What is a data centre?
It is many things to many people — from a bank’s transaction engine to a bitcoin mine. At heart, four types, all doing the same basic job.
Data centres come in all shapes and sizes — processing banking transactions, forecasting weather, even mining bitcoin. But at heart they are all the same thing: rooms or buildings that provide stable power and cooling to racks of computer servers, disk storage and network equipment. Those racks compute, store and connect — to each other and, via the internet, to the outside world. There are four types I see most.
Enterprise
The most common by far — a company's own data centre, anywhere from a "server room" with a few racks (and a misplaced mop bucket) up to a dedicated multi-megawatt facility run by a bank. When they are on site they are called "on-premise" or "on-prem". They are run by the company for its own business: a bank processing thousands of transactions a second, storing records, and talking to other banks.
Colocation
Colocation, or "colo", is outsourcing your computing to a third party. A provider such as Equinix or Teraco builds and operates the buildings, and companies rent space by the rack for their own servers. The C-suite likes it: it takes a big non-core asset off the balance sheet and removes an operational headache — banks know banking, and would rather do that than run data centres. Tenants range from SMEs taking one or two racks to banks taking hundreds, and hyperscalers taking anywhere from 25–100 racks up to 400-odd in South Africa.
The other thing colo does well is interconnection. Banks need to talk to each other, and to networks like Visa, constantly. If each bank sits in its own data centre across town, that traffic carries latency and network cost. Co-locate two banks in the same facility and they can cross-connect directly — much lower latency, and no charge to push data across the city. It is why corporates often sit in colo, their own enterprise data centre, and the cloud all at once.
Cloud (hyperscale)
Cloud data centres are built and run by the cloud providers — AWS, Google, Microsoft — for their own use. Hyperscalers build their own so they can concentrate computing power, dictate the design, and control the environment and costs. A hyperscale hall typically runs identical racks at a similar power density, tightly managed for efficiency — which lets them push boundaries (Meta has run data centres at 32 °C, something you can only do when you control the whole building). A colo, with many clients at different densities and older SLAs demanding 21 °C ± 1 °C, simply cannot move that fast. In Africa there is just one cloud operator so far — AWS, with three data centres in Cape Town — but as demand for cloud rises, others will follow.
Edge
The edge is the hardest to pin down — it means different things to different people. It might be Equinix building a 6 MW edge facility in Bordeaux, a smart mine putting micro data centres underground for its IoT, or Netflix installing a one-rack Point of Presence in your suburb. The common thread is moving computing closer to the customer to cut latency. I think the edge has a huge part to play in Africa, where a distributed model can lean on wireless networking to help solve the network build-out.