Daniel's been staring at the hard drive shelf in the computer store again. His question this time: we did a whole episode on LTO tape and how it refuses to die, so what about the spinning disk? His assumption was that the hard drive is the one that's actually struggling. It's louder than solid state, it's not as physically dense as tape, and it has moving parts that fail. He figured the only place left for it might be warm storage, big NAS boxes, and that would require some kind of step up from the consumer drives we used to bolt into desktops. So he's asking two things. What does hard drive adoption actually look like right now, and is the longevity coming from the enterprise side rather than the consumer side?
The framing is exactly backwards.
I had a feeling you'd say that.
The hard drive is not struggling. The hard drive is in the middle of the biggest boom it's had in a decade. Seagate just reported a record gross margin of fifty-two percent. Western Digital's fiscal year revenue was up thirty-six percent. Their entire production is sold out through twenty twenty-eight. Dave Mosley, Seagate's CEO, stood in front of investors two months ago and said, and I quote, disk drives ain't going away.
That's the kind of quote that ages well if you're right and terribly if you're not.
He's right. Here's the number that reframes the whole thing. In their last fiscal quarter, Seagate shipped two hundred eighteen exabytes of capacity. That's up thirty-four percent from the year before. And of that, one hundred ninety-five exabytes went to hyperscalers and large enterprises. That's eighty-nine percent of everything they shipped, going to the data center. The consumer desktop drive is not the story anymore. It hasn't been for years.
So Daniel's instinct was right about the enterprise side, he just underestimated how lopsided it's become.
Lopsided is the word. Laptop hard drives are nearly extinct. Desktop drives are a declining category. But the nearline enterprise drive, the big three and a half inch drive that sits in a rack and spins constantly, that's over sixty percent of all drives shipped now. And the unit count isn't even falling. Nidec, the Japanese company that makes about eighty percent of the world's spindle motors, counted one hundred nineteen million drives in twenty twenty-three and one hundred twenty-five million in twenty twenty-four. Unit sales went up.
Wait. The thing everyone assumes is being replaced by flash is shipping more units than the year before.
Because the demand curve moved. AI training and inference generate absurd amounts of data that has to live somewhere cheap. A thirty terabyte enterprise SSD costs twenty-two thousand six hundred dollars right now. Per terabyte, enterprise flash is eighteen point six times more expensive than hard drives. If you're a hyperscaler storing an exabyte of model checkpoints, training logs, or the raw corpus you scraped off the internet, you cannot put that on flash. The economics don't exist.
So the cloud is the thing keeping the spinning disk alive. That's a nice inversion of the story everyone told for fifteen years.
The cloud replaced the file server, not the archive. And now the cloud's own appetite is so big that it's consuming the entire hard drive supply chain. Seagate said their nearline capacity is allocated into calendar twenty twenty-eight. Customers are asking for planning horizons through twenty twenty-nine and beyond. This isn't a market that's hanging on. It's a market that's supply constrained.
Supply constrained because they're not building more factories.
Right. And that's a deliberate choice. Mosley said it plainly. Increasing the amount of data stored on every disk is the fastest, most capital efficient path. So instead of adding manufacturing lines, they're pushing density. The roadmap is the story. Seagate's HAMR drives, heat assisted magnetic recording, are shipping in volume at forty-four terabytes now. Fifty terabytes starts qualification late next year, volume in twenty twenty-eight. They're talking about eighty plus terabytes in twenty thirty-one and a hundred terabytes in the early twenty thirties.
A hundred terabytes on a single spinning disk. The drive I put in my first computer was forty gigabytes. That's twenty-five hundred times more.
And the technology to get there is wild. HAMR heats a tiny spot on the platter with a laser to make the magnetic material temporarily easier to write, then lets it cool so the bit stays put. The material they're moving toward is ordered granular iron platinum, which is a phrase that sounds made up but it's real. Western Digital is taking a more conservative path. They shipped their forty terabyte drive this month using something called UltraSMR with eleven aluminum platters, and their forty-four terabyte HAMR drive is coming in the first half of next year. Toshiba is the cautious one of the three, they're at twenty-eight terabytes now and will get to around forty by twenty twenty-seven.
Three manufacturers. That's the whole industry.
Three. And really it's a duopoly between Seagate and Western Digital with Toshiba trailing. Which is why the pricing conversation is what it is. Seagate's CFO said there's opportunity to keep pushing on pricing. Western Digital said the same thing in their own words. When you have three companies controlling the entire global supply of a component that the AI buildout cannot do without, you get what we're seeing now.
