Background Non-facility-based antiretroviral therapy (ART) delivery for people with stable HIV might increase sustainable ART coverage in low-income and middle-income countries. Within the HPTN 071 (PopART) trial, two interventions, home-based delivery (HBD) and adherence clubs (AC), which included groups of 15–30 participants who met at a communal venue, were compared with standard of care (SoC). In this trial we looked at the effectiveness and feasibility of these alternative models of care. Specifically, this trial aimed to assess whether these models of care had similar virological suppression to that of SoC 12 months after enrolment. Methods This was a three-arm, cluster-randomised, non-inferiority trial, done in two urban communities in Lusaka, Zambia included in the HPTN 071 trial. The two communities were split into zones, which were randomly assigned (1:1:1) to the three treatment strategies: 35 zones to the SoC group, 35 zones to the HBD group, and 34 zones to the AC group. ART and adherence support were delivered once every 3 months at home for the HBD group, in groups of 15–30 people in the AC group, or in the clinic for the SoC group. Adults with HIV who were receiving first-line ART for at least 6 months, virally suppressed using national HIV guidelines in the last 12 months, had no other health conditions requiring the clinicians attention, live in the study catchment area, and provided written informed consent, were eligible for inclusion. The primary endpoint was viral suppression at 12 months (with a 6 month final measurement window [ie, 9–15 months]), defined as less than 1000 HIV RNA copies per mL, with a non-inferiority margin of 5%. Findings Between May 5 and Dec 19, 2017, 9900 individuals were screened for inclusion, of whom 2489 (25·1%) participants were enrolled into the trial: 781 (31%) in the SoC group, 852 (34%) in the HBD group, and 856 (34%) in the AC group. A higher proportion of participants had viral load measurements in the primary outcome window in the HBD (581 [61%]of 852 participants) and AC (485 [57%] of 856 participants) groups than in the SoC (390 [50%] of 781 patients) group (p=0·0021). Of the 1096 missing observations, 152 (13·8%) were attributable to either deaths (25 [16%] participants), relocations (37 [24%] participants), or lost to follow-up (90 [59%]); 690 (63·0%) participants had viral load results outside the window period; and 254 (23·2%) did not have a viral load result. The prevalence of viral suppression was estimated to be 98·3% (95% CI 96·6 to 99·7) in the SoC group, 98·7% (97·5 to 99·6) in the HBD group, and 99·2% (98·4 to 99·8) in the AC group. This gave an estimated risk difference of 0·3% (95% CI −1·5 to 2·4) for the HBD group compared with the SoC group and 0·9% (−0·8 to 2·8) for the AC group compared with the SoC group. There was strong evidence (p<0·0001) that both community ART models were non-inferior to the SoC group (p<0·0001).