The African National Congress (ANC) faces a virtual wipeout in the next local government elections (LGE) in eThekwini (Durban).

This is according to polling* conducted by the Social Research Foundation (SRF), in conjunction with The Common Sense.

The polling, which was conducted in late February and early March, surveyed people nationally and in three metros, Johannesburg, Tshwane, and eThekwini, to determine the political state of play in the country.

People were asked who they would vote for if a municipal election was to be held that day.

Only 8% of respondents in eThekwini said they would vote for the ANC. This is a remarkable reverse for the party, which has since the establishment of the current system of local government in 2000 always been the biggest party in eThekwini, never winning below 40% of the vote. In 2011 it had secured over 60% of the vote in the metro.

Forty-four percent of those surveyed in the SRF/The Common Sense poll said they would vote for the uMkhonto weSizwe Party, with 28% saying they would vote for the Democratic Alliance. Eighteen percent said they would vote for the Inkatha Freedom Party.

No other party had the support of more than 4% of those polled, which was the poll’s margin of error in eThekwini.

*The Social Research Foundation Q1 2026 Market Survey was commissioned bythe Social Research Foundation supported by  The Common Sense and conducted by Victory Research using a nationally representative telephonic CATI survey of registered voters (N=2 222), with metro samples upsized to over 500 respondents each in Johannesburg (n=503), Tshwane (n=510), and eThekwini (n=503). Fieldwork was conducted between 16 February 2026 and 6 March 2026 using a single-frame random digit dialling sampling design that draws from all possible South African mobile numbers, ensuring equal probability of selection and near-universal coverage given SIM penetration above 250%, more than 90% adult phone ownership, and mobile network coverage of 99.8% of the population. Respondents were screened to include registered voters only, and turnout modelling assigned each respondent a probability of voting based on likelihood indicators, with the primary model assuming turnout of 56.0%. Data were weighted to ensure the national sample reflects the demographic profile of the registered voter population across language, age, race, gender, location (urban/rural), education, and income, while metro samples were weighted to the demographic composition of voters in each metro. Results are reported at a 95% confidence level with a design effect (DEFF) of 1.762, producing margins of error of 2.1% nationally and 4.4% for each metro sample.