What did data centers do before AI?

AI data centers are perhaps the most controversial topic in the tech world today. They are an integral part of using artificial intelligence programs like ChatGPT and Google Gemini, but are also known to cause problems. Their noise makes living near a data center a kind of dystopian nightmare, and the carbon emissions of a single data center can exceed those of a small country. These problems arose from a sudden and exponential expansion of data centers coinciding with the AI ​​boom, but are not really new. Data centers have been around for decades, dating back to the early days of computing. They began as individual rooms housing the machines of ancient mainframe computers and evolved alongside computers every step of the way.

The definition of a data center is quite simple: it is a physical space (originally rooms, now entire buildings) housing an IT infrastructure that operates 24 hours a day. Data centers have powerful generators to keep computers running even if the local power grid fails, and high-capacity Internet connections that allow far more traffic than the best WiFi routers.

They’re used to store, process, and transmit all kinds of data, from email exchanges to credit card transactions to the prompts of today’s chatbots. Every time you save something in the cloud or use a cybersecurity program to protect your data, you’re using a data center. So, where exactly did data centers come from and how have we become increasingly dependent on them?

The first data centers existed alongside the first computers of the 1950s and 1960s. The first version of a data center was the mainframe, a term for large supercomputers capable of performing complex calculations and usually requiring an entire room to house them. At that time, the concept of computer networks had not yet emerged, so each mainframe computer was like its own isolated data center. They developed during the 1970s and 1980s, but their next real evolutionary leap would not come until the microprocessor boom and the birth of the Internet in the 1990s.

This decade saw the advent of virtualization, whereby one physical device is used to create multiple virtual devices, such as networks or servers. In the early 2000s, virtual servers were available for rental at increasingly affordable prices, and suddenly the demand for data centers began to increase. Virtualization has helped usher in a centralized system of computer networks where almost all devices now rely on only a small handful of data centers.

Every step in the history of data centers has been driven by, and in turn driven, rapidly growing demand. First there was the microprocessor boom, then the cloud computing boom and now the AI ​​boom. Each new technology allows users to demand more from data centers, which have been forced to expand to entire campuses of buildings. Although they existed for decades out of the public eye, data centers are now literally too important to ignore.

Even today, with the AI ​​boom in full swing, not all data centers are designed to power artificial intelligence programs. There are a wide range of needs that data centers meet, as well as unique types of data centers to meet each goal. The Telecommunications Industry Association (TIA) divides data centers into four tiers. All tiers provide basic services such as power and cooling supplies 24/7. However, Tier 1 data centers only meet these basic needs, while higher tiers provide more, up to Tier 4, which operates completely isolated energy systems capable of operating in the event of a disruption.

Across these tiers, you can find data centers geared toward a number of specific goals. There are telecommunications data centers that support telecommunications networks like the Internet and cellular services. There are enterprise data centers, owned and operated internally by organizations such as banks and hospitals, that process highly sensitive data requiring additional security. And then, of course, there is everyone’s favorite: crypto mining data centers.

AI data centers are just the latest member of the vast data center family. They also come in several subtypes, namely training AI data centers and inference AI data centers. Training centers are, as the name suggests, dedicated to training AI programs by feeding them huge amounts of data, while inference AI data centers support active AI models and ever-increasing user demands.