Which is what, exactly, for the person walking into a store?
Retail hard drive prices are up forty to ninety percent since March. German retailer tracking showed prices up one hundred twenty-nine percent year over year. NAS drives specifically went up forty-six percent between September and January. External drives have more than doubled in under a year. The era of storage being the cheap part of the project is over for a while.
And that's the thing Daniel was actually wondering about without knowing it. He said you can still find hard drives in any mainstream computer store. He's right, but the reason is not that the consumer market is healthy. It's that the enterprise market is so hot that the retail channel is getting the leftovers.
The factories are running flat out for the hyperscalers. The retail drive you buy for your NAS is the same physical product line as the nearline drive going into the data center, and the data center is willing to pay more and sign multi-year contracts. The consumer gets whatever capacity is left over at whatever price the shortage sets.
So the homelab person building a big NAS is now competing with Meta for the same drive.
And losing. Seagate's nearline exabytes are committed through twenty twenty-eight. If you want a twenty terabyte drive for your Plex server, you're bidding against the people who bought up the entire production run.
Daniel's guess about warm storage is interesting though. He said hard drives would need a step up in technology for that role. The answer is that the step up already happened, and it happened in the enterprise.
It did. And there's a specific product direction that maps almost exactly to what he described. Western Digital is building what they call power optimized drives for active cold storage. The idea is data that has to remain accessible but doesn't need high performance. AI logs, checkpoints, the stuff that's written once and read occasionally. These drives cut power consumption by about twenty percent and they're positioned against QLC SSDs starting next year.
Active cold storage. That's a nicer phrase than warm storage.
It's a real tier now. The hyperscalers have always tiered their storage. Memory at the top, then performance flash, then capacity flash, then hard drives at the bottom. What's changing is that the bottom tier is becoming more differentiated. You have standard nearline drives for bulk capacity, and now you have this emerging category of drives that are tuned specifically for the cold but not frozen tier. Lower power, still spinning, still accessible.
So the hard drive isn't just surviving as a legacy format. It's getting new product categories designed for it.
And the performance side is moving too, which nobody expected. Seagate has the Mach two dual actuator drives, which are basically two independent read-write heads in one drive so you get double the throughput. Western Digital is talking about an eight times bandwidth roadmap with dual pivot drives in the lab. The hard drive is not sitting still. It's being actively engineered.
The thing I keep thinking about is the assumption baked into Daniel's question. He said there's only so much room for older formats. Tape does long-term retention, SSDs do fast retrieval, so the hard drive is squeezed in the middle. But the data center doesn't see it that way. The data center sees a hierarchy where every tier has a job and the hard drive's job is the biggest one.
The hard drive's job is to be the floor. Everything that doesn't need to be fast and doesn't need to be archived for fifty years lands on spinning disk. And that category is growing faster than any other because AI generates so much of it. When you train a model, you checkpoint it constantly. Those checkpoints are enormous. You don't need to read them at flash speed, but you do need to read them eventually. That's a hard drive workload.
And the model itself, the training data, the logs, the inference outputs. All of it has to live somewhere.
Gartner is forecasting three point one nine zettabytes of data center storage capacity needed by twenty twenty-nine. That's a nineteen and a half percent compound annual growth rate. The total hard drive capacity shipped is projected to go from one point two zettabytes in twenty twenty-four to over seven zettabytes in twenty twenty-nine. The hard drive industry has to nearly sextuple its output in five years.
Sextuple. With three manufacturers and no new factories.
Which is why the density roadmap is the entire game. They can't make more drives, so they have to make each drive bigger. That's why HAMR is existential for Seagate. It's not a nice-to-have. It's the only way they can meet the demand curve.
And the supply chain underneath is terrifyingly concentrated. One company makes eighty percent of the spindle motors. One company makes the glass substrates. About ninety percent of the rare earth refining for the magnets is in China.
The rare earth piece is a genuine vulnerability. China suspended rare earth export controls into late this year, and the hard drive industry is completely dependent on those materials for the voice coil motors and the spindle motors. If that tap turns off, the entire hard drive industry stops. No amount of HAMR engineering helps if you can't make the motor spin.
That's the part that keeps me up at night, metaphorically. The cloud's entire mass storage tier depends on a supply chain that runs through a handful of factories and one country's rare earth policy.
And the alternative is flash, which is also supply constrained. Phison's CEO warned of a NAND shortage through twenty thirty. So the escape hatch is closed too. If hard drives get more expensive or scarce, you can't just switch to SSDs. SSDs are also expensive and scarce.
