The spectacular growth of the public cloud continues and journalists constantly report that the number of enterprise data centers up for sale is greater than ever. So, are we really witnessing the death of the data center? Perhaps not and here’s why.
The reports of my death are greatly exaggerated
As companies move workloads into public clouds, they may seek to sell their enterprise data centers and the spike in the sale-leaseback market is a good indicator that even older data centers are still valuable. The buyers in this market are colocation businesses and their actions underscore what leading industry analysts are predicting, e.g. we need more data centers and aren’t building enough of them to satisfy current needs. Consider these market growth forecasts to support this assertion:
● IDC predicts global public cloud revenue growing from $180bn in 2018 to $277bn by 2021, or $97bn.
● Forrester predicts global public cloud revenue growing 22 percent CAGR from $178bn in 2018 to $323bn by 2021, or $145bn.
● Gartner’s cloud prediction is $176bn in 2018 to $278bn by 2021, or $102bn.
The cloud, whether public or private, is housed in data centers and these facilities must continue to evolve if they are going to support the aforementioned predictions. It’s also important to note that evolution, in and of itself, does not ensure survival for all technologies. For data centers to properly evolve so that all types of organizations can benefit, they must also become more cost-effective and efficient to operate, be more flexible to support higher power configurations (such as hyper-converged), and scale faster.
There is no disputing that the adoption of public and private clouds as well as colocation hosting, are all growing at enormous rates - especially when it comes to large public cloud providers such as AWS and Microsoft. Case-in-point: AWS revenues have grown at a CAGR of over 60 percent since 2010, with the upward growth continuing through 2018. AWS revenues increased by over 48 percent through the first half of last year and by 49 percent year-over-year in Q2-2018. Revenue surged from a Q3-2017 total of $3.66bn to $6.11bn in Q2-2018. Multiply that by four, and you get an annualized run rate of $24.4bn, or $10bn more than the prior year.
According to Microsoft’s FY18 Q4 report, server products and cloud services revenue increased $4.5bn or 21 percent, driven by Azure and server products licensed on-premises revenue growth. Azure revenue growth was 91 percent, due to higher infrastructure-as-a-service and platform-as-a-service consumption-based and per-user-based services. Server products licensed on-premises revenue increased five percent, mainly due to a higher mix of premium licenses for Windows Server and Microsoft SQL Server. In addition, Microsoft’s operating expenses increased by $683m or seven percent, driven by investments in commercial sales capacity and cloud engineering. Any way you slice it, Amazon, Microsoft and other providers are seeing enormous public cloud growth.
What does all this enormous cloud growth mean for infrastructure providers? One prominent analyst predicts the “death” of the enterprise-owned data center, as workloads migrate into hyperscale data centers owned by the cloud providers - but they can’t build enough capacity fast enough. Also, not all workloads are meant for a public cloud of shared resources, and some enterprises have investments in IT hardware that are strategically valuable for them to own and manage. These organizations just need to rid themselves of the cost and burden of managing their own data center facility - the prominent analyst now points to these facilities as competitive liabilities. So the plausible answer to support the growth of cloud lies with the colocation market.
Privately hosted clouds represent a larger revenue amount than their public cloud counterparts, and the private cloud resides, almost by definition, within enterprise or colocation data centers - not in the shared resources that reside in hyperscale facilities. 451 Research estimates that in 2018 private cloud revenue was about 43 percent greater than the public cloud revenue and growing only two percent less quickly. Also, while the hyperscale companies continue to build sites, they don’t have enough capacity to support this predicted growth on their own and have been increasingly turning to colocation providers for help.
This predictive model is great news to the colocation market and it's supplier OEMs that provide mission-critical products. But just as the hyperscale facilities need to adhere to the four best-practice principles (efficient, cost-effective to operate, more flexible to support higher or different power configurations, and scale faster), the same holds true for their colocation counterparts.
But how much capacity is needed if the analysts’ growth predictions hold true? Using a few “back-of-the-envelope” calculations, we can begin to see how much data center infrastructure capacity growth may be needed. To find out, let’s make a few basic assumptions:
● Assume a standard rack supplied at 8kW is used to support the servers & storage for hosting a cloud environment.
● Assume that each rack cabinet operating at that full load of 8kW, is able to deliver $45,000 in monthly recurring revenue.
Based on these assumptions, we can calculate the incremental cloud revenue growth and what it can mean for the infrastructure needed to support the expansion.
If we look at the median forecast and use Gartner’s incremental growth figures of $102bn over the 36 months between 2018 and 2021, and use the aforementioned assumptions, it translates to new incremental revenue of $2.83bn per month. And adding our assumption of $45k in MRC per rack cabinet, this translates to just over 62,888 new racks of equipment that will be needed each month. At 8kW per rack, this equates to 503,104 kW or 503.1MW of new data center infrastructure globally, each month. The infrastructure support can come from new facilities or retrofitted data centers, but it would be an incremental infrastructure necessity to some degree.
A common figure in the industry is that it takes an average of $8-$10m to build a megawatt of capacity. But let’s assume costs have been improving so, it only requires $8m in capital expenses per megawatt. This means it will require over $4bn in capital spending - each month - to build the required global capacity to support the forecasted revenue growth of the cloud.
It’s also worth noting IDC’s estimate that 60 percent of the global cloud capacity is built in the United States. So by extension, this means that approximately $2.42bn in CapEx is needed - each month - to add the incremental infrastructure required in the U.S. to support the growth of cloud systems. Is this real or are forecasters being over-exuberant?
Based on the reported data - it appears to be very real - since each analyst firm revised their 2018 estimates to be higher than their initial forecasts. How will the colocation industry be able to support this growth? Even if we take the lower estimate from IDC, it indicates that $2.3bn will be needed - each month - to build the incremental infrastructure to support the cloud growth in the U.S. However, if we can construct these facilities in a quick and cost-effective manner, the needed data center support can be realized. Yet how to do it?
Looking at hyperscale providers, one consistent method that is used by each is the prefabricated modular approach. Just like car manufacturers want their assembly plants to be efficient, cost-effective, and support high-quality processes, hyperscalers use modular designs in their data centers as it delivers those benefits of lower cost, higher efficiency, and better flexibility and scalability.
The Department of Energy’s National Energy Technology Lab (DOE-NETL) HPC data center build provides another solid example of a present-day, cost-effective facility and also gives us some necessary guidance toward achieving the goal of delivering hyperscale performance to a broader market. The DOE-NETL facility is a modular 1MW data center that can flexibly support up to 35kW per rack. The facility uses adiabatic evaporative cooling that delivers the required ASHRAE standard humidity and temperature ranges without the higher cost of a chilled water system and raised floor with CRAC units. This reduces the cost of the build to under $6m per megawatt, yet the data center has been averaging a PUE of 1.06 since it was commissioned back in 2012. The manager estimates the higher efficiency has delivered an operational cost savings of $450,000 per year. Even better, it was deployed on site in two weeks to help realize those savings faster.
Regardless of which forecast you use, a huge capacity of colocation data centers is needed over the next three years. In order to ensure the infrastructure growth is capable of meeting the analysts’ predicted requirements for all workload processing needs, future data center retrofits and new builds must be low cost, highly efficient, support higher power densities and have the ability to rapidly scale.
Modular designs and methods, such as used at the DOE-NETL and many hyperscale facilities, can deliver colocation data centers that are capable of supporting these requirements. And most importantly, these new facilities can be deployed quickly! So as you evaluate your next colocation provider, see how they stack up compared to a data center using modular methods.