These are just some ideas for things you might want to consider.
This is capacity planning, no different because it is Cognos. First thing is to decide what to measure - CPU utilization, disk I/O, RAM utilization, etc. Then baseline your current system. It helps if your organization has some monitoring tools here, as you probably want to sample it over time to get averages. Then rerun your measurements at regular intervals, maybe make a graph showing the changes over time. Based off of this you can guess at your future requirements and plan to adjust accordingly.
1900 reports a week over a 5-day 40-hour workweek would be a little less than one report a minute on average. Whether this is a lot or a little obviously depends on the particulars of your data and hardware, which affect how long each report takes and resources used, as well as whether the report runs are clustered around particular times of the day or month. If your reports are predictable and repeatable for multiple users, you might look at running some at off-hours on a schedule and saving the output version, or bursting them. There are things you can do database-wise to speed things up, or use cubes if appropriate. There are too many variables to give very specific advice, it depends on your situation.