The Past, Present and Uncertain Future of Collocation Data Centers
Colocation data centers, sometimes called Internet data centers came into existence in the late 1990's. Massive server farms, with racks of servers and cages as far as you could see, colocation data centers were going to revolutionize information technology (IT). No longer would corporations need to build their own data centers, most of their IT services could be outsourced to colocation data centers. With data centers located worldwide and worldwide connectivity from multiple fiber optic carriers, your data processing equipment could literally be located anywhere. Disaster recovery issues would be resolved once and for all, since you could mirror your data at one or more of their multiple locations.
These massive 100,000 to 200,000 square foot data centers would have reliability levels never seen before, (5-9's or 99.999% availability), infrastructure that was designed to meet power requirements for today's technology and tomorrow's technology that was still being developed, massive bandwidth requirements from multiple fiber optic carriers, dedicated electrical service from the utilities and projected occupancy and lease rates that allowed owners to recover the costs of these data centers in less than 5 years. Seemingly overnight, there were companies formed, funded and underway designing and building data centers worldwide. It was a race, since some early research indicated that the first data centers up and running in a city had a distinct advantage over those that came later. Speed to market was more important that cost. Projects that we used to complete in 2 to 3 years would now be designed and built in 8 to 12 months!
Some of the names Exodus, Global Centers, PSI Net, Above Net, WAAVE, Telegis and many others went from startups to recommended investments by stock market analysts in the span of 12 to 18 months. Some of these companies were spending 2 billion dollars or more per year in construction costs for new data centers. The market for colocation space was projected by some to be unlimited, with over 2 million square feet of raised floor space being built in the Bay area alone. Sacramento had its share with at least a dozen companies either building or contemplating building colocation data centers in the city or county.
Projected power requirements for these data centers were so great that many utilities embarked upon massive capital expenditures to meet what turned out to be a fictional requirement. Nationwide, only 10% of the projected power requirements ever occurred. This required utilities to oversize their infrastructure dramatically in certain areas. Many data centers built during that timeframe have 10 to 12 megawatts of capacity and are only using 1 to 2 megawatts.
So what happened? Why are none of those companies in business now? How could companies worth tens of billions of dollars at their peak, become a penny stock and then worthless in the span of 2 years? It's an interesting story and one that mirrors the collapse of technology stocks in general. A story that I was a part of as the Director of Global Data Center Design for a company that was building 23 data centers worldwide over a 2 year period of time. As the first employee hired, I was part of the initial decision making process. I saw firsthand the assumptions that were made, the development of the business model and the problems both self-made and beyond our control that led to the collapse of funding and the collapse of the demand for these data centers.
The first issue assumption that was made, that I believe was false, was that companies would jump at the opportunity to outsource their data processing to a colocation data center. This was the basis for much of the growth projections that were made for the exponentially increasing demand for colocation data center space. It was based on the fact that dot com companies who didn't have large established IT programs, quickly embraced the concept of colocation data centers. These dot coms were often growing at rates that made it impossible for them to plan, build and manage their own data centers. A colocation data center company with plans to build 2 million square feet of raised floor space on 3 continents allowed the dot coms to focus on their core business, which didn't include designing, building and managing data centers.
This assumption however, was also applied to large existing corporations and government entities. Corporations and government entities have professional IT staff that have been managing their data centers for 20 years or more. IT staff and managers who would fight tooth and nail the concept of outsourcing what they saw as their livelihood. While there were a few large corporations that did outsource their IT, these contracts typically went to large, well-established companies like IBM and EDS, not the smaller colocation data centers without an established track record. No significant government IT was ever outsourced to a colocation data center that I am aware of.
The next problem was the business model. Basically, there are three ways to make money in a colocation data center:
You can make money leasing the floor space. Rental rates for raised floor space at one time were as high as $150 per square foot, per month. This only lasted however until the amount of raised floor square footage on the market exceeded the demand. At that point, rates dropped to where today, some of the colocation data centers will give away the raised floor space if you sign up for a certain level of other services from them. Carrier Hotel estimates the current average rental rate for raised floor space nationwide to be in the $25 per square foot range.
You could also make money marking up the bandwidth requirements of your customers. Initially, when there was little competition, the mark up for bandwidth exceeded 200% for some colocation data centers and you were often required to use their preferred carriers. Once carrier neutral data centers came into existence, competition dropped the prices until today bandwidth is marked up minimally, if at all.
