Fujifilm plans to spend about 10 billion yen, roughly $63 million, to raise magnetic tape production capacity fivefold by fiscal 2028, Nikkei reported on Wednesday. The money goes to its Odawara plant in Kanagawa prefecture, south of Tokyo, and the target is measured by shipped storage capacity rather than cartridge count: five times the volume the company shipped in fiscal 2025. It would be Fujifilm's largest investment in the medium in about 20 years. The company has not publicly confirmed the plan.
The reason is the least glamorous part of the AI boom. Training runs generate datasets measured in petabytes, and most of that material is read once and then sits untouched for years, kept because regulators, reproducibility rules or future fine-tuning might demand it. A rack of idle disks keeps drawing power and needs cooling. A shelf of cartridges draws nothing. Fujifilm's own comparison puts tape as low as a quarter of the cost per terabyte of hard disk, and the gap has widened as memory and SSD prices climb.
Demand has already turned. The LTO Program, backed by HPE, IBM and Quantum, reported that tape shipment capacity rose 57 percent year on year in the first quarter of 2026, helped by continued LTO-9 volumes and the ramp of LTO-10. That reverses a decline: global LTO shipments fell 9 percent to 160.3 exabytes in 2025.
The product itself keeps getting denser. Fujifilm began shipping a 40-terabyte native LTO Ultrium 10 cartridge in January, which holds up to 100 terabytes at a 2.5-to-1 compression ratio, about a third more than the previous 30-terabyte version. The company has said its work on strontium ferrite particles could eventually support 400 terabytes per cartridge, a claim it has not demonstrated in a shipping product. LTO-11 and LTO-12 remain on the roadmap.
Supply is concentrated. Fujifilm says it holds the largest share of the data storage tape market, and Sony is effectively the only other supplier of the media, so a fivefold increase from one of two players raises the ceiling for the entire archive tier. That matters for hyperscalers building out cold storage as fast as they build training clusters.
The business is already visible in the accounts. Fujifilm's Advanced Functional Materials segment, which houses data tape, posted 42.8 billion yen in revenue for April to June 2026, up 14.1 percent year on year.
The counterargument is that tape only does one job. It reads sequentially, so pulling a specific file takes minutes and a robotic arm, not milliseconds; nobody puts tape in front of a user-facing workload. Revenue also depends on a small group of hyperscale buyers and library vendors, so a slowdown in AI capital spending would leave new coating lines idle. Rival long-term storage approaches, from Microsoft's quartz glass Project Silica to ceramic and DNA writing, remain laboratory work with no shipping supply chain.
What it means: as AI shifts from buying compute to keeping what that compute produced, the archive tier becomes a real market with a real bottleneck. The cheapest place to park a petabyte is still a plastic cartridge on a shelf, and two Japanese companies control whether you can buy one.
Comments (0)
Log in to join the discussion
Log InNo comments yet