For many on the enterprise side of the data center market, cloud and colocation are often simplistically presented as competing offerings to a common problem. Seeking outsourced solutions to their data center capacity requirements, they are presented with the choice of either tapping into the cloud in the form of Infrastructure as a Service (IaaS), Platform as a Service (PaaS), or Software as a Service (SaaS) offerings, or shifting their computing resources to a hosted environment as offered by colocation providers.
Today’s reality is more complex than that. Many enterprises are finding a combination of the two to be the best approach. They may tap into SaaS offerings for applications such as productivity tools or sales force management, while also using colocation to meet new capacity demands for sensitive or custom applications. Their challenge isn’t really choosing between cloud and colocation. It is creating a hybrid environment that seamlessly integrates edge and core facilities as well as cloud and colocation services, to deliver near real-time capacity management across all resources.
The enterprise isn’t the only place where cloud and colocation providers are co-existing. Colocation offers cloud providers a solution to the challenge of managing continued growth.
Cloud computing is still growing at an impressive rate. In 2016, IDC projected a 19% compound growth rate for cloud computing between 2015 and 2020 (Figure 1). That forecast now appears conservative. The analyst firm says the public cloud services market grew 28.6% in the first half of 2017, with revenues totaling $63.2 billion, a pace significantly higher than the projection of $99 billion for the full year.
Figure 1. Projection from IDC shows spending on cloud services continuing to grow at double-digit rates through 2020.
According to IDC, the SaaS segment comprises approximately 69% of public cloud market share and experienced a year-over-year growth rate of 22.9% in the first half of 2017. The IaaS segment, which accounts for 17.8% of the market, grew at a rate of 38.1%. The smallest segment, PaaS, grew at the fastest rate — 50.2%.
Also notable, this growth is increasingly coming from regions where cloud providers may not have a strong established presence. Asia/Pacific saw the highest regional growth at 38.9%. It now represents 11.5% of all public cloud services revenues and is being driven by particularly strong growth in China, which saw 55.6% year-over-year growth in the first half of 2017.
This creates challenges for cloud providers as user demands for faster access to data and applications is forcing them to add capacity in the specific locations where demand is growing. Yet it can be difficult to predict where growth will emerge with enough lead-time to acquire real estate, secure permits, and construct a facility. If they move too slowly, they miss the opportunity and lose ground to competitors.
As a result, cloud providers, in many cases, are choosing to focus on service delivery and other priorities over new data center builds. They are willing to pay a small price premium — the margins colos add to their services minus the lower cost structure colos may be able to achieve through their scale and efficiency — for the speed-to-market and agility colos can provide.
Cloud providers who partner with a colo can be more responsive to shifts in demand by limiting the capital investment required to service a particular market. If demand doesn’t grow as expected, the cloud provider can simply scale back its footprint in a facility without paying the price of low utilization. And, colos take complex decisions about data center location, in which access to large groups of users must be balanced with financial and reliability considerations, out of the hands of cloud providers.
These considerations create a strong business case for cloud providers to use colocation to expand their reach and fill gaps in their own data center networks. This trend is not limited to smaller players with limited scale and expertise. Even large cloud providers with significant scale and expertise in the area of data center development are recognizing the advantages of partnering with a colocation provider to quickly reach new markets and minimize latency.
That’s good news for colo providers who are operating in an increasingly competitive environment. According to Structure Research, the wholesale market for colocation services is already growing faster than retail. But the risk of underutilized capacity and missed opportunity that cloud providers are mitigating by partnering with colocation providers is being assumed by their partner. To capitalize on this trend while minimizing the associated risk, colos must accurately project future demand to identify the best locations for future sites and develop facilities that cost-effectively adapt to future, hard-to-predict changes in demand.
In 2017, Vertiv surveyed U.S. enterprise data center managers to determine what they look for in a colocation provider. Three factors stood out:
- Quickly scale capacity when required
- Provide edge connectivity
- Offer price transparency
These are valid issues that colos must continue to address, but, if we had conducted a similar survey with cloud providers, I suspect a new response would have risen to the top: capacity in the exact location I need it. To continue to drive down latency and ensure customer satisfaction, cloud providers must expand in the precise locations where their customer base is growing. Often, they don’t know where they will need capacity until demand has already emerged. Then, they need to move fast and a colo who has already secured real estate and navigated the permitting process and perhaps even constructed a shell provides a fast path to bringing capacity online where it is needed.
