Several valid responses are already posted. Let me add a few additional considerations for your requirements analysis, the most difficult issue for hardware sizing/capacity planning.
First of all, you need to identify the business solution you intend to run in the NW environment which means what components of NW stack you will using (like EP, XI, BW, etc.). You need to find the how the end user will access your system. In case an end user access outside of the LAN (say via WAN or dial-up) using https, it will have additional requirements for your hardware as you will need compression enabled for improved response time for WAN. Adding compression or SSL may add additional hardware if you based your sizing based on LAN.
Now for an example, if you have EP in your system, you need to understand how large (say in KB) an average portal page is. How it is connected to the backend system, if any. You need to find out the performance requirements in terms of a predefined “transaction (say an end user click and response)” or in terms of number of either logged-on users (may not be active always but logged on) or concurrent users (active users) or combination of them.
It will make your proper hardware planning challenging. One easy approach is after you define your test case, run some load test on your QA/Staging system. Use the load test data and find out the hardware requirements based on load test. If your requirement is larger than the staging system, you may have to take a staged approach. In that case you roll-out the system to a certain number of users and monitor the system and plan for additional users based on the monitored results.