Daniel's been digging into magnetic tape this week. Not the cassette kind. The kind that's quietly holding up a huge chunk of the world's long-term data while nobody's looking. He wants to know about LTO, Linear Tape-Open, and why it's still relevant in the era of cloud computing. His prompt runs through what makes it remarkable, how it's still in active development unlike, say, Blu-rays, which are frozen in time. How the generations have been carefully controlled and backward compatibility assured, and why people respect the format. He also mentions the long-running speculation that Google Takeout takes forever because your data's being pulled from tape. So the question is whether that theory holds up. And underneath all of it, there's a bigger question about how tape has stayed relevant while the cloud ate everything else.
And the answer to the Takeout thing is no, by the way. Just going to spoil that early.
We'll get there. But first, the obvious question. What actually is LTO, and why should anyone outside a data center care?
LTO Ultrium is a magnetic tape format developed by three companies, HPE, IBM, and Quantum. It's an open format, which is the part people miss. Multiple competing vendors license it, and their products get submitted to compatibility testing so one vendor's tape works in another vendor's drive. That openness is why it became the de facto standard for what they call medium-tier storage. And it's now in its tenth generation with a twenty-five year history. The current cartridges, the LTO-10, hold forty terabytes native. That's a hundred and twenty-two percent capacity boost over LTO-9. And a fifteen hundred percent increase over LTO-6, which launched just over a decade ago. So this is not a frozen format. It's not a museum piece. It's still moving.
Fifteen hundred percent in a decade. That's the part that surprises people who think tape is dead.
Right. And here's the tension that runs through the whole story. Tape is losing the density race to disk and solid state. Disk drives have passed thirty-two terabytes. SSDs are at a hundred and twenty terabytes plus. But tape keeps winning the economics race and the resilience race. And its compatibility discipline, the thing that built its reputation, just hit its first break at generation ten. That's the story Daniel's pointing at, whether he knows it or not.
So let's dig into the compatibility thing, because that's the technical heart of this. How has LTO managed twenty-five years of disciplined generational compatibility?
The rule, straight from the LTO Program's own documentation, is that drive generations one through seven can read tapes from two generations prior and write to tapes from the prior generation. So an LTO-7 drive could read LTO-5 tapes and write to LTO-6 tapes. That's the pattern. LTO-8 drives read and write LTO-7 and LTO-8 media. There's even a thing called LTO-7 Type M, which is an LTO-7 cartridge that's been specially formatted to hold nine terabytes instead of six, and LTO-8 drives can read and write those too. LTO-9 drives read and write LTO-8 and LTO-9 only. So the window narrows over time, but it's always there. Two generations back for reading, one generation back for writing.
So if you're an enterprise with ten years of LTO-6 tapes sitting on a shelf, you can still read them with an LTO-8 drive. You don't need to keep a museum of old hardware running.
That's the discipline. And it's why enterprises trust the format with decades-long retention. You're not betting on a single vendor's roadmap. You're betting on a consortium that has committed, in writing, to a compatibility contract. And then LTO-10 broke it.
How badly?
Completely. LTO-10 drives can only read and write LTO-10 media. That's it. No LTO-9, no LTO-8. The chain is broken. And the reason is a redesigned tilted head in the drive mechanism. The head is now angled relative to the tape path in a way that eliminates the need to pre-initialize cartridges before first use. Which is a real operational win, by the way. One of the annoyances of previous generations was that you had to run a full initialization pass on a new cartridge before you could write to it. That's gone now. But the cost is that the drive can't physically read the old tapes anymore.
So it's not a licensing decision. It's a physics decision.
It's a physics decision with a licensing asterisk. The LTO Program's own language says backward compatibility "could be implemented in future LTO generations based on market demand." Which is a very careful way of saying, we could bring it back if enough customers scream, but we're not promising anything. That's a genuine event in this world. Twenty-five years of a compatibility chain, and then a break. Not a footnote.
And that's the difference between LTO and something like Blu-ray. Blu-ray is static. It's a finished format. There's no Blu-ray generation fourteen roadmap. There's no consortium actively developing the next capacity bump. It just is what it is.
Right. LTO has a published roadmap through generation fourteen. That's the structural difference Daniel's pointing at. Blu-ray is a legacy format that still gets used. LTO is a legacy format that's still being developed. Those are different categories. And the roadmap is public. You can go look at it. Generation eleven, twelve, thirteen, fourteen. Capacity targets. Data rates. All of it.
