Seagate Technology reported the highest gross margin in its history on Tuesday. By the end of the call, the shares were down about 8 per cent in after-hours trading.
Revenue for the quarter ended October 2 reached $4.23 billion, 61 per cent above a year earlier and 17 per cent higher than the previous quarter, above the top of the company’s own guidance. Non-GAAP gross margin was 57.8 per cent, more than 18 points above the year-earlier level. Non-GAAP earnings of $7.68 a share were a record and nearly triple the prior figure. Net income came in at $1.76 billion, against $549 million.
The single number that explains the rest is gross profit per terabyte, which rose to $11.03 from $5.70. Seagate sells capacity, and for thirty years the price of a terabyte fell every year as the industry packed more bits onto the same platter. Getting paid more per terabyte is not a normal condition for a disk-drive maker. It is what a shortage looks like in a business where two companies make almost all of the nearline drives that go into data-centre racks.
The mechanism is straightforward once the demand is understood. Training a large model requires holding a corpus that is read repeatedly, and running one requires writing checkpoints that capture the state of the model at a given moment so a job can restart after a failure. Both are bulk reads and sequential writes, and both are far cheaper on a spinning disk than on flash. The hyperscalers buy nearline drives by the tens of thousands and place the orders months ahead.
Supply did not respond the way it once would have. After the last collapse in demand, the industry spent two years shutting lines, writing down inventory and promising investors it would not build ahead of orders again. That promise is being kept, which means the only way to add capacity is to raise price, and the buyers have no substitute at the same cost per byte.
Which is why the shares fell on a record. A gross margin of 57.8 per cent is not a level anyone expects to persist, and the question investors put to management was how much of the improvement is structural and how much is the top of a cycle. Per-terabyte pricing that has nearly doubled in a year invites the buyers to do arithmetic of their own.
The buyers have options, though none of them is quick. They can push archive data onto tape, where the cost per terabyte is a fraction of disk and the latency does not matter for a file nobody opens. They can move warm data to dense flash, whose price has also been rising as memory makers divert wafers to AI accelerators. Or they can renegotiate, using the threat of qualification with the other supplier, which every large buyer already runs as a matter of policy.
There is a physical limit that supports the company’s position. A nearline drive holds more data than it did five years ago because of a change in how the heads write to the media, and the transition to that technology took years and a great deal of capital to get right. Seagate and its competitor spent that money during a period when the business was losing money. The payoff arrives now, and it arrives on a product that takes roughly a year to design, qualify and build in volume.
What the market is pricing is the possibility that this is the peak. Storage has produced false dawns before: in 2010, in 2018, in 2021, each time a shortage that looked structural ended within four quarters as buyers worked through inventory they had double-ordered. The difference this time is that the buyers are not distributors holding stock for resale. They are operators building facilities that need to be filled, and their forecasts are tied to capital spending plans already announced.
The caution is that the same customers write the contracts. Annual agreements with volume commitments give the supplier visibility for twelve months and give the buyer a lever at renewal. A buyer who has watched per-terabyte pricing double will spend the year qualifying a second source and testing how much of its archive can tolerate tape latency. That work does not show up in a quarterly report. It shows up two years later as a flat price.
Seagate’s results also carry information for the rest of the hardware chain. Drive prices feed directly into the cost of a storage server, and storage is the part of a data-centre bill that grows with the data rather than with the compute. Every percentage point of per-terabyte increase lands on the operating budget of the customer, and the customers have been told for a decade that storage gets cheaper every year.
For now, the company is selling everything it can make at a price it could not have charged two years ago, and the argument among investors is about how long that sentence stays true. The results answer the question about the past quarter. They do not answer the one about the next eight.
