Results are for one weather year. Cost figures: system cost and avg energy cost cover fuel, carbon, start-up, new-build capital, fixed O&M for the existing fleet, REIPPPP payments and grid reinforcement; rooftop PV is excluded. Load shedding is costed separately at R12.30/kWh, Eskom’s cost-of-loadshedding figure of R9.53 in 2020 rands escalated to 2026, and shown beneath system cost when it occurs. Replacement cost instead prices the mix at full lifecycle cost, as if it were built new today: generation per MWh, and storage and gas on the capacity the scenario holds. Neither is a tariff. Methodology in the README.
Run the full model
network-aware MIP · HiGHS · runs in your browserHourly dispatch
modelledHourly dispatch chart: generation by carrier for a representative 168-hour week. Tabular summary below.
Annual electricity mix
modelledLoad shedding risk
modelled60 outage-yearsNational engineBars: probability the year reaches at least stage S, across 60 simulated years.
Carbon budget
modelled · vs South Africa’s NDCGeneration capacity build rates
capacity expansion · least-cost schedule to 2030Least-cost build schedule – how much of each technology, in which year, and in regional mode where. Uses each coal unit’s retirement date, technology costs and build-rate limits.
Set build limits per technology – GW added per year
What happens if the grid build lags? – cost of delay
Everything above assumes the Transmission Development Plan lands on schedule. NTCSA itself notes the first five years carry high certainty and beyond 2030 is uncertain – and 52% of the mapped projects sit in Concept, Pre-Concept or generic Planned rather than Execution. This slips those and re-solves.
Provincial electricity mix
capacity: data, mid-2026 · generation: the engine at start-up, not the scenario · demand shares: source being confirmedGrid queue pressure
SAPVIA & NTCSA GCCA · pipeline vs headroomPermitted but not built
environmental approvalsLevelised cost comparison
R/kWh over each plant’s life at the use stated in its control, fuel and carbon included · several are assumptions with no published source, marked in the controlsThe no-gas frontier
modelled · worst of twelve weather yearsPrice-setting technology by hour
modelled · energy-only shadow priceHourly shadow price for a representative 168-hour week, each bar coloured by the technology setting the price in that hour.
Capture price forecast
modelled · energy-only shadow price · PPA value over five yearsBattery revenue benchmark
modelled · energy-only shadow price · lithium-ion 4h · what a BESS would earnBattery revenue split
modelled · lithium-ion 4h · arbitrage vs ancillary vs capacityCapture rate projections
modelled · energy-only shadow price · cannibalisation · wind vs solarBattery revenue as the fleet grows
modelled · revenue per MW · at this scenario’s yearWhere a battery earns most
modelled · lithium-ion 4h · nodal · needs full model runCarbon capture & storage retrofits for the coal fleet
policy test · changes every outputModel assumptions & caveats
What this is: a national-level digital twin of the South African power system. The default hourly demand, wind and solar profiles come from 2025 Eskom data; the twelve weather years (2014–2025) behind the risk panel, the extremes and the pathway checks are Renewables.ninja/MERRA-2 wind and PVGIS-SARAH3 solar. Fleet parameters – EAF, OCGT load factor, rooftop PV, structural demand level – are compared with Eskom’s weekly system status reports (): EAF ≈ % calendar year to date, OCGT load factor ≈ %, GW behind-the-meter rooftop PV (NTCSA’s August 2026 estimate less the ground-mounted wheeled plants this model counts as utility supply), post-2023 structural demand decline. Transmission and distribution losses are not modelled separately, and should not be: the demand series is Eskom’s transmission-level demand – what generators must send out – and is therefore already gross of downstream losses. Adding a loss factor on top would double-count them. Output is compared with Ember's metered generation (calendar 2025, Ember's monthly data). Koeberg runs at % sent-out by default. Wind and utility solar carry no availability derate: the national profiles are Eskom’s own metered hourly output, so availability, losses and real curtailment are already inside the per-unit series, and the regional profiles are Renewables.ninja, whose PV already carries a 10% system loss. The wind series divides each hour’s metered output by that hour’s measured installed capacity, from Eskom’s hourly data. Compared fleet-normalised, the model’s wind capacity factor lands within 1% of Eskom’s measurement; compared as energy it reads about 20% above Ember, a gap from fleet size and weather year. Rooftop PV carries a derate: a rooftop fleet takes whatever orientation the roof has, is shaded, is never cleaned, does not track, and sits in Gauteng and the Western Cape rather than the Northern Cape; the comparison panel shows the resulting capacity factor. NTCSA’s rooftop figure is itself an inference – NTCSA derives it from residual load on sunny versus cloudy days and counts only systems under 100 kWp – and analysts have questioned whether it runs high. The remaining gap is not calibrated away: Ember reports metered output, already net of the network curtailment Eskom applies to Cape wind, and it does not count privately wheeled plants or fully capture behind-the-meter rooftop. Calibrating the derates to close that gap would bake curtailment into a technical assumption and then double-count it whenever a scenario curtails. The validation panel shows the four carriers against Ember so the difference stays visible.