There's no cheap tier left.
There's no cheap tier left. That's the real story of twenty twenty-six storage. The AI buildout consumed the cheap tier.
Let me go back to something you said earlier. Seagate's gross margin went from thirty-seven percent to fifty-two percent in a year. That's not a company fighting for survival. That's a company with pricing power.
Record margins, revenue up forty-seven percent year over year, net income up two hundred eighty percent. And Western Digital just completed its fiscal year as a pure play hard drive company after spinning off its flash business into Sandisk. Their revenue was twelve point nine billion dollars, up thirty-six percent. The hard drive is not the legacy format. The hard drive is the profit center.
Pure play hard drive company. A year ago that would have sounded like a death sentence. Now it sounds like a smart trade.
The flash business went to Sandisk and the hard drive business stayed with Western Digital, and the hard drive business is the one with the forty-four percent year over year growth. The market has completely repriced what spinning disk is worth.
What about the consumer side though? Daniel's question was partly about the drives in the computer store. If the enterprise is consuming everything, what's actually on the shelf?
The same drives, just more expensive and harder to find. The desktop drive category is declining but it still exists. The laptop drive is essentially gone. What's growing on the consumer side is NAS and external backup. And that's the part of the consumer market that's getting squeezed the hardest because it wants the same high capacity drives the hyperscalers want.
A twenty terabyte NAS drive and a twenty terabyte nearline enterprise drive. Are they actually the same product?
Not exactly, but they share the same platform. The NAS drives have firmware tuned for vibration tolerance in multi-drive enclosures, and they're rated for always-on operation. The enterprise nearline drives have similar characteristics plus some data center features. But the underlying platters, heads, motors, all of it comes from the same factories. When the factory is sold out to the hyperscaler, the NAS drive gets scarce too.
So the homelab enthusiast is feeling the squeeze directly.
And the squeeze is not ending soon. Analysts don't expect normalization before late twenty twenty-seven. The supply is committed on long-term agreements. If you want a high capacity drive for your NAS, you're paying a premium and you're waiting.
Daniel said the hard drive sounds like something from a past era of computing. The clicking, the spinning, the mechanical failure risk. But the enterprise didn't abandon it for those reasons. The enterprise never cared about the noise.
The enterprise cares about dollars per terabyte per year. And on that metric, the hard drive still wins by a factor of eighteen. The failure risk is managed with redundancy. A hyperscaler doesn't care if a drive fails. They have erasure coding across the entire fleet. A dead drive is a maintenance event, not a data loss event.
Which is the thing consumer users still don't fully internalize. RAID is not a backup. One drive in a NAS is not a backup. The enterprise treats every drive as disposable because the system is the durable thing.
And that's the philosophy that lets the hard drive keep winning. The individual drive is unreliable, sure. But the aggregate fleet is the most cost effective storage ever built.
What about the reliability data? Backblaze publishes those drive stats every year. What's the current picture?
The annual report I'd want to cite is behind a paywall I couldn't get through this week, but the broad pattern has held for years. HGST drives, which are now Western Digital, tend to be the most reliable. Toshiba tends to lag. And the failure rates overall are low single digits per year for the big enterprise drives. The horror stories about drives dying constantly are mostly from the consumer desktop era, not the nearline era.
The nearline drives are built differently.
Different firmware, different vibration handling, different error recovery. A desktop drive would sit there retrying a bad sector for thirty seconds while your game freezes. A nearline drive is configured to fail fast and let the storage system handle it. Time limited error recovery. It's a completely different philosophy.
So the same physical object is a different product depending on what firmware it's running.
The platters don't know what market they're in. The firmware decides whether the drive is a consumer product that tries really hard to read a bad sector, or an enterprise product that gives up quickly and lets the RAID controller deal with it.
And the firmware is why you can't just buy a cheap drive and shuck it from an external enclosure for your NAS anymore. Or you can, but you're getting consumer firmware.
The shucking era was fun while it lasted. People were buying external drives on sale, pulling the bare drive out, and putting it in their NAS. The drive inside was often a NAS-grade or enterprise-grade drive at a consumer price. The manufacturers caught on and started putting cheaper drives in the externals. And now with the shortage, the externals are expensive too, so the whole arbitrage is gone.
The arbitrage is gone. That's a nice summary of the whole consumer moment.
