The System Operator’s new ten-year outlook says South Africa can get through the 2030 coal exit without new gas. Read the assumptions, and the cliff is still there. It’s three years further down the road, and it rests on promises nobody has kept yet.

On 30 October, the National Transmission Company of South Africa published its Medium-Term System Adequacy Outlook. For the first time, it looks ten years ahead, to 2036. The headline is good, and much of it is earned. Eskom’s fleet availability has recovered from 56% in 2023 to 66% in the year to March 2026. Load-shedding hasn’t been implemented since May 2025. Under the Reference Case, unserved energy stays below the 20 GWh adequacy limit from 2029 and falls to zero in 2031 and 2032.

A year ago, the same System Operator told a different story. Its 2026–2030 Outlook assumed 6 GW of gas-fired power by 2030. If that gas slipped, it warned, unserved energy would pass 4 TWh in 2030, and the diesel peakers would run at 45%. Stage 4 sheds up to 4,000 MW. Even at that full depth, 4 TWh takes 1,000 hours to shed.

So what changed in twelve months?

The exit didn’t. Last year’s Outlook (2026-203) scheduled 8.74 GW of Eskom shutdowns in 2029 and 2030. The 2027-2-36 MTSAO schedules 8.75 GW. Cahora Bassa’s 1,150 MW contract still ends in 2030, and this edition adds the 1,005 MW Avon and Dedisa peakers, whose contracts also end that year. Look further out, and the withdrawals keep coming: 16.2 GW of Eskom plants by 2036, plus 3 GW of the first renewable projects reaching the end of their contracts between 2033 and 2036.

Everything that changed sits on the replacement side. Gas has left the 2030 picture; none is now assumed before 2032. The coal fleet is also assumed to run much better. Last year’s Outlook used a 60% average availability. This year’s Reference Case climbs a point a year, from 66% to 70% by 2031, and holds there to 2036. Its downside case holds at 65%, roughly where the fleet stands today.

That assumption carries the 2030 result. Apply ten points of availability to the 31.7 GW of coal left after the shutdowns and you’ve found about 3 GW of capacity in a maintenance plan. It’s the cheapest power station in the country. It’s also one we’ve watched collapse: availability fell from 77% to 56% in five years.

What happens when the promise isn’t kept? The Outlook answers that itself. Its Perfect Storm case holds availability at 65%, adds no gas, and assumes high demand and slower rooftop solar. Unserved energy breaks the limit in 2033 at 169 GWh and reaches 943 GWh by 2036, 47 times the System Operator’s own adequacy criterion of 20 GWh a year. That criterion is the equivalent of five hours of Stage 4. The Perfect Storm reaches about 236.

The cliff hasn’t gone. It’s moved three years to the right.

Some readers will object that demand is falling, not rising. Statistics South Africa’s figures show electricity distributed through the grid fell 3.9% in the first half of 2026, and the Perfect Storm uses the Outlook’s high-demand path. That’s a fair point, and lower demand would delay the breach. But most of the load leaves the grid in daylight, because what replaces it is mostly solar. The Outlook itself says rooftop solar “reduces daytime net demand but steepens the evening ramp”. A business running on its own panels still leans on the grid at seven on a winter evening. On the high path, peak demand rises from 35.8 GW to 49.1 GW by 2036. The coal retirements remove firm capacity, the kind that counts after dark.

Others will say the Perfect Storm stacks four bad assumptions on top of each other. They’re right, and the Outlook says so itself. That’s why the case it didn’t run matters most. Gas arriving late while the fleet performs well “would be expected to produce outcomes between the Reference Case and Perfect Storm,” the System Operator writes, “but this case was not modelled.”

