I did some math yesterday on the cost of virtualizing our "Internet infrastructure" vs. leaving it as it is.
Right now we run five servers -- the household file server, two production web/email/DNS boxes, a development web box, and an NFS host to catch disk-to-disk backups. One machine is a P4, the others are PIIIs. The combined power load of these five machines, plus switching, firewall, KVM, etc. is 450 Watts.
I looked at the ROI of retiring all of the PIIIs (the three web boxes and the NFS host) and recreating them as virtual machines running on a single new machine. The most economical way to get the platform I wanted was assemble it from parts (my old friend JDR Microdevices and my new acquaintance Provantage to the rescue). A 2.4GHz Core2 quad (isn't that a contradiction?) motherboard, small server case, 4GB of RAM, a pair of 500GB Barracuda drives, a "RocketRAID" SATA RAID controller and other miscellaneous giblets cost $1,157.32 (including shipping). The four PIII servers draw 236 Watts. Based on the power draw of our 2GHz P4 file server (109W, with four internal drives and an external) I estimated that the new machine would draw 80W.
Doing the math, 236W - 80W = 156W saved by virtualizing four boxes onto one. At that rate, that will save 1,367kWhr (kilo Watt hours) per year. Our electric rate is just under 19 cents/kWhr, yielding a savings of $256.27 per year. Verses $1,157.32 for a new server, that comes out to a 4.5 year ROI.
Of course, the above analysis only looks at power savings. If you factor in avoidance of upgrades for individual machines, the picture is even better. For example, one of the production web boxes has redundant drives, the other doesn't. Upgrading the non-redundant machine would cost roughly $275 (two 400GB SATA drives and a SATA RAID card, plus shipping). Also, that machine is the busiest of the production web boxes -- it's reasonable to think that a 700MHz PIII won't have enough 'oomph' (especially if one of the sites it hosts continues to surge in popularity) in the next twelve months or so. Even if I took the cheap route and picked up another $175 'used Dell Optiplex' special at the Pratt Surplus Store (which is where my flock of Dell PIIIs came from), that's still $175... Avoiding those two upgrades saves an additional $350, making the break-even on the new hardware come at two years one month.
So, there you are ... that's the business case for virtualization, writ small. Replicate this many times over across a large datacenter, and now you know why everyone is buzzing about this.