And they just cut those targets.
They did. In November of twenty twenty-five, the LTO Program reduced the future roadmap numbers. LTO-11 went from seventy-two terabytes down to seventy. LTO-12 from one forty-four down to one twenty. LTO-13 from two eighty-eight down to two ten. And LTO-14 from five seventy-six down to three sixty-five raw. That's a thirty-six percent cut at the top end. Blocks and Files, the storage trade publication, read that as tape technology limitations in action. The LTO Program framed it as a balanced combination of density, reliability, performance, and cost efficiency. Two framings of the same fact. And both are true, honestly.
So the density race is being lost, and they're managing expectations.
Yes. But here's the thing. The forty terabyte LTO-10 boost came from a new aramid-based tape substrate. Thinner and smoother than the previous PEN film. That lets you fit longer tape in the same standard cartridge. So there's still innovation happening at the materials level. It's not like they're just sitting there watching disk drives pass them. They're fighting for every terabyte. But the physics of magnetic tape is harder than the physics of flash. That's just the reality.
So that's the mechanism. But here's the question Daniel raised. If tape is losing the density race to disk and SSD, why does it keep winning? Why is it still the archival backbone?
Economics first. At multi-petabyte scale, tape's cost per terabyte is a fraction of flash or even hard disk. The practitioner numbers from people actually buying this stuff tell the story. Used LTO-5 drives sell for about a hundred and fifty dollars on eBay. SAS host bus adapters, thirty dollars. One commenter on Hacker News said he sustains a hundred and forty megabytes per second writes on an old i5 server. LTO-9 media runs about eighty dollars for eighteen terabytes. That's half the gross cost of spinning disk, his words. LTO-8 media is sixty-five to seventy euros for twelve terabytes. A brand new mLogic LTO-8 drive is about thirty-three hundred dollars. So the entry point is real but not insane, and the media cost per terabyte is unbeatable at scale.
And then there's the resilience side.
Right. Tape has structural advantages that disk and SSD fundamentally cannot match. Hardware encryption on the cartridge itself. WORM capability, write once read many, which gives you immutability. You literally cannot overwrite a WORM tape. That's your ransomware insurance. And true offline air-gapping. The tapes can physically leave the library. They can sit on a shelf in a vault. There's no network path to them. No attacker can reach them. No production failure can corrupt them. That's the thing cloud storage can't give you. Your cloud data is always online, always reachable, always one compromised credential away from deletion.
So the cloud didn't kill tape. The cloud made tape more necessary.
And now there's the AI angle, which is the newest twist. Skip Levens at Quantum, who works on the LTO Program, wrote a piece in TechRadar just this week arguing that a dataset does not need the same performance at every stage of the AI pipeline. The training data doesn't need to be on flash. It needs to be affordable and durable and accessible when you need it. Tape is the affordable capacity tier for AI datasets. And HPE's Stephen Bacon said in November that AI has turned archives into strategic assets. That's the reframing. Archives used to be cold storage. The place data went to die. Now they're the fuel depot for model training.
And the fuel depot is growing fast.
The Stanford AI Index from twenty twenty-five found that LLM training dataset sizes are doubling every eight months. Doubling. Every eight months. That's a compounding retention problem. You can't delete the old training data. You might need it for retraining, for evaluation, for compliance, for legal discovery. So you keep it. And the cheapest place to keep it is tape. That's why the LTO Program is marketing directly at the AI crowd now. They see the opening.
So the economic case is clear. But let's talk about the operational reality, because tape is not just slow disk. There's a thing called shoe-shining.
Shoe-shining. This is the part that separates people who've actually run tape from people who've read about it. LTO is linear. The tape moves at over a hundred inches per second. If the write buffer empties, if your data source can't keep up, the drive has to decelerate, rewind, re-accelerate, and get back up to speed. That's called shoe-shining because the tape is literally polishing the head while it does that. It's terrible for the media and terrible for throughput. So you need fast staging storage feeding the drive. You can't just point a slow network share at a tape drive and expect it to work. The tape will outrun you.
So it's a fundamentally different operational model. Not a slower version of disk. A different thing entirely.
Completely different. And that's why the abstraction layer matters. LTFS, the Linear Tape File System. It's an open source, open standard layer that makes tape behave like a drag and drop drive. You mount a tape and it looks like a filesystem. You copy files to it. You copy files off it. No proprietary backup software required. That's a huge part of why tape is usable outside mainframe-style workflows. And it's open, so it works across vendors.
