Background Comprehensive and representative data on resource use is critical for health policy decision making but often lacking for HIV infection. Privacy preserving probabilistic record linkage of claim and cohort study data may overcome these limitations. Methods Encrypted dates of birth, gender, study center and antiretroviral therapy (ART) of the Swiss HIV Cohort Study (SHCS) from 2012 and 2013 were linked by privacy preserving probabilistic record linkage with claim data from the largest health insurer covering 15% of the Swiss residential population. We modeled predictors for mean annual costs adjusting for censoring and grouped patients by cluster analysis into 3 risk groups for resource use. Results The matched subsample of 1196 patients from 9326 SHCS and 2355 claim records was representative for all SHCS patients on ART. Corrected mean total cost (SE) in 2012 and 2013 were USD 30'462 (582) and USD 30'965 (629) and mainly accrued in ambulatory care for ART (70% of mean costs). The low risk group for resource use had mean annual cost of USD 26'772 (536) and USD 26'132 (589) in 2012 and 2013. In the moderate and high risk groups annual costs for 2012 and 2013 were higher by USD 3'526 [1'907; 5'144] (13%) and 4'327 [2'662; 5'992] (17%) and USD 14'026 [8'763; 19'289 (52%) and 13'567[8'844; 18'288] (52%), respectively. Conclusions In a representative subsample of patients from linkage of SHCS and claim data, ART was the major cost factor but patient profiling allowed to identify factors related to higher resource use.