We consider the capacity planning of telecommunication networks with linear investment costs and uncertain future traffic demands. Transmission capacities must be large enough to meet, with a high quality of service, the range of possible demands, after adequate routings of messages on the created network. We use the robust optimization methodology to balance the need for a given quality of service with the cost of investment. Our model assumes that the traffic for each individual demand fluctuates in an interval around a nominal value. We use a refined version of affine decision rules based on a concept of demand proximity to model the routings as affine functions of the demand realizations. We then give a probabilistic analysis assuming the random variables follow a triangular distribution. Finally, we perform numerical experiments on network instances from SNDlib and measure the quality of the solutions by simulation.