Price formation: coal sets the marginal price in almost every hour of the default scenario, so the shadow price is nearly flat and storage has little to arbitrage. Battery revenue, capture-price spreads and the value of flexibility are therefore small at default settings and grow once enough wind and solar are added to push coal off the margin. The model carries an explicit operating reserve on NTCSA’s ancillary services rule: MW split into instantaneous, regulating and ten-minute reserve to , then % of wind and solar output above what the fleet would give is added. At today’s fleet it rarely binds and does not set the price; it bites once coal is tight. Unit-level forced outages: availability comes from a two-state Markov process over the individual coal units rather than a flat derate, so several units can be out at once. The EAF slider still sets the annual average; what changes is the shape. A fixed seed keeps the headline scenario reproducible; the risk panel varies it. Part-load heat rate: a coal unit burns more fuel per MWh the further it is backed off, so emissions follow fuel burned rather than energy sold. Still simplified: unit trips are drawn independently, where real ones cluster (common-mode failures, coal quality, a stressed fleet), and intra-regional network limits below the ten-region corridor level remain out of scope. New build from the sliders is sited against GCCA connection headroom, so none goes to the Northern Cape and Hydra Central, which have had zero solar headroom since GCCA 2025. Beyond about 19.9 GW of new solar the model reports that the scenario has outrun national connection headroom instead of quietly absorbing it.
Demand response is modelled as two separate products, because they behave differently. Interruptible load – contracted industrial customers Eskom can drop, historically the smelters – defaults to MW because it exists, is called before any shedding, and is reported apart from unserved energy: a managed reduction under contract is not a blackout. It is not free, and the compensation is booked into system cost. Shiftable load – pumping, irrigation and EV charging moved within the day, strictly energy-neutral – defaults to zero (household water heating sits in the virtual power plant below), because at meaningful scale it does not yet exist here. It is a scenario lever that absorbs midday output, so it matters most beside a high-solar build.
Virtual power plants are modelled separately again, because they aggregate behind-the-meter assets the utility does not own – rooftop solar, household batteries, and above all controllable electric geysers. Cape Town has an RMI-supported feasibility study for a municipal VPP, and eThekwini’s Project Smart Solar with Plentify, funded by AFD and the EU, is connecting residential PV, batteries and geysers into a city-wide VPP. The controllable pool is sized at 4 GW – residential is about 17% of consumption and the geyser 40–50% of a household bill, averaging ~1.8 GW, but geyser load is what makes the morning and evening peaks, so it contributes far more than its average into them. Enrolment defaults to zero.
Carbon cap: the build optimiser can be run subject to an emissions ceiling rather than only reporting against the NDC, which is how the IRP poses the question. When the cap binds, the model reports what the last tonne cost to avoid.
The model’s own winter peak runs higher than Eskom’s highest reported evening peak ( GW) because it is driven by the 2025 demand series and includes firm exports (it leaves storage charging out); the System adequacy and validation panels show the gap.
How dispatch works: merit order – rooftop PV nets off demand → wind / utility solar / CSP → nuclear, hydro, Cahora Bassa imports → coal → pumped storage & batteries → gas CCGT → diesel OCGT → interruptible load → unserved energy. Ahead of a forecast shortfall, spare coal and gas charge the stores. One stage of load shedding ≈ 1,000 MW of unserved demand.