In today’s dynamic market, colos have to do more than demonstrate the ability to develop and operate data centers efficiently and reliably. They must be able to determine the right location for those data centers to meet capacity needs before they fully emerge. Projecting future capacity demands requires a combination of art and science and has become an important differentiator for colo providers. They have become adept at identifying locations with favorable tax incentives and available electrical power, and where the user base — both retail and wholesale — is growing. By being in the right location at the right time, they make themselves more attractive to cloud providers seeking a partner to help them manage growth.
Identifying the right location for expansion is half of the challenge; the other half is developing facilities that can adapt to changes in capacity and density quickly and cost effectively.
We are already seeing changes in data center building patterns designed to increase flexibility. Fewer organizations are embarking on large facilities with rigid multi-year plans. They have learned that designs are often antiquated six months into a multi-year plan. Building smaller facilities that can scale modularly as needs — and designs —change is more cost-effective in the long run. It also allows resources to be developed closer to their customers.
Building smaller does compromise some of the economies of scale that come with larger developments. Some costs, such as the BMS control platform and fire controller, don’t scale with facility size the way other costs do. However, the increase in initial costs can be offset by a corresponding decrease in operating costs. Full utilization of the facility will be achieved much faster, saving the colo from having to support unused capacity while the market develops or new customers are recruited and brought on board.
Figure 2. Floor plan for an approximately 2-MW facility with more than 280 racks at an average density of 6 kW.
In addition, sound planning during the initial phase of the development can reduce costs in future stages. For example, Figure 2 shows the floor plan for an approximately 2 MW data center that features three UPS units in a catcher bus architecture and redundant cooling. When developing the second stage of this project, if it is required, the potential exists to leverage the redundancy in the first module to reduce infrastructure costs in the second.
Another benefit of building smaller data centers is the ability to find appropriate real estate in the right locations. It is much easier to find a small site in a prime location than it is to find a plot for a larger facility in the same location. Modular growth can still be achieved by finding other smaller sites near the first.
Potential changes in rack density also need to be considered as colos expand their network with an eye on the growing cloud market. Trends such as virtual reality, machine learning, and autonomous vehicles may require higher density racks than the 10 kW design spec many colos currently build to. But, like emerging capacity, it is difficult to predict exactly when some of these trends will impact the market. Build for density too early and you have the same problem as building out capacity before demand exists: underutilized assets.
Figure 3 shows a more than 10 MW data center designed to support about 300 high-density racks. This data center provides five times the capacity as the one in Figure 2 with about the same number of racks, but requires a much different floor plan.
Figure 3. Floor plan for a 10-MW facility supporting about 300 high-density racks. This facility is able to pack more capacity in a smaller space than Figure 2, but requires a dramatically different layout.
Admittedly, this is a somewhat extreme example of high density but it does illustrate the challenge of adapting to density. Accommodating higher densities is not as simple as adopting a modular approach to development — the amount of floor space required for supporting infrastructure can be much greater in a high-density data center. However, there are things that can be done from a provisioning standpoint to enable certain sections of a facility to support higher density in the future.
Indirect evaporative cooling, for example, will be unable to meet the demands of 20 kW racks, but moving to direct evaporative cooling with a chilled water tower, and provisioning the data center with cooling loops under the racks that can be supplemented with chip- or rack-based cooling at the appropriate time, creates the foundation to support increased density as demand emerges.
On the power side, bus bars can be sized for higher densities on select rows, branch circuits can be added, and scalability of the power modules can be planned for during the initial development. These add costs to the development and a colo would need to have some confidence that demand for higher density racks will materialize, but it is much more cost effective to provision for density during the initial development than to make the required changes after the demand emerges.
It seems likely that cloud providers will increasingly rely on colos to meet shifting capacity requirements, particularly in emerging markets. This represents a business opportunity for colos, but comes with risk. Colos could be left with underutilized assets if projections don’t pan out or demand shifts in a way a particular facility can’t support. Colos can mitigate this risk by developing modular facilities that can be easily scaled in terms of both capacity and density