I know. In 2026, that’s practically heresy. We’re surrounded by SSDs. NVMe drives are ridiculously fast, they’re tiny, they’re silent, and they’ve transformed what a computer feels like compared with the mechanical storage we grew up with. And I love them. My operating system is on an SSD. My games are on SSDs. Applications, scratch storage and anything that benefits from low latency belong on flash. But I’ve noticed something over the last few years.
We’ve stopped talking about what storage is good at and started talking about which storage technology is simply better. That’s the wrong question. Because SSDs didn’t make hard drives obsolete. They made hard drives specialized.
SSDs Have a Massive Advantage
There is no argument here. A hard drive has a mechanical actuator moving a head over a spinning platter. Every time it needs to access a different location, there is physical movement involved. An SSD doesn’t have to do that. That’s why random access is where the difference becomes enormous. An operating system isn’t loading one giant sequential file every time you click something. Applications constantly access relatively small pieces of data from different locations. Games do the same thing while loading assets and managing their enormous collections of files. This is exactly where SSDs shine: low latency, high IOPS and excellent random-access performance.
Then there is sequential throughput. A modern NVMe SSD can push several gigabytes per second. A mechanical hard drive might manage a few hundred megabytes per second under good conditions. If you’re loading a game, launching Windows, compiling code, editing media or working with an application that constantly hits the storage subsystem, the SSD wins. And it isn’t close.
But here’s the thing: not every storage workload looks like that.
Your Gaming PC Might Be SSD Crazy. The Internet Isn’t.
Walk into almost any PC hardware discussion and you’ll find someone declaring hard drives obsolete. Meanwhile, some of the largest storage systems on the planet are still deliberately built using them. That’s because once you get beyond the desktop, storage becomes an infrastructure problem.
Imagine you’re responsible for storing petabytes—or exabytes—of data. You don’t just care about latency. You care about capacity per drive, capacity per rack, power per terabyte, cooling, physical footprint, drive count, failure domains and total cost of ownership. That’s a completely different optimization problem.
Google’s Colossus storage system is a good example. Google describes Colossus as using a mixture of flash and spinning-disk storage, with SSDs used for the data that benefits most from their performance and HDDs providing much of the underlying bulk capacity. That’s not nostalgia. That’s engineering. The storage system isn’t asking, “Which drive is fastest?” It’s asking, “Which drive makes sense for this data?”
And that’s a much better question.
Capacity Changes the Equation
This is where hard drives get really interesting. An SSD’s performance advantage doesn’t disappear when you scale storage up, but the economic value of that performance depends heavily on the workload. If you need a database to respond in milliseconds, paying for flash makes sense. If you need to retain enormous quantities of data that might be accessed occasionally, the calculation changes.
A 20TB drive doesn’t just give you 20TB. It means you need fewer physical drives to store a given amount of data. Fewer drives can mean fewer drive bays, less rack space, fewer interfaces, less supporting infrastructure and potentially lower power and cooling requirements for the same usable capacity.
That’s why areal density matters so much in enterprise HDD development. The goal isn’t simply to make a hard drive’s benchmark score look better. It’s to put more terabytes into the same physical infrastructure. And the industry is still pushing this aggressively.
In March 2026, Seagate announced that its next-generation HAMR platform, Mozaic 4+, had been qualified for production by two hyperscale cloud providers, with capacities reaching 44TB per drive. Seagate says those drives are already shipping in volume.
That’s the real reason I’m still interested in hard drives. The technology isn’t standing still.
And If You Want Cheap Storage…
If you’re looking for a lot of storage without spending a fortune, used enterprise HDDs are another option. They’re all over eBay and the used-PC-parts market, often with capacities that would be expensive to buy new.
You do need to know what you’re looking at. Drive hours, interface, recording technology and the seller’s reputation all matter. But there’s a huge secondhand market for enterprise drives.
You’re obviously taking on more risk buying used hardware, so they’re not something I’d recommend for your only copy of important data. But for bulk storage, backups or a home server where you have redundancy, used enterprise HDDs can be ridiculously cheap.
Sometimes Slow Is Fast Enough
This is where my own experience with mechanical storage fits. My primary PC has a 6TB Western Digital Blue HDD that transfers at roughly 150MB/s in real-world use. That’s laughably slow compared with NVMe. But it’s being used for a workload where that speed is perfectly adequate.
The data doesn’t need to be accessed with sub-millisecond latency. I’m not running an application from it. I’m not trying to load a 150GB game from it as quickly as possible. It’s storing terabytes of media and backup files that don’t make sense taking up space on my SSDs.
