Glossary
Cold Site
A cold site is a backup data center that has space, power, cooling and network connectivity, but no computing equipment installed. If your primary site is lost, you bring in hardware, build it, restore your data and start running there.
It is the cheapest disaster recovery option and the slowest. Realistic time to get production running again is days to weeks, and most of that is not spent on technology.
What you actually get
A cold site contract typically covers:
- Floor space, usually specified as a number of racks or cabinets.
- Conditioned power to a stated density, in kilowatts per rack. This number matters more than it looks — a room provisioned at 5 kW per rack cannot host modern dense servers that draw two or three times that.
- Cooling sized to match that power draw.
- Network connectivity — cross-connects, carrier access and usually a choice of providers.
- Physical security and access, including who is allowed in and how quickly they can get badged.
What is not included: servers, storage, network switches, your software licences, and your data.
Some providers offer a variant where they hold equipment on retainer and deploy it when you declare. That removes the longest delay, costs considerably more, and is closer to a warm site than a true cold one.
What has to happen before you are running
| Stage | Typical time | What drives it |
|---|---|---|
| Declare and activate the contract | Hours | Who has authority to declare, and provider response terms |
| Acquire hardware | Days to weeks | Supplier stock and lead time — the longest stage by far |
| Ship, rack and cable | 2–5 days | Distance, site access hours, and how many hands you have |
| Install and configure | 2–7 days | Whether the build is automated or done from documentation |
| Restore data | Hours to weeks | Dataset size, where the backup lives, and read-back speed |
| Test and cut over | 1–3 days | Application dependencies and how much you are willing to verify first |
Hardware is the long pole
Enterprise servers and storage on a normal purchase cycle run four to twelve weeks from order to delivery. Emergency procurement compresses that to days or weeks, but at a premium and with whatever configuration the supplier happens to have in stock rather than the one you specified.
The scenario that breaks the plan is a regional event. If a flood, storm or grid failure has taken out other organisations in the same area, all of them are calling the same distributors in the same week. Availability that looked fine in the plan is gone precisely when it is needed.
This is why cold site recovery estimates built from a technology standpoint are usually optimistic. The systems work is measured in days; the supply chain is measured in weeks.
Where the data comes from
Cold site recovery is entirely a restore, so the location and format of your backups sets the second half of the timeline.
Tape in an offsite vault. Cheap to hold and slow to start. The tapes have to be requested, retrieved and physically transported, which is commonly 24 to 48 hours before you can begin reading anything — and you need compatible tape hardware at the recovery site, which is itself something to procure.
Backup in cloud or remote object storage. Nothing to transport, but you are limited by bandwidth and egress. Pulling 100 TB across a 10 Gbps link at realistic efficiency takes on the order of a day and a half, and cloud providers charge for the data leaving.
Backup already replicated to the cold site. The fastest option, and the point at which the site stops being genuinely cold — you are now paying to keep storage running there.
What it costs
A cold site is the cheapest of the three tiers because the expensive part is absent. The recurring bill is rent, power for an empty room, connectivity, and whatever the provider charges for holding the option.
The costs that arrive later are less visible. Emergency hardware is bought at emergency prices. And because nothing is running, there is no meaningful way to test the arrangement — you cannot fail over to a room. The plan stays a document until the day it is used, which is also the day you find out what was missing from it.
Who uses a cold site
Organisations that can be down for weeks without existential damage, and would rather spend the money elsewhere. In practice that tends to be smaller organisations where the alternative is no recovery site at all, archives and secondary systems inside larger ones, and cases where a regulator or insurer requires a documented recovery location without attaching a recovery time to it.
It also covers a specific risk well: total, long-term loss of a building. Fire, flood or a lease ending badly. For that scenario a cold site is a reasonable answer, where a hot site would be paying continuously for speed you do not need.
What the storage layer decides
Because recovery is a restore, two properties of the backup copy set the outcome.
The first is whether it survived. Ransomware and accidental deletion propagate through replication and backup jobs alike, so the copy that matters is the one that could not be altered. Scality RING and Scality ARTESCA both support S3 Object Lock in governance and compliance modes, which keeps a retained copy intact even when production credentials are compromised.
The second is how fast it reads back. Restore throughput is what turns a dataset size into a number of days, and it is the part of cold site recovery the storage platform actually controls. ARTESCA is built for this role — backup-first object storage validated from 20 TB to 8.5 PB — and the number worth knowing about any backup platform is not how fast it ingests, but how fast it gives the data back.