The consumer market for hard drives in twenty twenty-six is a market of people who need spinning disk specifically. Homelabbers, photographers, video editors, people with big media collections. The people who just want storage for their computer buy an SSD. The hard drive has become a specialist product even on the consumer side.
Which loops back to Daniel's warm storage idea. He was imagining a step up in hard drive technology for the NAS use case. The step up already happened, it's just being consumed by the data center first.
And the consumer version of that step up is the drive you can't get because the hyperscaler bought it all.
So where does this leave the format in five years? If the roadmap hits a hundred terabytes, what does the hard drive look like in twenty thirty-one?
It looks like the foundation of the entire AI infrastructure. The hyperscalers will have exabyte-scale hard drive farms that are the cheapest possible storage per terabyte. The consumer market will be a niche of people who need bulk local storage and are willing to pay for it. And the technology will be at a level that would have seemed impossible ten years ago.
A hundred terabytes on a single drive. You could put the entire Library of Congress on a handful of drives.
The Library of Congress is about twenty terabytes of text. So five drives, with room to spare.
I want to go back to the pricing story for a second, because there's a tension there. Three manufacturers, sold out capacity, record margins. That's the setup for a price fixing conversation.
There's already a class action lawsuit against the hard drive suspension assembly makers. The component suppliers, not the drive manufacturers themselves. But the structural conditions are there. Three companies, no new capacity, everyone pushing on pricing. Whether that crosses into illegal coordination is a different question, but the outcome for buyers is the same.
The buyer has no leverage.
The buyer has no leverage. The hyperscalers have some leverage because they sign multi-year contracts and they're the primary customers. But even they're paying more. The real pain is downstream. The homelabber, the small business, the photographer who needs a twenty terabyte external drive. Those buyers are paying whatever the market sets.
And the market is setting prices that doubled in a year.
More than doubled for externals. The FPS Review tracked external drive prices doubling in under a year. That's not a shortage blip. That's a structural shift in what storage costs.
The era of storage being the cheap part of the project is over for a while. That's the line that sticks with me.
It's the right line. Storage used to be the thing you didn't budget for because it was always cheap. Now it's a line item.
What about the argument that this is all temporary? The AI buildout is a bubble, the demand will crash, and hard drives will go back to being cheap.
Even if the AI training demand cools off, the data doesn't go away. Every model checkpoint, every training corpus, every log file from the last three years is still sitting on disk. The data center storage capacity requirement is forecast to grow at nineteen and a half percent annually through twenty twenty-nine. That's not a bubble number. That's the floor.
The data is the durable thing. The models might change, the architectures might change, but the training data has to live somewhere.
And the cheapest somewhere is a hard drive. That's the whole argument. It's not that hard drives are the best at anything. It's that they're the cheapest at the thing that's growing the fastest.
So Daniel's question, is the longevity coming from the enterprise side rather than the consumer side, the answer is yes, emphatically, and the consumer side is actually being harmed by the enterprise side's success.
That's the shape of it. The hard drive is not dying. It's being consumed. And the consumer is feeling the squeeze of being the secondary customer.
The secondary customer. That's the phrase for the person in the computer store looking at the hard drive shelf.
The drives are there. They're just more expensive, harder to find, and they're the same drives the data center wanted first.
Let me ask you the technical question Daniel was circling. The warm storage use case. What does a purpose built warm storage hard drive actually look like?
The Western Digital power optimized drives are the closest thing. Twenty percent less power, tuned for the active cold tier, positioned against QLC SSDs starting next year. The idea is that AI logs and checkpoints don't need to be on flash, but they do need to be accessible faster than tape. So you build a hard drive that's optimized for that specific middle ground. Lower power means lower operating cost, which matters when you have a million drives spinning.
A million drives. The scale is the thing that's hard to hold in your head.
Seagate shipped two hundred eighteen exabytes in one quarter. That's two hundred eighteen million terabytes. If you stacked those drives, you'd have a tower that reaches into space. The scale is hard to comprehend.
And that's one quarter. From one company.
One company. Western Digital and Toshiba are also shipping. The total industry output is staggering.
So the hard drive is not the legacy format. The hard drive is the growth format. That's the headline.
The hard drive is the growth format, and the consumer is the afterthought. That's the full headline.
What about the failure modes? Daniel mentioned mechanical failure as a reason the hard drive feels outdated. Is that still a real concern at the enterprise level?
It's managed. A hyperscaler with a million drives expects a certain number to fail every day. The system is designed around that. The individual drive's reliability matters less than the fleet's statistical behavior. And the fleet behavior is actually pretty good. The big enterprise drives fail at low single digit rates per year.
