Why South Africa’s Power Prices Keep Rising?

South Africa already has the legal tools to test whether a cost belongs in the electricity tariff, meaning that the historical failure has been using them.

Municipal tariffs went up again on 1 July 2026. A typical Johannesburg prepaid household now pays about 413 c/kWh including VAT, whereas a ferrochrome smelter on a negotiated pricing agreement pays 62 c/kWh from 1 June 2026. That rate had been 87.74 c from 1 January, establishing a stark political gap between two prices for the exact same electron. Since 2007 the standard tariff has risen 937% against consumer-price inflation of 155%. Had the 2007 price of about 19.6 c/kWh only tracked inflation, it would have sat near 51 c by 2025; Instead, the live NERSA standard average, from 1 April 2026 stands at 240.28 c/kWh.

Our analysis traces this path across three phases.

First, capital. Between 2008 and 2011 tariffs rose about 25% a year to finance Medupi and Kusile, before the 2012 regulatory period was re-opened to 16% to cover project-execution risk. The Eskom board revised cost-to-complete to R145bn for Medupi and R161.4bn for Kusile, about R306bn together. Against the first six-unit Medupi approval of R92bn in 2006 and Kusile’s R77.8bn in 2007, this final figure stands about 80% above what had been budgeted.

Then the regulatory clearing account put past under-recoveries back into later tariffs: 3.32 c/kWh in 2015, 4.70 c in 2016, 8.17 c in 2022.

Then diesel. As energy availability fell from 84.9% in FY2008 to 54.6% in FY2024, the combined diesel bill for Eskom and its suppliers peaked at R33.9bn in a single year, with Eskom-owned diesel costing 14.51 c/kWh during that peak period.

Every cent of the standard tariff was approved by the regulator. Because approval merely records who pays, it does not constitute a finding that the underlying cost was efficient.

Our evaluation of this year’s determination shows what a rigorous cost test already takes out. Eskom asked for R30.8bn more for operating expenditure, carbon tax and arrear-debt recovery, but NERSA allowed none of it. This intervention kept 15.99 c/kWh out of the tariff, comprising 8.49 c on opex (R16,353m), 2.87 c on carbon tax (R5,534m), and 4.63 c on arrears (R8,914m).

Importantly, NERSA did not refuse Eskom’s R17.6bn OCGT variance in the 2022/23 RCA, choosing to allow that variance in full. The net RCA of minus R232m represents a separate allocation line payable to customers, instead of a diesel haircut.

Fleet availability has since recovered to 67.79% year-to-date, while the FY2026 diesel allowance is set at R6.4bn. We treat these as generation-recovery facts, instead of the product of a NERSA refusal.

Municipal cost-of-supply work has also shifted. For the 2026/27 cycle, NERSA processed 176 licensed-distributor applications with published cost-of-supply studies, and subsequently published the formal reasons for decision on 30 June 2026.

Our results isolate exactly what is still embedded within this year’s Eskom tariff. Of the 199.73 c/kWh allowed, 12.03 c is identifiable allowed-imprudence: 3.32 c of diesel and 8.71 c of past clearing-account balances. Another 32.13 c is unclassified capital charges (16.20 c depreciation and 15.93 c return) on a regulatory asset base of about R995.5bn that public data cannot split into prudent plant and overrun. At the municipal layer, Johannesburg’s 29.92% losses add 74.01 c/kWh of physical excess on its own bulk purchases, whereas Cape Town, buying on the same wholesale path, adds only 2.22 c because it holds losses at 10.81%, near the 10% target. This operational variance provides definitive proof that the pricing gap is a function of local management instead of physics. 

The Electricity Regulation Amendment Act 38 of 2024 (in force 1 January 2025) changes market architecture: unbundling, wheeling, a wholesale platform. However, it does not repeal the municipal section 229 surcharge, it does not extract Medupi and Kusile from the asset base and it does not end the clearing account. Complying with the new Act re-labels the customer bill, where enforcing the old cost test actively shrinks it. Cabinet’s pricing policy reflects this identical distinction: itemised bills show a cost, but they do not remove it.

We define three necessary actions going forward:

1. Apply the statutory prudency test to February’s R12bn asset-base add-back, which is already inside the live 8.76% increase from 1 April 2026, and publish the split of the 32.13 c unclassified capital block.
2. Refuse every municipal tariff application that lacks a credible cost-of-supply study, thereby holding the line NERSA set when it published the 2026/27 municipal decisions on 30 June.
3. The National Treasury must strictly ring-fence revenue on every special industrial price. The 62 c ferrochrome rate works because it is not recovered from standard customers, insofar as a discount that is not ring-fenced merely reallocates the bill. On the ferrochrome book, this shortfall is about R18bn a year at standard prices, proving that only taking cost out cuts the price.
    

A Johannesburg household and a ferrochrome smelter buy the same electron. The gap on the bill is a chain of regulatory approvals instead of an engineering variable. The Electricity Regulation Act has given the regulator the cost test for two decades: use it, on record, this year.