If you gave me $500 million today, I could spend it down at a rate of $25 million a year, for 30 years (the age of the data center that's now being replaced), assuming that I kept the unspent funds in a form that returned me annual after-tax yield of 3%…with $2,000 per work day left over for coffee and doughnuts in the break room.
I believe my answer might therefore be, "I wouldn't buy a data center at all."
Wait, it gets better: the same story says that the federal agency in question "will need closer to $800 million to fund a new IT infrastructure, including the new data center-the physical building, power and cooling infrastructure, IT hardware, and systems applications. (This is addition to a $72 million backup facility currently under construction…)"
So, we're really talking about something much closer to a billion dollars — much more, if we bring back to present value all the costs of planning the project, spending the money, and operating the facility once it's built. A gigabuck is just the ante. That means the equivalent stream of payments is at least $50 million a year.
The agency in question has 62,000 employees. For $50 million a year, I could buy that agency:
- 62,000 Force.com User Licenses (Enterprise Edition) @ $50/user/month = $37.2M/year
- 1,000 AWS SimpleDB machines @ $0.14/hour x 24×365 hours/year = $1.2M/year
- SimpleDB data storage/transfer @ 1,000 GB/month in/stored/out per SDB machine = $6.2M/year
- 62,000 Google Apps Premier accounts @ $50/user/year = $3.1M/year
That's only $47.8M/year, and I'm vastly over-buying on the Force.com licenses: not every employee needs the all-day, every-day access of the per-user/per-month pricing plan. That $37M/year could likely be cut in half. That would get us down to something more like $30M/year for email, personal productivity, collaboration tools, database management and custom application development and deployment. No extra costs for data backup, air conditioning, or washing the data center's windows.
I'm spitballing my estimate of additional computation and storage that might usefully be bought from Amazon Web Services: I've budgeted for 1,000 virtual machines that are each defined as equivalent to a 1.7 GHz Xeon processor, each of them handling 1,000 GBytes of data with 100% monthly turnover. That's one million x 10243 bytes to represent a U.S. labor force of about 153 million people, or roughly 7 MBytes per worker. I don't know what your employment history looks like, but I believe that mine can easily be represented within that allowance.
So, would you run a major federal agency's IT entirely in the cloud? If you don't, it's not because you can do it more cheaply yourself. The cloud's economies are real — and you don't spend the money up front.