To further shine a light on the cloud/hyperscale market, I’d like to piggy-back off my article “The Industrial Age of Hyperscale: The Three-Legged Stool to Getting to Hyperscale TCO,” by expanding on the current direction of the wholesale data center market.
That Was Then, This Is Now
Over the years, the wholesale market has catered to the enterprise user, which is a completely different buyer than the cloud providers. Enterprise users would cling to the design standards of uptime, utilizing continuous generators and upsizing of equipment to support load.
These design features and branding — complemented by a nice office space — had a significant impact on the final cost per MW. Since the enterprise generation is moving on, with 87% of the users now in the wholesale and cloud sectors, that market fizzled. Now, the scramble is to support hyperscalers within wholesale facilities that were designed for enterprise users. The two buyers clash.
For wholesale providers to become competitive, they need to get their price points around $122 to $125 per kW. This simply cannot be done with the prototype models they have built in the past. A fully built-out hyperscale campus typically comes in at $6 million per MW (not including the utility).
Here are a few of the changes wholesalers should consider when leasing to the hyperscale market.
Metal Buildings- The precast buildings designed for the wholesale market do not apply. The average cost for shell-and-core buildouts is approximately $123 per square foot for precast, whereas metal buildings are $74 per square foot. And while the overall impact of shell/core versus the mechanical, electrical, and plumbing (MEP) cost is minimal in the overall cost per MW, you need to take the mindset of milking a mouse.
Distributed Redundant Design- The philosophy originally identified as a “catcher block” design is the most scalable; it also has the lowest first cost. This still holds true, but at the rate that the hyperscalers are growing, you should hedge your bet on using capacity. Distributed redundant (not static switch) is the most cost-effective at total buildout. In some cases, it’s more about changing your mindset toward the hyperscale mindset.
Think Campus Size- Instead of thinking of building 40-MW facilities, think campus size of 180 MW-plus. This may push your land purchases farther out, away from the Ashburn areas. Many campuses are located in remote areas; this has both pros and cons.
Speed-to-Market and Build Program- Cloud providers are rolling out applications and programs faster than ever before. Instead of building out 2-MW suites of capacity, consider 28-MW chunks of builds. This fits in with the hyperscale program and current mindset. The hyperscaler doesn’t have the mindset of “stranded capacity.” You may be thinking this is just bad business at high risk, but examine all your options first.
Cooling- Consider direct evaporative or indirect evaporative cooling to get power usage effectiveness ratings (PUEs) under 1.2 at full load.
The above suggestions may seem outlandish to some, but if you look back only five years, the industry wouldn’t have believed the hyperscale campuses of today would be the size and volume they are now.
Question for the audience: There are a lot of investors who want to get into data center acquisitions for their portfolio — just ask Miles Loo. Once the hyperscalers have taken all of the tax incentives and depreciations, do you think that there will be a market to acquire hyperscale campuses? Let me know what you think; email me at firstname.lastname@example.org.