Introduction: Early empirical broad-spectrum antimicrobial therapy is recommended for patients with severe infections, including sepsis. β-lactam/β-lactamase inhibitor combinations or carbapenems are often used to ensure coverage of likely pathogens. Piperacillin/tazobactam is proposed as a carbapenem-sparing agent to reduce the incidence of multidrug-resistant bacteria and superinfections. In the recently published MERINO trial, increased mortality from piperacillin/tazobactam was suggested in patients with bacteraemia with resistant Escherichia coli or Klebsiella species. Whether these findings also apply to empirical piperacillin/tazobactam in patients with other severe infections, including sepsis, is unknown. We aim to assess the benefits and harms of empirical and definitive piperacillin/tazobactam vs carbapenems for patients with severe bacterial infections. Methods and analysis: This protocol has been prepared according to the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols statement, the Cochrane Handbook and the Grading of Recommendations, Assessment, Development, and Evaluation approach. We will include randomised clinical trials assessing piperacillin/tazobactam vs carbapenems in patients with severe bacterial infections of any origin. The primary outcome will be all-cause short-term mortality ≤ 90 days. Secondary outcomes will include all-cause long-term mortality > 90 days, adverse events, quality of life, use of life support, secondary infections, antibiotic resistance, and length of stay. We will conduct meta-analyses, including pre-planned subgroup and sensitivity analyses for all assessed outcomes. The risk of random errors in the meta-analyses will be assessed by trial sequential analysis.