From mainframes to modern colocation: the history of data centres

Posted on in

Most histories of data centres give you a neat timeline and move on. ENIAC appears, cloud computing arrives a few decades later, then AI takes over the final paragraph.

The more useful story is in the structural shifts between those milestones. Computing moved from single machines in dedicated rooms to shared facilities, then to virtualised infrastructure and hyperscale cloud. In the UK, commercial colocation and internet exchange infrastructure developed together, helping turn London into one of the world’s major connectivity hubs.

And through all of it, one thing has remained true: the hardware has always depended on the physical environment around it.

When did data centres begin?

There is no perfectly clean answer to the question of what the first data centre was, partly because dedicated computer rooms existed before anyone widely used the term.

ENIAC, completed in 1945 and publicly unveiled in February 1946 at the University of Pennsylvania, is a sensible starting point for the physical concept. Built by J. Presper Eckert and John Mauchly, it occupied roughly 1,800 square feet and needed dedicated power and mechanical cooling to operate. The machine could not function without the room around it.

The physical idea therefore predates widespread commercial use of the term “data centre” by decades. Some use appeared in IBM contexts as early as the 1960s, but the term became much more common commercially during the 1990s.

Early facilities were largely private. Governments, military organisations, universities and major corporations housed their own systems in rooms built around the hardware they owned. These were not shared environments.

In 1954, Jean H. Felker’s TRADIC at Bell Labs demonstrated that transistors could viably replace vacuum tubes in computing. It was an airborne military computer rather than a data centre system, but it showed that computers could become smaller, cooler and more reliable.

Commercial mainframes were developing at the same time. IBM introduced the 701 in 1952 and the 650 in 1954. The IBM 704, first installed in 1955, pushed scientific computing further with floating-point hardware.

By then, organisations outside government were increasingly dedicating rooms to computing equipment, and those early computer rooms already had characteristics that remain familiar today: controlled temperature and humidity, specialised electrical supply and a building environment treated as part of the technology itself.

The difference was ownership. Every machine room was an internal asset, designed for one organisation. Shared colocation was still decades away.

Evolution of data centres: a decade-by-decade timeline

  • 1940s-1960s: the mainframe era

Mainframes dominated institutional computing. Universities, government agencies and the largest corporations were among the few organisations with the budget and need to operate them.

IBM became the dominant hardware supplier, while data centre design evolved around the physical requirements of large machines. Raised-floor construction became standard during the 1960s because it made cable routing and airflow management easier, and the approach remained common for decades.

  • 1970s-1980s: minicomputers and networked infrastructure

Minicomputers lowered the barrier to business computing.

Systems such as the DEC PDP-11, introduced in 1970, and the VAX, launched in 1977, brought computing within reach of a much wider range of organisations.

ARPANET had already begun connecting computing facilities across institutions from the late 1960s, laying some of the groundwork for networked infrastructure.

More businesses built internal machine rooms. UPS systems became standard protection against electrical interruptions, and raised-floor design was firmly established.

The important change was scale. Computing was no longer confined to governments, universities and the very largest corporations.

  • 1990s: commercial colocation, Telehouse Docklands and LINX

The internet changed what organisations expected from their infrastructure.

Email, web hosting and early e-commerce pushed businesses of all sizes to install servers, while growing connectivity requirements made the old model of isolated internal machine rooms less practical.

Commercial colocation emerged as an alternative. Instead of every organisation building its own power, cooling, security and connectivity environment, multiple customers could place equipment inside one specialist facility.

The history of colocation in the UK is closely tied to London Docklands.

Telehouse North opened in Docklands in 1990 as a purpose-built colocation facility. A few years later, the same location became central to another important part of UK infrastructure history.

In November 1994, five UK internet networks began exchanging traffic at Telehouse using a shared switch. That early exchange became the London Internet Exchange, or LINX.

The practical reason mattered. UK internet traffic between domestic networks could otherwise be routed through exchanges in the United States, adding cost and unnecessary distance. Direct interconnection kept more traffic local.

LINX became a formal company in 1995 and expanded rapidly as internet use grew. Its development helped turn London data centres into something more than buildings containing servers. Interconnection itself became part of the value of the location.

