By Hector Diaz -- The construction of multi-megawatt hyperscale data centers by colocation companies has been accelerating globally. A shift in the construction techniques they use has allowed them to shorten their time to market and reduce construction costs. In a nutshell, the shift has been from stick-built construction, where everything is assembled mostly or entirely on-site, to semi-prefabricated construction, where major pre-assembled modules are delivered to the construction site.
Time to market is paramount for colocation companies. Many are organized as Real Estate Investment Trusts (REITs) and make no money until tenants start deploying equipment racks to new data centers. And results of the recent push to shrink time to market have been impressive. CyrusOne, for example, claims to have set construction speed records at their Sterling II data center in Northern Virginia. The company said it completed a 30MW buildout in 180 days – from breaking ground to commissioning. It credited the accomplishment to its use of “massively modular” power distribution and chiller units.
This column’s author, Hector Diaz, senior partner of iDiaz Advisors and a Data Center Institute board member, will be co-hosting a workshop on colocation strategies and techniques at Data Center World in Phoenix on Tuesday, March 19. To sign up, go here and join more than 1,600 other data center professionals at the show.
Modular power and cooling units are typically assembled in factories run by system integrators. The modules can be assembled simultaneously with construction of the concrete data center shell and delivered on-site when needed. This concurrent fabrication results in:
- Shorter time to market through parallel construction and fabrication paths
- Increased reliability of prefabricated modules, as quality control can be tighter in a manufacturing environment than in a field construction environment
- Optimized deployment of capital, as capacity can be added to match projected data center occupancy
Capacity is typically added in well-defined increments. Today, those tend to be somewhere in the 2MW to 2.5MW range. This is all driven by economies of scale. If you look at the dollars-per-kW ratio for a generator set, the best “bang for the buck” is somewhere in the stated range. The dollar amount per kW will be higher if you deploy in smaller chunks. If you want to save space in your equipment yard and have fewer but larger generators, the dollars-per-kW goes up as well, plus, you are now dealing with special-order items with longer lead times. Economies of scale apply to all the other major components for the MEP (mechanical, electrical, and plumbing) fit-out of a new data center.
I have talked with top-tier colocation companies that claim their construction cost for a concurrently maintainable multi-megawatt data center is $6.5 million to $7 million per 1MW. There are large data center providers that have not achieved these record-breaking efficiencies, citing per-MW construction costs in the $7 million to $10 million range – even while using pre-fabricated power and cooling modules. That’s still much lower than the $12 million to $17 million a company pays for a typical under-1MW enterprise “stick build” data center.
As a result, I generally advise enterprise IT groups against building data centers unless they are looking at IT loads greater than 2MW. Surprisingly, many IT groups don’t understand the true cost of building a data center, which includes eventual depreciation of all the capital sunk in construction, utilities, and facilities maintenance – costs typically paid by the real estate and facilities group, not by the CIO’s organization.
This article is part of a series of columns written by members of the Data Center Institute advisory board, a think tank of distinguished data center professionals who publish research, whitepapers, and speak at industry events offering insight and forecasting on important data center trends.