The open source ecosystem has caught up too.
Proxmox Backup Server ships open source tape support now. Single drives and robots. With deduplication, compression, and encryption built in. So the home lab person can run a legitimate tape backup setup without paying enterprise software licensing. That's the thing that changed. Tape used to require a six-figure library and a dedicated backup administrator. Now a hobbyist can do it with a used drive and an open source tool.
Which brings us to the Google Takeout question. The folklore says Takeout is slow because Google has to pull your data from tape. Is that true?
No evidence. And I looked. There's no documentation from Google that says Takeout is tape-backed. No confirmation anywhere. Google's own help page says archives may be delayed for account safety reasons. Advanced Protection Program users get their archives scheduled two days out. That's a deliberate security delay, not a tape mount time. The Hacker News commenters who've dug into this attribute the slowness to server-side job compilation, data held across several servers, and migration time. Your Takeout archive is a bunch of zip files assembled from multiple Google services. That's a job that takes time to run. It's not sitting on a tape waiting for a robot arm.
The tape theory is folklore.
It's folklore. And it's the kind of folklore that sounds plausible because people know Google uses tape somewhere. Which they do. Google has been public about using tape for cold storage in their data centers. But Takeout specifically? No. The mundane explanation is just job scheduling and account safety. The archive sizes people report, fifty to five hundred gigabytes typically, one user had sixty-six zips totaling a hundred thirty-one gigs, another had a hundred sixty-seven gigs across seven archives. That's not a tape retrieval problem. That's a zip file assembly problem.
The thing people respect about LTO isn't the Google connection. It's the discipline.
It's the discipline. And the economics. And the air gap. And the fact that it's still being developed. People on Hacker News who actually use this stuff talk about it with genuine affection. One commenter said LTO tapes changed his life, or at least his mental health. The peace of mind of having an offline copy that can't be ransomware'd. That's the thing. It's not the sexiest technology. It's the one that doesn't fail you when everything else does.
Where does this leave the format? The compatibility break, the roadmap cuts, the density race being lost. Is LTO in trouble?
I don't think so. The economics are still too good. The air gap is still too valuable. The AI retention problem is still growing. But the compatibility break is a signal. It says the old assumptions are being renegotiated. The LTO Program is saying, we might not be able to promise two generations of backward compatibility forever. We might need to make tradeoffs. And the roadmap cuts say the same thing from the density side. Tape is not going to outrun flash. It's going to have to win on cost and resilience, and it's going to have to be honest about what it can't do.
The AI reframing is the bet they're making. That archives are strategic assets now, not cold storage.
That's the bet. Whether it holds as disk and SSD keep outpacing tape on capacity, that's the thing to watch. If someone figures out how to make a hundred terabyte hard drive that costs fifty dollars, tape's economic case gets thinner. But that hasn't happened yet. And the air gap advantage doesn't go away no matter how cheap disk gets. You can't air gap a disk array the way you can air gap a tape on a shelf.
There's something else I want to pull on. You mentioned the tilted head design that broke compatibility. That's a real engineering tradeoff. They traded backward compatibility for faster initialization. Was that the right call?
I'm not sure. The pre-initialization step was annoying. You'd buy a new LTO-9 cartridge and have to run a full pass before you could write to it. That's hours of drive time wasted. The tilted head eliminates that. So operationally, it's a win. But the cost is that every LTO-9 tape in the world is now unreadable by the current generation of drives. If you're an enterprise with a vault full of LTO-9 tapes, you now have a hard dependency on keeping LTO-9 drives alive. That's a real risk. Drives fail. Vendors stop manufacturing them. The compatibility chain was your insurance policy, and now it's gone.
The break isn't just a technical footnote. It's a risk management problem for anyone who trusted the old contract.
The LTO Program knows this. That's why their language is so careful. "Could be implemented in future LTO generations based on market demand." They're leaving the door open to restore compatibility if customers demand it. But they're not committing. And that ambiguity is itself a cost. You can't plan a twenty-year retention strategy around a maybe.
Which is the kind of thing that makes a storage administrator nervous.
It should. The whole value proposition of LTO was that you could trust the roadmap. You could buy LTO-8 drives knowing that LTO-10 drives would still read your tapes. That trust is now conditional. And conditions are not what you want in a format that's supposed to outlast your career.