Two days before the Outlook was published, that unmodelled case stopped being hypothetical. The first gas IPP bid window sought 2,000 MW. Four bids offered 2,830 MW. None qualified, mainly because no bidder could guarantee gas supply. South Africa imports most of its gas from Mozambique and has no LNG import terminal. The Minister’s 7 October determination for 5,000 MW of gas creates a mandate to procure. It doesn’t create molecules. The Outlook concedes as much in its closing pages: “Gas delivery risk has increased.”

The second gap is legal. On 31 March 2025, the Minister of Forestry, Fisheries and the Environment capped Eskom’s emission-standard exemptions at Kendal, Lethabo, Majuba, Matimba and Medupi at 1 April 2030, and Tutuka’s at 5 June 2030. After those dates, each station must comply, win a fresh exemption or stop emitting at today’s levels. That decision stands until a court overturns it. Last year’s Outlook said this risk, covering more than 22 GW, “has not been considered”. This year’s doesn’t mention emission standards at all. Yet it assumes the fleet, including those six stations, runs at 65% to 70% until 2036.

So the Outlook’s comfort rests on the very stations whose exemptions run out in 2030. The 8.75 GW is the least the system loses in that window. It isn’t the most.

Before the Outlook appeared, I wrote an open letter to NERSA. Its argument was that the coal exit is dated in law and the replacement isn’t dated at all. I asked the regulator to have the System Operator answer five questions about its assumptions.

The Outlook answers some of them. It counts none of the Minister’s 9,600 MW determination, models every battery as four-hour storage and ends Cahora Bassa in 2030 in every case. It doesn’t give a station-by-station shutdown list, discount for the delivery record, or put probabilities on any scenario. That last gap matters most. Without probabilities, nobody can tell a consumer how likely the bad case is.

I’ll own one miss. I expected the test to come in 2031, the first full year after the exit. The Outlook shows no shortfall that year; the risk has moved to 2033. The letter’s core point stands, and the failed bid window proved it: the exit has a date, and the entry still doesn’t.

I’ve built a measure for this problem, the Cliff Intensity Index. Divide how fast firm capacity retires by how fast replacement can be integrated, then adjust for institutional friction. A score of 1 means replacement keeps pace. South Africa scores 2.25. The working paper has been public on SSRN since December 2025 and is now under peer review.

When the new Outlook landed, I tried to break the index. I credited the fleet with the full recovery to 70% as if it were new capacity. I added Avon and Dedisa to the retirements. I stripped out all institutional friction. The index still doesn’t fall below 1.45. Retirement still outpaces replacement.

That fits the Outlook. A system that retires faster than it rebuilds doesn’t fail on the day the plants close. It spends its margin, and fleet performance decides how long the margin lasts. At 60% availability, it ran out in 2030. At 65% it runs out in 2033. If availability slips again, it runs out sooner.

I’m not arguing for coal, or against gas or batteries. I’m arguing for planning that shows its working. Three steps would do it.

First, the System Operator should run the case it skipped: gas delayed beyond 2032, with availability at 65% and at 70%. It should publish the year the 20 GWh limit breaks under each.

Second, it should run a case in which the six stations’ exemptions lapse unresolved, and label every post-2030 megawatt from them as conditional on a legal step that hasn’t happened.

Third, as I asked in the letter, NERSA shouldn’t concur with any procurement until battery capacity is stated in megawatt-hours, with commissioning windows set against the 2030 exit. The Outlook’s own climate-stress test shows why. Cut wind and solar output by 40% for a week, and four-hour batteries can’t sustain supply while the diesel peakers run far beyond a peaking role.

The good news in this Outlook is real, and Eskom’s people earned it. But a plan that needs availability to rise five years running, gas to arrive on time and a legal deadline to quietly disappear is a hope with a spreadsheet attached.

The 2030 cliff has been rescheduled. We have three more years to build the bridge. We shouldn’t spend them celebrating.

Koko is Managing Director of Matshela Energy, a former Eskom Interim Group Chief Executive (2016-2017) and a doctoral candidate at the UNISA Graduate School of Business Leadership. This article draws on his doctoral research.