When drafting an SLA, it's more important to agree with the customer what they expect (and afford) vs what your willing to support within the constraints of the equipment and budget you have.
For example: a single non-clustered server is not suitable for a customer that wants 99.999% uptime and 24 hour on-call support and 1 hour "Return to operations" on a major failure. It's not technical reasonable to accept that and the customer needs to understand that.
Yes, Windows 2003 Server is reliable and can perform very nicely. Brand name servers come with proven reliability and rock solid warranties. Regular monitoring and TLC on a server can keep it going for many years.
You need to "hope for the best, but plan for the worst".
You'll also need to accurately calculate your availability statistics and have the calculation agreed with the customer (1 hour downtime at 2am is a different "cost" to 11am on a Tuesday).
You'll need to incorporate all the additional equipment that is required to keep a server alive (networking, switches, firewalls, operator time, backups).
Finally, you'll need to test your contingency plans, and keep your infrastructure flexible so you can solve the fault in several different ways.