Reflections On 20 Years in The Data Center Industry

Juan Font, President and CEO of CoreSite and SVP of American Tower, explores "industry earthquakes" reached during two decades of data center industry evolution and the sea change triggered by artificial intelligence.

At the turn of the millennium, large telecommunications carriers needed to be able to interconnect — meaning, exchange traffic with each other. Say you were in Berlin and trying to call a friend in Madrid: Deutsche Telekom needed to originate the call and then hand it off to Telefónica for you to have that conversation.

As you can imagine, this process was not terribly efficient. So, what we call “carrier hotels” sprang up; these were single locations where multiple networks were connected, such as One Wilshire^®, in Los Angeles, or 60 Hudson, in New York. Similar carrier hotels were developed around the world, all acting as hubs for carriers to interconnect. 

Some nascent data applications were running in these carrier hotels, but most of the traffic was voice-only, relying on very low-density gear. (DC power was still the main driver; AC power hadn’t really taken off.) And, for a while, this worked fine for everyone’s needs … until, suddenly, the world changed.

Phase 1: The Digital Era 

Remember MySpace? In the early aughts, early social networks like LiveJournal and Six Degrees were gaining momentum; MySpace emerged in 2003, soon to be eclipsed by Facebook (2004) and YouTube (2005). Peer-to-peer file-sharing networks such as Napster, Gnutella and Freenet were also popular, internet portals like AOL and Yahoo became dominant and more and more people started creating their own websites. Unsurprisingly, the infrastructure in most carrier hotels proved woefully inadequate to support these changes; the servers on which these applications ran simply consumed far more power than their voice-only predecessors. 

The result was the creation of data centers: purpose-built facilities that could support these power- and data-intensive workloads. Increasingly, corporate customers wanted access to all the connectivity of carrier hotels, but on a high-density, modern, scalable platform. Data center campuses, where two or more data centers in close geographic proximity housed many large racks, supported the workflows of many different enterprises in the same or a connected facility — and this became a winning business model. Both enterprise and personal internet use became the norm.  

Phase 2: The Cloud

The next big inflection point was the advent of a brand-new public utility known as “the cloud,” in about 2006. Computer services such as data storage, compute, analytics, database programs and servers all suddenly became accessible via the internet, without the need for any active management by individual users. Amazon, Microsoft, Google and Oracle joined the fray and, very quickly, cloud computing became a commodity that customers consumed on demand. The public cloud required dedicated data centers, which gave rise to the term “hyperscale.”

Multi-tenant data centers had to respond aggressively to this wild uptick in demand for faster, more powerful resources. So, they adapted to the emergence of hybrid IT use cases to serve organizations that were transferring an ever-increasing number of workflows to the cloud. With the influx of enterprises that suddenly demanded bigger, faster, more powerful and resilient digital infrastructure, data centers came into their role as third-party hosts for multiple, aggregated network providers. Moreover, they pivoted to heighten both physical security and access to digital security in an increasingly tight data regulatory environment. 

Phase 3: The AI Boom

In a sense, AI is an evolution of the cloud — it’s not new, conceptually. The IT challenge du jour is finding ways to support increasingly complex inference models and analytics. However, there is a vast difference in the underlying architecture between CPUs, which power your laptop, and GPUs, which are required for today’s AI models. The high-performance chips that power GPUs can consume 10 times more power than their non-AI predecessors, and with that massive uptick comes a host of increased demands for electricity, cooling, security and installation space. 

Companies are rapidly turning to colocation and competing for a strained supply of data center space to gain access to these resources at scale. Data center providers are under significant pressures: to expand (including building new data centers) to meet the increased demand, to deliver both robust connectivity and low latency, to increase power and rack density, to have power redundancy and resilience. All of this makes today’s challenges all the more consequential, even if they’re not inherently novel.

Where We Are Now

Of these three data center industry earthquakes, if you will, there is no question that the third has the greatest magnitude. And for enterprise leaders across all other industries, it has presented a host of difficult questions: What workloads are imperative to our business? Which of these must be driven by AI? Can we afford the necessary infrastructure and resources to support these workflows? Do we repatriate these workloads on-premises or shift to colocation? Can we lease the space we need? When? For how much? 

We’re in a period wherein more data is being created than can feasibly be transported over vast distances, and this ratio is only widening. Consequently, enterprise leaders must think months or years ahead to secure the resources necessary to implement their AI transformations.

Within our data center industry, this period represents both a massive opportunity and also an admonishment of sorts. The disruption caused by GenAI has shown that we simply cannot become complacent. Sometimes an event like the AI revolution happens, spurring not linear but exponential growth. The changes are earth-shattering — but we cannot be caught on the back foot and expect to survive. Agility is essential in the Fourth Industrial Revolution, as we fundamentally and permanently alter the ways we work, communicate and live.

Know More

Enjoy this article? You can see all of Juan's contributions (over the past two years) in his area of the Forbes Technology Council website.