Whether antibiotic de-escalation reduces the risk of subsequent antibiotic resistance is uncertain. We sought to determine if beta-lactam (BL) antibiotic de-escalation is associated with decreased incidence of new Gram-negative resistance in hospitalized patients with sepsis. In a retrospective cohort study, patients with sepsis who were treated with at least 3 consecutive days of BL antibiotics, the first 2 days of which were with a broad-spectrum BL agent defined as a spectrum score (SS) of ≥7 were enrolled. Patients were grouped into three categories: (1) de-escalation of beta-lactam spectrum score (BLSS), (2) no change in BLSS, or (3) escalation of BLSS. The primary outcome was the isolation of a new drug-resistant Gram-negative bacteria from a clinical culture within 60 days of cohort entry. Fine-Gray proportional hazards regression modeling while accounting for in-hospital death as a competing risk was performed. Six hundred forty-four patients of 7742 (8.3%) patients developed new gram-negative resistance. The mean time to resistance was 23.7 days yielding an incidence rate of 1.85 (95% confidence interval [CI]: 1.71-2.00) per 1000 patient-days. The lowest incidence rate was observed in the de-escalated group 1.42 (95% CI: 1.16-1.68) per 1000 patient-days. Statistically significant reductions in the development of new gram-negative resistance were associated with BL de-escalation compared to no-change (hazards ratio (HR) 0.59 [95% CI: .48-.73]). De-escalation was associated with a decreased risk of new resistance development compared to no change. This represents the largest study to date showing the utility of de-escalation in the prevention of antimicrobial resistance.