The third way to make money was selling the customers managed services. The argument was that all of the services required to manage a client's systems and data could be done more cost effectively by colocation data center staff than by the clients staff. These managed services included daily or weekly backups, mirroring of data at other sites for disaster recovery purposes, software and hardware upgrades, etc. This certainly made sense where clients had equipment and data in multiple locations. These additional services in some cases doubled or tripled the revenue a colocation data center received per square foot. It seemed like this would give colocation data centers the long term source of revenue they'd been looking for.
It wasn't long however before problems began to appear with the managed services business model. The expertise that it required to provide these services required colocation data centers to hire some highly skilled and therefore highly paid staff. Providing staff that could manage software, hardware and network problems or upgrades, and qualified help desk personnel that could interact with the customers 24 hours a day/7 days a week was an expensive undertaking. These staff had to be highly leveraged so that they had billable hours the majority of the time, in order for a colocation data center to make money on the services they provided. When the dot coms crashed and the demand for colocation data center space and the managed services they required collapsed, there were a lot of data centers with a lot of highly paid people with little work to do.
Another big problem for colocation data centers was the cost of building these facilities. Construction costs started out high and went higher. While some of the early data centers were built for $500 per square foot or less, the last generation of data centers being built were costing $2,000 per square foot or more. At $2,000 per square foot, a 100,000 square foot facility was costing $200,000,000. With that kind of capital investment up front, recovering the cost of a facility was going to be measured in decades, not years.
What drove the construction costs up to these levels? There were a number of factors that had an impact. You paid a premium to design and build facilities in 8 months instead of 24 months. Data center designs were also being driven by marketing concerns, not actual customer requirements. The higher the power densities, the better the data center. Despite the fact that the average electrical load for data centers is still less than 30 watts per square foot, colocation data center designs started out at 70 and went up to as high as 200 watts per square foot. Complexity was confused with reliability. Redundancy was added on top of redundancy. All of these factors and more drove the construction costs to levels that made the business model unworkable.
Once projected to dominate the market, colocation data centers now make up less than 10% of the data center space currently constructed. Corporate data centers, financial services data centers, educational and government data centers make up the vast majority of data centers in use today. The movement to outsource data processing from corporate data centers to colocation data centers simply never occurred at the levels that were predicted. Without a growing demand for their services and with few outside sources of funding, many colocation data centers are now sitting empty or were never completed.
In my opinion, the future of colocation data centers is now clear. The companies that are doing well are ones that have kept their capital costs to a minimum. Companies who didn't buy into the hype and designed cost effective data centers. Companies that only built out a portion of their ultimate capacity and are adding raised floor square footage as they lease the space out that they already have built. Companies that have purchased bankrupt colocation data centers for ten cents on the dollar and don't have the large capital expenditures that have to be amortized. While the focus in the past had been on reliability and speed to market, the focus now is on operating costs and energy efficiency. Colocation data centers are now being run like any other business, with a business model focused on current, not future earnings.
Service offerings are more realistic now than they were 3 years ago. Most successful colocation data centers have conceded that they are not going to be the provider of outsourced data processing services for large corporations or government entities. They have begun to focus on the small to mid size companies that have the most need for their services. There is also a growing market for providing disaster recovery and business continuity services for companies of all sizes. Colocation data centers with multiple locations and diverse networks are a great place to mirror critical data, use as a backup site or set up electronic vaulting of backup tapes.
Sacramento at one time had a dozen companies designing or building colocation data centers here. As the market for colocation data center space dried up, most of the sites have either been sold or converted] back to office buildings. There are only two large collocation data centers not associated with carriers left in Sacramento, Herakles Data and Raging Wire Telecommunications. Both are located in the Natomas area and both mirror this new economic reality. LANSET Communications is another smaller colocation data center located in Rancho Cordova.
Herakles Data Center is a 92,000 square foot facility, with 60,000 square feet of raised floor built out. The raised floor is divided into two separate east and west colocation areas. Designed and built for WAAVE data centers, Herakles purchased the facility in 2001 and has been operating it for the last two years. Herakles is privately held company. Designed by EYP - Mission Critical Facilities, the data center is typical of the latest generation Internet data centers with 5? 9's reliability levels and a minimum of N+1 redundancy levels.
Raging Wire Telecommunications is located in a 200,000 square foot facility that has 30,000 square feet of raised floor currently built out, along with offices and warehouse space. The balance of the 200,000 square feet is unfinished. Raging Wire is a privately held company. The data center was designed by Ellerbe Beckett, with 5?9's availability levels and redundancy levels that range from N+1 to N+5.