2 SEP 2026 — LTO tape shipments fell 9 per cent in 2025, from a record 176.5 exabytes to 160.3. The first quarter of 2026 was 57 per cent higher than the same quarter of 2025, so the decline had already reversed before the figure was published. And exactly two companies in the world make the cartridges.
The numbers
The LTO Program, an alliance of HPE, IBM and Quantum, published the 2025 figures on 2 September. Shipped capacity was 160.3 exabytes against 176.5 in 2024, which had been an all-time high. Q1 2026 came in 57 per cent above Q1 2025.
LTO accounts for roughly 85 per cent of tape sales. The new cartridge generation holds 40TB and works in existing drives built for 30TB media. Only Sony and Fujifilm manufacture LTO cartridges.
Storage analyst Tom Coughlin attributes the 16-exabyte fall to the evolving international trade environment, saying it prompted more cautious purchasing as businesses adjusted sourcing and inventory decisions, and expects recovery if the uncertainty resolves. He notes that constrained power availability for data-centre buildouts favours tape, which consumes less power than any other storage medium.
A 9 per cent dip followed by a 57 per cent quarter is a pause
The two figures describe a purchasing pause rather than a technology in decline. A market losing to alternatives does not recover by 57 per cent in a single quarter.
What it looks like instead is deferral. Tape is bought in bulk against a capacity plan, the purchase can be delayed a quarter or two without operational consequence, and it is therefore among the first line items postponed when a procurement team is uncertain about pricing or supply, then bought later against the same plan.
Reading a single annual figure as a trend is the error here, and it is one that has been made about tape nearly every year since 2005.
Exabytes are not cartridges
The measure being reported is shipped capacity, which moves for reasons that have nothing to do with how many units were sold.
A generation change from 30TB to 40TB media means the same number of cartridges carries a third more capacity. Depending on where in the cycle a year sits, exabytes can rise while units fall, or the reverse.
A 9 per cent decline in exabytes is therefore not a 9 per cent decline in tape purchases. Neither the LTO Program nor its analysis separates the two measures. Anyone using this figure to model the market needs the unit numbers, which are not in the announcement.
Two suppliers is the fact worth alarming about
Sony and Fujifilm are the only two manufacturers of LTO cartridges anywhere, which is a smaller number than most people assume when they read that tape is a mature commodity.
That is a tighter supply concentration than in semiconductors, underneath the medium most organisations use as their offline last resort against ransomware. Tape is chosen precisely because a cartridge on a shelf cannot be encrypted over a network, which makes it the copy that has to survive when everything else has failed.
The precedent for what a two-supplier market does under stress is not comforting. A single factory incident at either company removes half of world supply, and the historical example is close at hand: a 2011 fire at a Fujifilm facility disrupted tape supply for months. A backup strategy that rests on a medium with two producers has a supplier dependency that belongs in the risk register alongside the cloud provider and the ransomware scenario it exists to answer, and it is rarely recorded there.
The trade explanation fits, and it is not the only one
Coughlin's account is plausible, and it does not distinguish between two possibilities. Cautious purchasing because of trade uncertainty would produce the same dip and surge as a pause ahead of a media generation change.
A buyer who knows 40TB cartridges are coming and will work in existing drives has an obvious reason to defer buying 30TB media. That is not caution about tariffs; it is ordinary product-cycle behaviour, and it would produce a dip in 2025 and a surge in 2026.
Both explanations are consistent with everything published, and the 57 per cent first quarter is equally consistent with either. Which one dominates matters for anyone forecasting from these numbers, and the announcement does not contain enough to tell them apart.
The power argument is about to matter more
Coughlin's observation about power is the part with the most consequence for this region. A cartridge on a shelf draws nothing. A disk array holding the same data draws power continuously, whether or not anything reads it.
For a data centre where the binding constraint is the grid connection rather than floor space or capital, that difference converts directly into capacity available for compute, since power not spent spinning archival disk can be sold as something else.
We reported that operators are installing on-site gas turbines to bypass multi-year grid connection queues. Moving cold data to a medium that consumes nothing at rest is a much cheaper way to approach the same power constraint.
View author profile → · Editorial policy