Low single digits. So a million drive fleet loses maybe thirty thousand drives a year. That's eighty-two drives a day.
They have hot spares and erasure coding and automated replacement. A drive dies, the system rebuilds, a technician swaps it, the dead drive goes to a shredder. It's a logistics operation, not a data loss event.
The shredder is the part that always gets me. These drives are destroyed after a few years because the data on them is too sensitive to let them leave the building.
That's another reason the hard drive keeps selling. The drives wear out or get retired, and the hyperscaler buys new ones. The replacement cycle is built into the demand.
The hard drive is not just storing the data, it's being consumed by the data's lifecycle.
The drive is a consumable. That's the framing the enterprise uses. The drive is a consumable, like a filter or a battery. You buy it, you use it, you replace it.
Which is wild when you think about the consumer framing. The consumer thinks of a hard drive as a durable good. You buy it, you keep it for years, you worry about it failing.
The enterprise thinks of it as a consumable. That's the difference in perspective that explains everything about why the hard drive is thriving in the data center while the consumer market shrinks.
The consumer is worried about the drive failing. The enterprise has already budgeted for the failure.
That's the mental model shift. If you think of the drive as a consumable, the mechanical failure risk stops being scary. It's just the cost of doing business.
Where does this leave the person who just wants to back up their photos?
Paying more and hoping the drive lasts. The consumer is stuck with the consumer mental model and the enterprise price environment. That's the squeeze.
The squeeze. That's the whole episode in one word.
Hilbert: The squeeze is the wrong word. The right word is the floor.
The floor.
Hilbert: The hard drive is the floor. Everything in that data center sits on it. The flash, the memory, the tape, all of it is sitting on a lake of spinning disks. And the lake is getting deeper every day. I worked at a colocation facility in the early two thousands. We had a customer who ran a backup service. Rows and rows of shelves, every shelf full of drives. The drives were the cheap part. The shelves cost more than the drives. Now the drives cost more than the shelves.
The drives cost more than the shelves. That's the whole pricing story in one sentence.
Hilbert: The customer was a man named Gerald. He'd come in every Tuesday and swap out the dead drives. He had a cardboard box full of them. He'd take them home and hit them with a hammer. He didn't trust the shredder company. Said they might resell them. So he hit them with a hammer in his garage.
That's a very specific form of data security.
Hilbert: It worked. Gerald never had a breach. But the point is, Gerald's drives were the cheap part. He didn't care if one died. He had a box of spares and a hammer. Now Gerald would be paying forty percent more for the same drive, and the hyperscalers would have bought out the whole production run before Gerald even got to the store.
Gerald's use case is the one that's getting squeezed hardest. The small operator who needs bulk local storage but doesn't have hyperscaler purchasing power.
Hilbert: Gerald's use case is gone. He retired in twenty nineteen anyway. But the person doing what Gerald did now is paying double and waiting months. The floor got expensive.
The floor got expensive. That's the consequence nobody was expecting. The cheapest tier stopped being cheap.
Hilbert: The cheapest tier stopped being cheap because the richest customers started buying it all. That's the whole story. The hard drive was the one thing in the data center that was still cheap, and then the AI people showed up and bought it all. Now there's no cheap thing left. The floor is a luxury item.
A luxury item with a hammer in the garage.
Hilbert: Gerald would have appreciated that.
The thing I'm left with is the inversion. Everyone assumed the hard drive was the past and the SSD was the future. But the hard drive is the future's foundation. The AI revolution runs on spinning disks.
The spinning disks are running out. Not the technology, the supply. The factories can't make them fast enough, and nobody's building new factories because the density roadmap is more profitable. So the hard drive is simultaneously the most important storage technology and the most constrained one.
The one thing I'd take from this is that the hard drive didn't survive by getting better at being a hard drive. It survived by being the cheapest thing in a world that suddenly needed to store everything. The enterprise didn't keep buying hard drives out of nostalgia. It kept buying them because there was no alternative.
The consumer who walks into a store and sees a hard drive on the shelf is looking at the leftovers of a market that's been consumed by something much bigger. The drive is there, but it's not really for you anymore.
That's the whole episode. The hard drive is alive, it's thriving, and it's not yours.
We'll be thinking about Gerald and his hammer. Thanks to Hilbert Flumingtop for producing. This has been My Weird Prompts. Email us at show at my weird prompts dot com. We'll be back soon.
