Friday, May 18, 2012

Facebook IPO Won't Be Good for Data Centers

Facebook's "Initial Public Offering"  will be good for Facebook, many of its employees, the banks underwriting the offering and many other social networking firms, as their valuations get a boost from Facebook's valuation. At some level, it will be good for firms selling products that many instant millionaires will be buying.


But the IPO probably won't be good for several data centers that currently sell services to Facebook, as Facebook is transitioning its hosting to its own facilities.


Facebook currently spends more than $70 million a year leasing “plug-and-play” data center space from Digital Realty Trust, DuPont Fabros Technology, CoreSite Realty and Fortune Data Centers.  


As a result of its new liquidity, Facebook will be able to migrate from  third-party space to its own facilities. 


Facebook is said to represent between four percent and 20 percent of total annual revenue for these providers, who might start to see the migrations in a couple to several years. 


Facebook is said to spend $30.1 million a year on four facilities it leases from Digital Realty in Silicon Valley and northern Virginia. That represents 4.6 percent of Digital Realty’s annual revenue. Digital Realty has leases with Facebook with an average of 86 months (about seven years) remaining, according to Data Center Knowledge. 


Facebook is the second-largest  tenant for DuPont Fabros, with annualized base rent of about $22 million, which is at least 20 percent of DFT’s annualized base rent. Facebook’s leases, primarily in northern Virginia, have an average remaining term of 6.5 years. (77 months).


Facebook is the largest single customer for CoreSite, paying $11.5 million in annualized rent for 74,112 square feet of space in three facilities, including an entire building in Santa Clara. Facebook represents 12.6 percent of CoreSite’s lease revenue. The Santa Clara lease expires in March 2016.

No comments:

Costs of Creating Machine Learning Models is Up Sharply

With the caveat that we must be careful about making linear extrapolations into the future, training costs of state-of-the-art AI models hav...