The respect people have for LTO is earned, but it's also being tested.
It's being tested by physics. The density race is real. The materials science is getting harder. The aramid substrate got them to forty terabytes, but the next jumps are smaller than they promised. That's not a failure of will. It's just the reality of trying to pack more magnetic domains onto a ribbon that's already incredibly thin. There are limits.
Meanwhile the cloud is still there, still convenient, still online.
The cloud is the right answer for a lot of things. But the cloud is not an archive. It's a service. Services change pricing, change terms, get acquired, shut down. A tape on a shelf is a physical object. It doesn't care about your subscription status. It doesn't require authentication. It's just there. That's the thing the cloud can't replicate. Physicality.
Daniel's question about why tape has remained relevant despite cloud computing has a pretty clear answer. The cloud didn't replace the archive. It replaced the file server. The archive is a different job.
The archive job is growing. AI training data, compliance retention, medical records, scientific datasets. The stuff that has to exist in twenty years. That's tape's territory. Always has been. The AI boom just made it more obvious.
Before we wrap, I want to make sure we've addressed everything Daniel asked. The format itself, the active development, the controlled generations, the backward compatibility, the respect it earns, and the Google Takeout theory. I think we've hit all of it.
We haven't talked about the sound.
The sound?
Hilbert: You're both wrong about one thing. Missing. Everyone talks about capacity and cost. Nobody talks about the sound.
Hilbert: I spent eighteen months in the late nineties working night shifts at a regional insurance company's data center in Ohio. My entire job was swapping LTO-1 cartridges in and out of a robotic library and logging the barcodes by hand. The library at three in the morning had a sound. The whine of the picker arm when it moved. The specific click when a cartridge seated. You could hear it across the room. I can still hear it.
Hilbert: The company migrated off tape in two thousand four. They lost a chunk of claims data in the migration. I think about that sometimes. They spent all that money to move to disk, and the disk array ate their history. The tapes would have been fine.
Hilbert: I still have one of the cartridges. LTO-1. It's on my coffee table. I use it as a coaster.
You have a tape cartridge with insurance claims data on it, sitting on your coffee table, under a drink.
Hilbert: It's a good coaster. Flat. Heavy enough not to slide. And I'm pretty sure there's still claims data on it. I never erased it. Nobody asked me to.
Somewhere in Hilbert's apartment is a twenty-five year old magnetic tape containing the medical and financial records of people who filed insurance claims in Ohio in the late nineties. And it's currently protecting a coffee table from water rings.
Hilbert: It's the most reliable thing I've ever owned. Never failed once.
I don't think we need to resolve whether that's a problem.
Hilbert: It's not a problem. It's a coaster. The data's fine. Tape doesn't degrade if you keep it dry. That's the whole point.
He's right about the sound, though. The people who run tape libraries talk about the sound. It's a sensory thing. The whine and the click. It's like the people who miss the sound of a modem handshake. The machine was doing something physical, and you could hear it working.
That physicality is the thing we keep coming back to. A tape is a thing. It has weight. It has a sound. It sits on a shelf. The cloud is an abstraction. Useful, but an abstraction. Hilbert's coaster is the most concrete possible reminder that tape is real in a way that cloud storage never will be.
Hilbert: My brother-in-law sold tape drives for a while. He said the same thing. People would come in asking about cloud backup and he'd hand them a cartridge and say, this is what your data looks like. Most of them bought the tape drive. He's not to be trusted, but he had a point.
The one thing I'm taking from this is that the compatibility break at LTO-10 is the moment the format stopped being able to promise what it used to promise. Twenty-five years of disciplined backward compatibility, and then physics forced a choice. The question is whether that's a one-off or the start of a pattern. The LTO Program's own language leaves it open.
The thing that changed for me is the AI reframing. Archives went from cold storage to strategic assets. That's not just marketing. If training datasets are doubling every eight months, the cheapest durable storage wins. Tape wins that race today. Whether it keeps winning as disk and SSD keep outpacing it on capacity, that's the open question.
Which is a good place to leave it. Thank you to our producer, Hilbert Flumingtop, for keeping the show running and for the coaster.
This has been My Weird Prompts, the human AI collaboration podcast.
If you enjoyed this episode, leave us a review wherever you listen. It helps other people find the show.
We'll be back soon.
