Three companies make essentially all hard drives on earth: Western Digital (42% market share), Seagate (40%), and Toshiba (18%). When WD CEO Irving Tan says they're "pretty much sold out for calendar 2026" with "firm purchase orders from top seven customers" and long-term agreements extending to 2028, he's describing nearly half the global supply being locked up.
Lead times for enterprise nearline drives have gone from a few weeks to over 52 weeks. HDD average selling prices are at their highest point since 1998. And this is happening despite total exabyte shipments hitting records: WD alone shipped 215 EB in the most recent quarter, up 22% year over year. The drives are moving faster than ever and still can't keep up.
Why now? Hyperscaler capital expenditure for 2026 is approaching $700 billion. Amazon announced $200 billion in capex, the largest single-company infrastructure spend in history. Google guided $175-185B (up 140% from 2025). Meta guided $115-135B. About 75% of all of this goes to AI infrastructure: GPUs, servers, and the storage to feed them.
SSDs can't absorb the overflow. Enterprise SSD prices (30TB TLC drives) surged 257% in nine months, from about $3,000 to nearly $11,000. HDDs went up only 35% in the same period. The cost ratio between SSDs and HDDs per terabyte widened from 6.2x to 16.4x. Micron's 2026 enterprise NAND supply is already "fully committed." So even companies that would prefer SSDs are buying hard drives because that's what's available at a price that doesn't break the budget.
The technology bottleneck compounds this. Seagate is the only company shipping HAMR (heat-assisted magnetic recording) drives right now, their Mozaic 3 platform at 30TB per drive. The next step, 40TB drives using Mozaic 4 , starts volume production in the first half of 2026. Western Digital delayed its HAMR entry to the second half of 2026, citing prohibitive per-unit costs at low volume. HAMR requires a fundamentally different manufacturing process: iron-platinum media, glass substrates instead of aluminum, and a near-field transducer laser integrated into every read/write head. You can't just flip a switch and double capacity.
The comparison that keeps coming up is Chia coin in 2021, when cryptocurrency miners panic-bought hard drives and the network storage grew from 600 petabytes to 10 exabytes in a single month. Prices spiked and crashed when the speculation ended. The difference here is that the buyers are Microsoft, Google, Amazon, Meta, and Oracle, with purchase orders running through 2028, building physical data centers (Meta's Hyperion project in Louisiana: 5 GW capacity, $27 billion; the Stargate project in Texas: 7 GW planned, $400 billion ). This isn't speculative hoarding. The demand is structural, funded by $121 billion in new debt issuance in 2025 alone.
WD saw this coming. They spun off their flash/SSD business into SanDisk Corporation in February 2025, making Western Digital a pure-play hard drive company for the first time. Seagate's FY2025 revenue hit $9.1 billion (up 30%), with net profit jumping from $335 million to $1.47 billion. WD's most recent quarter was $3.02 billion in revenue, with gross margins at 43.9%. These companies are printing money right now.
The overcapacity question is real though. Morgan Stanley and others have warned about an AI infrastructure bubble. Grid interconnection delays average 7 years. The enterprise software sector has lost about $2 trillion in market cap since late January 2026 on fears AI will cannibalize existing products. The Magnificent Six lost over $1.35 trillion in a single week in early February. Everyone agrees AI is transformative. The question is whether $700 billion a year in spending will generate returns before the debt comes due.