry – Pricing Power, Cloud Demand & Next Leg of the Cycle
References
2. Sandisk's IR: SanDisk details its growth strategy and long-term financial model
All insights in this article are based on information provided by Third Bridge experts.
For media enquiries, please contact: comms@thirdbridge.com