At some point, faster stops being particularly useful. If a file takes 30 seconds instead of 3 seconds to transfer, that’s noticeable. If the alternative is spending considerably more money for performance I don’t actually need, the calculation gets less obvious.
This is why I don’t think comparing an HDD and an NVMe drive by sequential throughput tells us very much by itself. They’re solving different problems.
Then There’s the Part Nobody Talks About Enough: SMR
There’s another reason I’m reluctant to reduce HDDs to “old, slow storage.” Modern hard drives have some pretty interesting recording technologies. The one consumers really need to understand is SMR — Shingled Magnetic Recording.
Traditional CMR — Conventional Magnetic Recording writes tracks independently, with enough spacing that a track can generally be modified without disturbing its neighbors.
SMR overlaps those tracks, much like shingles on a roof. That allows manufacturers to increase areal density and store more data on the same platter. It’s a clever engineering tradeoff. But there is a catch.
When existing data needs to be modified, overlapping tracks can mean that the drive can’t simply rewrite one track in isolation. The drive may have to reorganize and rewrite neighboring data. Drive-managed SMR hides much of this complexity through its firmware, mapping and caching. For some workloads, that’s perfectly acceptable. For others, it can become a nightmare.
SMR Is The Reason You Think Hard Drives Suck
This is where my experience with SMR finally made sense. You start a large write. The drive looks great. 150 MB/s. Maybe 200 MB/s. “Okay, this isn’t bad.” Then you keep writing. Eventually the available cache fills and the drive has to spend more time doing the internal work required to manage those shingled tracks than actually writing your new data.
The result can be a dramatic reduction in sustained write performance. The important distinction is burst performance versus sustained workload performance. A short benchmark can make an SMR drive look fantastic. A long, write-heavy workload can tell a very different story.
Imagine buying a car advertised as being capable of 150 MPH. You get on the highway. It hits 150 MPH. Great. Five minutes later: “Okay, we’re doing 10 MPH now.” “Why?” “You kept driving.”
That’s basically what SMR felt like the first time I encountered it. The drive wasn’t necessarily broken. I had simply discovered that the number I saw at the beginning wasn’t the whole story.
SMR Isn’t Entirely Bad
And this is where I don’t want to make the same mistake in the opposite direction. It exists because increasing areal density is valuable. For workloads dominated by sequential writes, infrequent changes and periods of idle time, drive-managed SMR can be perfectly reasonable.
The problem comes when you put that drive into a workload involving constant rewriting, heavy random writes or other sustained write activity and expect it to behave like a CMR drive. That’s when CMR vs. SMR becomes an important specification rather than an obscure storage acronym. And I think consumers should know which one they’re buying.
One Other Advantage: Sudden Power Loss
Here’s another advantage that doesn’t show up in benchmark charts. HDDs can be more forgiving than SSDs when power suddenly disappears, particularly with data that had already been written. A sudden power loss can still cause data corruption on either type of drive, especially if something was actively being written, so this isn’t a replacement for a UPS or backups.
It’s also not something I’d choose a hard drive for. These situations are relatively rare, and modern SSDs have their own protections and failure characteristics. But it’s another reminder that storage performance isn’t the only thing that matters.
HDDs Didn’t Lose. They Got a Different Job.
I don’t think the future of storage is SSD versus HDD. It’s both. Put the workloads that need extremely low latency and high IOPS on flash. Put massive amounts of less latency-sensitive data on high-capacity magnetic storage. Use caching and storage tiers when the workload benefits from both.
That’s exactly what modern systems are doing. Google isn’t replacing every HDD with an SSD. It’s putting the data that benefits from flash on flash while retaining enormous quantities of data on HDDs.
And Seagate isn’t developing 44TB HAMR drives because somebody forgot to tell them SSDs exist. They’re developing them because the world keeps generating data faster than anyone wants to delete it.
SSDs are amazing. They’ve earned every bit of the hype they get for the jobs they’re good at. But a storage device isn’t automatically better because it’s faster. Sometimes you need extremely low latency. Sometimes you need millions of IOPS. Sometimes you need several gigabytes per second. And sometimes you need an enormous amount of storage that doesn’t have to respond in milliseconds.
That’s why I like hard drives. Not because they’re old. Not because they’re nostalgic. Not because I think they’re secretly faster. I like them because they still solve a problem that SSDs haven’t made disappear.
*whirrrrr… tik-tik-tik…* (hard drive writing this article to disk.)