That is one of the defining changes in modern data centre history: a facility could offer power, cooling and physical security, but the ecosystem around it increasingly mattered too, including carriers, exchanges, low-latency routes and access to other networks.

The EIA/TIA-568 standard, published in 1991, reinforced another part of that transition by establishing a vendor-neutral approach to structured cabling. Organisations could mix equipment from different suppliers without rebuilding the physical layer every time the hardware changed.

  • 2000s: purpose-built facilities, virtualisation and resilience

Internet service providers and web businesses drove demand for purpose-built, multi-tenant data centres with redundant power and multiple connectivity options.

Compliance requirements added pressure. Regulations such as Sarbanes-Oxley in 2002 encouraged organisations to put greater emphasis on controlled, auditable IT environments.

Virtualisation reshaped the computing layer at the same time, changing how much hardware a given workload actually needed. VMware was founded in 1998, released VMware Workstation in 1999 and launched ESX Server in 2001. ESX had a much greater effect on data centre operations because it allowed multiple isolated workloads to run on one physical server.

Before widespread server virtualisation, individual applications commonly occupied individual machines. Consolidation increased hardware utilisation and reduced the number of physical servers required for the same amount of computing.

Resilience also became easier to describe and compare.

The Uptime Institute’s Tier classification system, originating with work published in the 1990s, gave operators and buyers a common language for facility redundancy and maintainability. Tier I through Tier IV became widely recognised descriptions of different resilience levels.

The dot-com crash in 2001 briefly left substantial commercial data centre capacity unused. Broadband adoption, e-commerce and continued internet growth absorbed much of that capacity over the following years, and the market consolidated rather than disappearing.

Energy efficiency then became a much bigger operational issue.

Power Usage Effectiveness, or PUE, was introduced in 2006 and promoted by The Green Grid from 2007 as a way of comparing total facility energy use with the energy consumed directly by IT equipment.

A PUE of 2.0 meant the facility was using roughly as much energy for cooling and supporting infrastructure as the IT equipment itself. Industry efficiency has improved substantially since then, particularly in newer hyperscale facilities.

  • 2010s: cloud, colocation growth and infrastructure at scale

Cloud computing changed the relationship between organisations and physical infrastructure.

The end user increasingly consumed computing as a service rather than interacting directly with a physical server. Behind that abstraction, providers were building larger and more specialised facilities.

Power, fibre, land and efficiency became increasingly important to site selection. Cooling evolved as rack densities rose, while liquid cooling began gaining ground in high-performance computing environments.

Edge computing developed alongside centralised cloud infrastructure, using smaller facilities closer to users for latency-sensitive applications.

Cloud therefore drove infrastructure in two directions at once: towards enormous centralised campuses and towards smaller distributed facilities closer to the workloads they served.

Mainframe computer centres vs modern data centres

The difference is not simply scale. It is architectural and commercial.

A mainframe room was generally built around one organisation’s hardware. Power, cooling and floor layout could all be designed for a particular machine or family of machines. Replacing the system could make much of the surrounding environment redundant.

Modern data centres are much more hardware-neutral. One facility can support x86 servers, ARM-based systems, storage platforms, network equipment and, in some cases, mainframes alongside one another, with the building providing the resilient physical platform rather than dictating the hardware that runs inside it.

Ownership changed just as significantly.

Early computer centres were internal cost centres. The organisation using the computing equipment generally owned and operated the room around it.

Colocation separated those responsibilities. An organisation could retain control of its own hardware while using a specialist provider for the facility, power, cooling, physical security and connectivity around it.

That change is central to the history of colocation. The client no longer needs to build a data centre simply to control its own infrastructure.

Modern colocation has continued to develop from there. The facility still matters, but so do the operational services around it: engineering support, service management, migration planning, connectivity and the people responsible for keeping the environment visible once equipment is deployed.

Instead of every organisation duplicating power, cooling, security and connectivity infrastructure internally, these capabilities can be shared within a purpose-built facility, enabling tenants to benefit from the economies of scale, industry influence and specialist expertise that dedicated operators provide.

Cloud computing and its impact on data centre infrastructure

Cloud pushed that separation further.

Amazon Web Services launched S3 and EC2 in 2006. Microsoft announced Azure in 2008 and brought it into general availability in 2010. Google was expanding its own cloud platform over the same period.

