It sounds like you could compete with Dell and that you should start a company. Maybe then you realise that 1kUS$ isn't that much for a system.
The saving is in that he isn't starting a company. So he has no costs for inventory, advertising, shipping, distributor discounts, etc. If he is building a dozen systems he can't win on cost over the hardware lifetime, but if he is building hundreds, he says he needs that many, he probably can shave quite a bit of the cost. But how much hardware is going on those machines, to drive the cost that high? Dell sells a reasonable office machine for just under $600, without massive discounts. Companies like eMachines go lower than that, and have similar performance. There is something driving up the cost we haven't been told.
The next obvious question is how much of the cost is software, and how much of that (possibly including the OS) is required cost? The old "easier if they're all the same" argument is usually made by a salesman or lazy purchasing agent, and often doesn't match reality. Data entry jobs which are poking numbers into web forms or spreadsheets don't require proprietary software. That doesn't mean that there may not be some need for commercial software, just that there's a lot of tasks in most enterprises which don't. And the "retraining cost" FUD is just that, people doing data entry, or any activity where the browser is the computer, need to learn login and start application from a menu or icon. Just like Windows. And free software will read/write most proprietary formats, so the need for a proprietary data format doesn't mean proprietary software is necessarily needed. One size does not fit all, there is probably room for saving in software, too.
This might even be a case for thin clients and a few servers, and get the cost way down, not enough information to guess, but a possible large saving. The problem is convincing management that the best approach is finding the most cost effective solution, not in finding the best price on the "way we always did it."