Unit commitment: coal is committed at individual unit level – units in groups across stations – in 4-hour blocks, respecting each unit’s minimum stable level ( of capacity, fleet-weighted ), minimum up and down times, and ramp limits. Until no unit runs below of capacity, the level that reproduces Eskom’s lowest midday coal output in 2026; from units can run down to their design minimums. Against a unit-commitment optimisation on the same rules (24 October 2026), the engine’s cost is 0.9–2.7% above the solver’s lower bound in each of seven weeks of 2026, and 1.3% above the optimum on October’s least-cost build for 2035.
| Coal | R/MWh (Eskom FY2026 coal unit cost, environmental levies excluded) · +R80 VOM · tCO₂/MWh (our figure; Eskom’s FY2026 report implies 1.11) · GW installed (Eskom FY2026 nominal) · real per-unit retirement dates to 2051 |
| Diesel OCGT | R/MWh fuel · +R70 VOM · 0.78 tCO₂/MWh · 3.4 GW |
| Gas CCGT | R/MWh fuel · +R35 VOM · 0.348 tCO₂/MWh · LNG-fired, the central gas case: 13.25% of Brent at USD 80 in 2024 dollars, USD 11.74/MMBtu at the coast with shipping, plus USD 0.25 of regasification, × R16.50/USD ÷ 0.293071 MWh/MMBtu ÷ 52% efficiency, the same in every year. The LNG terminal is a fixed R/kW-yr on new gas burning LNG, the midpoint of the sourced R354–707 range (a choice). Carbon added separately. Spot ($23/MMBtu) is about R2,490/MWh; pre-Hormuz consensus ($12) about R1,300. The national model treats gas plants as coastal (Richards Bay): no inland transport is charged. The inland adder (the Lilly pipeline at R9.00/GJ, NERSA 2025/26, once its flow is reversed; about USD 2.8/GJ by truck before then, from an undated source) goes into the regional optimiser. The 2040 pathways also run five gas price cases (low, central, Brent USD 100, high, and IRP 2025 as published), each delivered at the coast; all but the IRP case carry the LNG terminal as a fixed cost per kW of new gas. Shipping is their least-sourced input: a 2018 charter breakeven adjusted with US inflation and an estimated sea distance, and charter rates have likely risen since the Hormuz disruption. |
| Imports (Cahora Bassa) | R/MWh (Eskom FY2026, international purchases) · MW must-take at % · |
| Variable O&M | Applied, and shown above per source. Thermal only: wind, solar and battery carry their O&M as fixed R/kW-yr instead, the standard convention. Anchored to NREL ATB figures. Hydro and imports get no adder – hydro has no fuel so its R30/MWh already is consumables, and the import price is contracted and embeds the supplier’s own O&M. |
| Existing fleet | Wind 4.7 GW · Utility solar GW (grid-contracted + wheeled ) · Hybrid 0.34 GW (RMIPPPP, contracted dispatchable 05:00–21:30; modelled hourly as 05:00–22:00) · Rooftop GW (the 2026 mean) · CSP 0.6 GW · Batteries GW · Pumped storage GW |
| New-build capex | Wind R– · PV R– · Rooftop R– · Battery (4h) R– (BESIPPPP bid window 3 aged to 2026; inclusions not stated) · CCGT R– /kW, 2026. Declining per BNEF to 2035 (solar −30%, storage −25%, onshore wind −23%); CCGT rises, having hit a record high in 2025 |
| Network | 439 real Eskom transmission lines · 185 substations with verified coordinates and kV ratings (NTCSA shapefile, OSM, Eskom GPS, DBSA RFP 008 register) · GCCA 2025 connection headroom by region · 11 Renewable Energy Development Zones (GN 114 / Gazette 41445 2018 and GN 142/144/145 / Gazette 44191 2021), held as centre points and equivalent-area radii rather than gazetted polygons. Per-substation capability bands shown in the Grid connection tab are a GridTwin estimate derived from that regional figure plus network topology – they are not published by Eskom or NTCSA. |
| Planned build | 221 projects from NTCSA’s Transmission Development Plan 2025–2034, all nine provinces, each with its published commissioning year and delivery phase |
| Renewables vs Non-fossil | “Renewables” = wind + utility solar + rooftop PV + CSP + hydro. “Non-fossil” adds nuclear and (mostly-hydro) imports. |
Risk and variability: the risk panel dispatches 60 synthetic years, each drawing both an outage path and a real weather year. Coal availability is drawn unit by unit around the EAF slider, with station-level trips that take several units out together, calibrated to Eskom’s measured 2023–2025 availability; weather is drawn from twelve real years (2014–2025: Renewables.ninja/MERRA-2 wind at up to twelve sites per region, PVGIS-SARAH3 solar) and cycled so each year is used equally. The two are sampled independently, since unplanned outages track plant condition rather than the weather. Separately, Show wind & solar extremes re-solves the scenario against the same 12 real weather years. A single weather year can over- or under-state both adequacy risk and curtailment.