AWS, Microsoft Azure and Google’s cloud services allowed organisations to consume compute, storage and networking without owning the hardware underneath them.

At infrastructure level, this accelerated the software-defined data centre. Networking, compute and storage could be abstracted from individual physical devices and managed programmatically.

For the end user, infrastructure looked increasingly intangible. For operators, it became more physical than ever.

Cloud providers built ahead of demand, creating large campuses with substantial power requirements and dense network connectivity. Global data centre floor space expanded even as many individual businesses reduced their own on-premises server rooms.

The Open Compute Project, launched by Facebook in 2011, accelerated another shift by making data centre hardware designs openly available and encouraging operators to rethink server, rack and facility efficiency.

Large-scale cloud also brought energy efficiency into sharper focus. When a campus consumes enormous amounts of electricity, relatively small improvements in cooling or electrical efficiency have significant operational consequences.

Renewable energy procurement became part of site selection and long-term infrastructure planning. Major cloud operators made increasingly ambitious commitments around renewable electricity and carbon reduction, influencing how and where new facilities were developed.

The rise of hyperscale data centres

Hyperscale is the industrial-scale version of the same trend.

Google, Amazon, Microsoft, Meta and other large technology companies began operating facilities containing thousands of servers and enormous amounts of computing capacity.

Definitions vary, but hyperscale generally refers to facilities designed for very large, rapidly scalable computing environments rather than conventional enterprise estates.

Many hyperscale operators design their own servers, racks, cooling systems and networking equipment. That is a significant departure from the standardised, off-the-shelf model that shaped many commercial data centres in the 1990s and early 2000s.

Scale changes the geography too. Large campuses need substantial electrical supply, dense fibre connectivity and enough land to expand, requirements that have concentrated hyperscale development in a relatively small number of global markets while also encouraging operators to look beyond traditional city-centre locations.

Colocation providers have adapted too. Large shell-and-core facilities can now be built to hyperscale specifications and leased to cloud providers on long-term contracts.

The boundary between hyperscale cloud infrastructure and commercial colocation has therefore become less tidy than it once was.

2020s: AI rewrites the power and cooling assumptions

AI is changing what modern data centres need to support.

Training and inference workloads built around dense GPU clusters require much more power per rack than conventional enterprise computing.

Traditional colocation environments were commonly designed around rack densities of roughly 5 to 10 kW. AI inference deployments can require 40 to 60 kW per rack, while some training environments reach 100 kW or more.

That is not incremental growth. It is a step change.

Cooling has had to change with it.

At those densities, conventional air cooling becomes increasingly difficult and inefficient. Direct-to-chip liquid cooling, immersion systems and rear-door heat exchangers are becoming much more important as operators prepare facilities for high-density compute.

Power availability has moved to the front of site selection too. An AI campus may need electrical infrastructure and grid connections that take years to secure, and in established markets, including the UK, the availability and delivery date of power can determine whether a development is viable before questions of land or construction are settled.

Modular designs are one response. Factory-built data centre components can be deployed more quickly than traditional construction in some circumstances, helping operators add capacity while permanent facilities are still being developed.

The broader direction is clear. AI and high-performance computing are pushing facilities towards greater rack density, more demanding cooling and much larger power requirements.

The technology has changed. The physical constraints have not disappeared.

From data centre history to modern colocation

The history of data centres is really a history of changing relationships between computing and the infrastructure around it.

Mainframes needed specialised rooms. Networked computing made connectivity more important. Commercial colocation turned the facility into a shared service. Internet exchanges made location and interconnection part of the product. Virtualisation increased utilisation. Cloud moved infrastructure behind an API. Hyperscale changed the economics of construction. AI is now rewriting assumptions about power density and cooling.

But the fundamentals remain recognisable.

Business-critical infrastructure still needs reliable power, effective cooling, resilient connectivity, physical security and an environment people can operate with confidence.

That is where modern colocation sits today.

Datum provides secure, service-enhanced colocation from Farnborough and Manchester, with the facility infrastructure supported by service management, in-house engineering and teams that stay close to the environment after deployment.

Get in touch to talk through your infrastructure requirements, or find out more about our colocation services.

Back to Insights