You think bits are cheap? Think again.
I meant after the book has been formatted and edited. And the costs are more of a one time fee than a per unit cost as with real books.
I see you don't work with technology for a living, specifically IT. Let me illuminate you a bit, if possible.
Here's a non-comprehensive list of some of the costs associated with producing anything digitally, including ebooks, but (hopefully) excluding the overlap with dead-tree versions of books (remember, this is not comprehensive or perfect):
1. You need data center space. You either lease space in an existing facility or you own your own building, whichever is appropriate for your company's needs. Neither option is cheap. Owning your own building comes with extra expenses, such as property tax, zoning and inspection issues, facilities maintenance, insurance, and more.
2. You need electricity. To run the servers, storage devices, air conditioning, lights (so the admins can see what they're fixing), network equipment inside the data center, network equipment connecting the data center to your ISP(s). For a large publishing house such as Macmillan, the yearly bill for electricity in their data centers (they probably have more than one) is likely somewhere between 6 digits and 7 digits long (in front of the decimal), assuming US dollars.
3. You need people to keep the data center running for you. You're paying for them whether you're renting space or own it.
4a. You need system administrators to keep the computers, namely the servers and storage devices, running.
4b. You need network administrators to keep the networking gear running. In some cases, the system admin and network admin may be the same person or people but this doesn't scale terribly far so it won't last long.
5. The servers. Servers needed to mediate access to the network-based storage devices (think Netapp or Hitachi SAN). Even if the storage devices are inside the servers, rather than network (ethernet) or SAN (fiberchannel) based. Enterprise servers are not cheap; even for the low-end servers, you may spend a small fortune on enterprise server-class RAM, especially if you need the new 8GB 1266MHz stuff.
6a. The storage backend. This equipment, in the case of a large company such as Macmillan, will not come from any place like Newegg or Fry's. It's very likely that it's purchased from Netapp, Sun (nee, Oracle), HP, Dell, Hitachi, etc; basically, companies that sell stuff you can run a large business on 24x7 for years on end. This sort of storage is never cheap for anything you'd want to rely on, even for SATA drives; usually, it's based on SCSI or SAS disks. Although, the drives are really the cheap part compared with the infrastructure required to make them useful.
6b. You will need someone (or several someones) to administer the storage backend, both logically and physically. This may be the same person as the system or network admin, or it may not be because it can get extremely complex and consume a lot of time to manage it.
7a. You will be paying for throughput used between your servers and your ISP, definitely upstream and possibly downstream, generally by the megabit. Probably somewhere in the neighborhood of $10 and $30 per megabit, depending on too many variables to list here. Note: This is megabits,
not megabytes. This includes data transferred between your data centers that doesn't go over a private WAN link. Some of the non-WAN costs can be mitigated by using content distribution networks, such as Akamai, but then of course you're also paying Akamai for their services.
7b. The cost of private WAN links between data centers, if you have them. These are, however, almost always cheaper than paying by the megabit for traffic that transits the public Internet, but they're still a non-trivial cost.
8. Support contracts for the hardware.
9. Support contracts for the storage management, and other, software you will almost certainly be using for the petabytes or exabytes of storage space you're managing across all of your data centers.
10. Backups for all that data. You don't think these guys sit around with one copy of everything, do you? They don't. Granted, backups may be worked into the storage systems linked across data centers but there's still whatever software is being used to manage those backups.
11. Annual or bi-annual hardware upgrades to at least some gear; annual or bi-annual hardware replacement of outdated gear.
12. Costs associated with disposing of that outdated gear.
13. Periodic costs for software upgrades that fall outside the support contract. Also, costs for acquiring new software as becomes necessary.
Data storage may cost next to nothing for you, the average end-user/consumer, but it is
far from cheap for publishing businesses of the size we're discussing with respect to the iPad.