Capacity expansion: the build-schedule optimiser is a linear programme solved in-browser with HiGHS. It minimises discounted system cost. Costs are in constant 2026 rands. Years are weighted at % real (Meridian Economics’ review of the IRP 2023 uses 8.2% real). Each technology’s capital is charged at its own real pre-tax cost of capital for state-backed procurement, set from South African bids (REIPPPP, BESIPPPP) with the gaps between technologies from the UK’s 2025 hurdle-rate study: solar and wind %, lithium and pumped storage %, gas %, nuclear %, transmission %, with interest during construction where a cost is quoted before financing. IRP 2025 uses one rate, 11.3%, without stating the basis (Gazette p.34, with all costs ‘in January 2024 rand terms’), which we treat as real. Results are tested with one rate for all, private finance for wind, solar and lithium, and higher and lower costs of capital. The optimiser runs from 2026 to a horizon of 2030, 2035 or 2040 (the Horizon control), using representative days rather than all 8,760 hours, so it sizes the build; Run the full model checks the resulting mix by unit and region. Reliability is checked year by year: where the hourly engine sheds more than the standard allows, the failing days are added to the optimisation as a stress window for the model year in which they failed, as NREL’s ReEDS model checks resource adequacy for each solve year (ReEDS documentation; Mai et al. 2024, doi:10.2139/ssrn.4841668). In national mode storage keeps an hourly state of charge: each representative day stands for a block of identical real days, its hourly level must stay between empty and full on the first and the last day of the block (which covers every day between them), and the level carries over from one representative day to the next. In regional mode it also decides where, subject to per-region GCCA headroom, real corridor limits, and a per-region annual build cap of 45% of the national rate.
What it is not: the instant model has no network constraints (the MIP does, across 10 regions) and no operating-reserve co-optimisation. Costs exclude existing-fleet capex, the existing network and retail; new wind, solar and offshore carry a charge for the network they need. It answers bulk adequacy, not distribution reliability, where most outages start. Everything runs in your browser, except the rooftop tool's address and roof lookups, which go to Google. The TDP is a published plan, not an investment commitment.
How the model compares to Eskom’s published figures
How the model compares to other published models
How the model works
Two engines answer two different questions. The instant engine re-dispatches the whole system in about half a second and treats South Africa as a single node, so it does not see where power flows on the network. The full model is a mixed-integer unit-commitment solve across the ten GCCA supply areas, run on 52 representative days with a 2% optimality gap, or over a full year. Both use the same fleet register. The full model uses regional weather profiles and PyPSA-RSA’s per-station costs, so the two differ on inputs as well as on the network.
Unit commitment is modelled per station rather than per technology: minimum stable levels, minimum up and down times, and start-up costs, with per-unit parameters taken from PyPSA-RSA, except the minimum stable level of of the groups, which is our own 0.50 (unsourced; PyPSA-RSA's workbook has 0). Coal carries a part-load heat rate penalty, so backing a unit off raises its fuel burn per MWh and its emissions with it. Storage dispatches on opportunity cost rather than fuel cost, and each technology keeps its own state of charge and round-trip efficiency. Solver is HiGHS, compiled to WebAssembly and running in your browser – nothing is sent to a server.
