Should You Build Your Own Off-Site Backup Server or Rent One?

You need a few hundred terabytes of somewhere-else storage for backups, and the instinct is obvious: buy a dense storage chassis, fill it with drives, ship it to a colo, and you own the whole thing outright. That instinct made a lot more sense two years ago than it does right now.

This is a genuinely common request, and the underlying workload is usually the same shape: write-heavy, rarely read, needs to survive for years, and doesn’t need much CPU or RAM to do it. The question is whether building the box yourself is actually the cheaper, simpler path once you look past the sticker price of the hardware.

The DIY Instinct, and Why It’s Reasonable

A 36-bay dense storage chassis is a well-understood, well-supported category. Supermicro and others have sold this form factor for years specifically because it hits a sweet spot for organizations that want maximum capacity per rack unit without paying for compute they don’t need. If your team already runs infrastructure and has colocation space, buying one of these and shipping it to a second site as an off-site replication target feels like the straightforward move.

The math looks simple at first glance: chassis plus drives, shipped once, owned forever. No monthly bill, no vendor relationship, no recurring cost beyond power and colo space. For a team with the operational muscle to rack, cable, and monitor hardware already, that’s a real, legitimate option, and it’s worth taking seriously rather than assuming a rental is automatically better.

What the Hardware Actually Costs Right Now

Here’s where the calculation gets less simple than it did a year or two ago. Hard drive prices have moved sharply in 2026. Enterprise HDD pricing has climbed roughly 46-50% between September 2025 and March 2026, driven by AI-related demand pulling NAND production capacity away from traditional drives, tariff exposure across the supply chain, and buyers shifting from SSD to HDD as SSD costs rose even faster. A high-capacity enterprise drive that priced out around $11/TB in late 2025 is running closer to $16/TB as of mid-2026.

Run that against a real-sized build. Thirty-six bays at 20TB each is 720TB raw. At today’s roughly $16/TB, that’s approximately $11,500 in drives alone, before the chassis, dual CPUs, RAM, a RAID or HBA controller, and networking gear that goes around them. And that’s the list price for drives shipping today. Component pricing in 2026 has been moving during the weeks between requesting a quote and placing an order, not just year over year, so a number you priced out last quarter may already be stale.

The Costs That Don’t Show Up on a Parts List

The purchase price is the easy part to calculate and the smallest part of the real commitment. A few things that come with owning the box outright rather than renting equivalent capacity:

• Redundant power and networking. A single chassis with a single power supply and a single network path is a single point of failure for your entire replication target. Building in real redundancy means a second PSU, a second switch port, and ideally a second uplink, all added cost on top of the base build.
• Spare drives on hand. With 36 drives in one chassis, a failure isn’t a rare event over a multi-year service life, it’s a routine maintenance item. Budgeting for cold spares, and the shipping and labor to swap them, is part of the real cost, not an exception to it.
• Colocation space, power, and remote hands. An off-site box needs an actual off-site location, with its own power draw, cooling load, and either your own staff traveling to it or a remote-hands contract for anything that needs physical access.
• The upgrade problem. When larger, cheaper drives hit the market in two or three years, a fixed 36-bay chassis with old drives in it doesn’t get bigger. You’re either living with what you bought, or planning a second capital purchase and a data migration between them.
• Your team’s time. Firmware updates, RAID rebuild monitoring, capacity planning, and the eventual hardware refresh all draw on staff time that has a real cost even when it doesn’t appear on an invoice.

None of this means owning the hardware is a bad decision. It means the honest comparison isn’t drive cost versus monthly rental rate, it’s total multi-year ownership cost, including the failures, against a fixed monthly number that already has all of that built in.

How Ceph Handles the Redundancy Question Differently

A single 36-bay chassis, however it’s configured, concentrates risk in one box. RAID protects against individual drive failures, but a chassis-level failure, a backplane, a controller, a PSU that takes the whole unit down, has no fallback if that’s your only replication target.

Ceph’s approach spreads that risk across multiple nodes instead of concentrating it in one. Using erasure coding rather than simple replication, a properly configured cluster can tolerate multiple simultaneous drive or node failures while using a fraction of the storage overhead that full replication would require, closer to 18% overhead than the 200% overhead of triple-replicating everything. We cover the mechanics of this in more detail in our breakdown of OpenMetal’s storage tiers from hot to cold, including what that overhead math looks like at real cluster sizes.

The practical difference for a replication target specifically: a single chassis failing is a full outage of your off-site copy until you fix it. A properly built multi-node Ceph cluster losing a node is a degraded state the cluster keeps serving through while it rebuilds, not a phone call at 2am about your only backup being unreachable.

What Renting Actually Solves

Renting equivalent capacity converts that whole list of hidden costs into one fixed number, and shifts the operational burden, drive failures, firmware, hardware refreshes, spare parts logistics, onto the provider rather than your team. OpenMetal’s storage servers are built around exactly this profile: HDD-dense configurations for cold and archival data, paired with an NVMe tier for the metadata operations that keep even a mostly-cold cluster responsive, with drive failures and hardware refreshes handled as part of the service rather than a project your team owns.

Predictable egress pricing matters here too. Cloud cold-storage tiers often charge the most exactly when you need to retrieve data, whether that’s a real disaster recovery event or simply migrating away from a provider later. OpenMetal’s 95th-percentile bandwidth billing avoids that specific trap: pulling a large volume of data during an actual DR event doesn’t turn into a retrieval bill on top of everything else going wrong that day.

Who Should Actually Build Their Own

• Build your own if: you already have colocation space and staff who manage physical hardware routinely, your capacity needs are stable enough that the upgrade problem won’t bite you for years, and the multi-year ownership cost genuinely comes in lower once you’ve priced in redundancy, spares, and labor honestly.
• Rent if: you want a fixed monthly number instead of a large upfront purchase followed by an unpredictable maintenance tail, you don’t have spare operational capacity to own physical hardware at a second site, or your capacity needs are likely to grow in ways a fixed chassis can’t absorb without a second purchase.
• Either way: price drive costs at today’s rate, not a number from last year’s budget, since 2026’s hardware market has moved enough that stale assumptions will make either option look wrong.

Getting Started

Current storage server configurations, including HDD-dense options built for exactly this kind of off-site archival and backup workload, are on our storage cluster pricing page. For a deeper look at how OpenMetal structures storage across hot and cold tiers in the same cluster, see our storage tier architecture breakdown.

FAQ

Is it cheaper to build my own storage server or rent one?

It depends on your team’s existing operational capacity and how honestly you price the full multi-year cost, not just the drives. Drive prices alone have risen roughly 46-50% since late 2025, and a full ownership comparison needs to include redundant power and networking, spare drives, colocation costs, and staff time, not just the parts list.

How much have hard drive prices increased in 2026?

Enterprise HDD pricing has climbed approximately 46-50% between September 2025 and March 2026, driven by AI-related demand shifting NAND production capacity, tariff exposure, and buyers moving from SSD to HDD as SSD costs rose. Confirm current pricing before budgeting, since it has continued moving through 2026.

What happens if a drive fails in a 36-bay chassis I own?

With 36 drives in service over a multi-year period, individual drive failures are a routine event, not a rare one. RAID protects against data loss from a single drive failure, but budgeting for cold spares, replacement labor, and rebuild time is a real, recurring cost of ownership.

Does Ceph handle hardware failure differently than a single storage chassis?

Yes. Ceph distributes data across multiple nodes rather than concentrating it in one chassis, so a node failure degrades cluster performance during rebuild rather than taking the entire storage target offline, which matters specifically when that target is your only off-site backup copy.