Corridors between supply areas are represented by a DC power flow, not a transport model. Susceptances are derived from the real line geometry in NTCSA’s published transmission shapefiles: length is computed from the routes, and reactance is assigned from voltage class using standard overhead-line values, which is the method PyPSA and PyPSA-Eur use. Parallel circuits add susceptance. Series compensation is applied to the long high-voltage lines, because Eskom has compensated its backbone since 1975 and ignoring it makes those lines look electrically longer than they are; the thresholds were tested for sensitivity rather than assumed. The HVDC circuits to Cahora Bassa are excluded from the power flow entirely and treated as a fixed import, since HVDC flow is set by converter controls rather than by impedance.
Reactance here is assigned rather than measured, and conductor bundling and per-scheme compensation settings are not public.
Every price here is a shadow price: the marginal cost of serving the next MWh in that hour. South Africa’s wholesale market, SAWEM, has traded inside Eskom since April 2026, with commercial launch targeted for 2027 (Krutham, October 2026); until its prices are published there is nothing to compare these against. They are what the system would pay if dispatch were priced rather than administered. The market is designed to clear to a single national price, so a single-node price is its basis; the locational signal lands on curtailment and network charges instead.
Hourly wind and solar profiles are per supply area across twelve weather years, 2014–2025, so a scenario can be tested against a bad wind year rather than an average one. Wind comes from NASA MERRA-2 via Renewables.ninja, sampled at up to twelve sites per region chosen from environmental authorisations and gazetted development zones, then calibrated so the national mean matches Eskom’s metered output. Solar is PVGIS satellite data at 5 km. Demand is Eskom’s measured 2025 hourly profile, domestic only, scaled by the growth slider. The 2040 pathways grow demand on the IRP 2025 reference path year by year, from the forecast behind the IRP (UCT Energy Systems Research Group demand model, IRP 2023–24 revision; Merven, doi:10.25375/uct.26942134.v5, CC BY 4.0), applied as growth on measured 2026 demand. Their levels therefore sit about 11% below the IRP’s own: on the same basis the IRP’s forecast put 2026 at 246 TWh against about 219.5 TWh measured. Each comparison with the IRP’s build is also run at IRP 2025’s own demand, its Figure 2 Moderate forecast. IRP 2019 expected 306 TWh by 2030, demand on the IRP’s basis was about 239 TWh in 2024, and Eskom’s sales have fallen from 224 TWh (FY2012) to 178 TWh (FY2026).
Every change runs a test suite before it ships: per-carrier reconciliation against Eskom’s published outturn, capacity factors checked separately so that energy and CF errors cannot cancel and hide each other, hour-by-hour energy balance invariants across eleven scenarios, cross-panel consistency so two panels cannot report different numbers for the same thing, and a sweep asserting that every control still moves something. The two panels above run the same kinds of comparison on the page: the Eskom panel sets the ‘2025, as modelled’ run against 2025, and the published-models panel runs each study from the start-up settings, each at the year the study names.
Every data file the model uses is served with the page, because the model runs in your browser. The model logic, calibration, cost assumptions, code and presentation are © 2026 Nick Hedley and licensed under CC BY-NC-ND 4.0: you may share them with attribution for non-commercial purposes; commercial use or distributing modified versions needs written permission. Our compilations of public facts (project names, capacities, technologies, locations and dates) are licensed under CC BY 4.0. Files derived from other people's data carry the most restrictive terms among their inputs, stated in each file and listed in LICENSE: Eskom's website terms for Eskom-derived data (non-commercial; publication needs Eskom's consent, which has been requested); NTCSA's website terms for GCCA-derived data (non-commercial; publication needs NTCSA's consent); CC BY-NC 4.0 for Renewables.ninja-derived profiles; and PyPSA-RSA's AGPL-3.0-or-later, whose terms for our use are being confirmed with Meridian. Analysis derived from Eskom data is ours, not Eskom's. LICENSE lists each file in nodal/ with its licence; a file in public_data/ or the root with no licence of its own is CC BY-NC-ND 4.0, unless derived from a source LICENSE lists. If you are building something similar, get in touch to compare notes.