Context: Ramaphosa’s call to the Presidential Climate Commission
In a special session of the Commission held on 20 February 2026, Cyril Ramaphosa called for an accelerated, inclusive transition that treats energy security and climate action as mutually reinforcing, and explicitly urged faster deployment of renewables, expanded transmission infrastructure, enabled storage solutions, and greater private-sector participation.
He also framed “justness” as a multi-dimensional requirement: “redistributive, restorative and procedurally just,” and used the speech to emphasise vulnerable groups and worker livelihoods as central design constraints, not afterthoughts.
This acceleration message is tightly coupled to South Africa’s implementation challenge: turning high-level plans and pledges into grid-connected projects, while simultaneously preventing communities affected by coal phase-down from being stranded. The President named the need to address the redevelopment of Komati Power Station in Mpumalanga as an example of unfinished “justice work” that must not be left behind.
A practical bridge between “acceleration” and “justice” is measurable delivery. The President’s own checkpoints point to a delivery stack:
- (a) more clean generation,
- (b) more transmission to unlock resource regions,
- (c) more flexibility (storage + dispatchable resources), and
- (d) governance and social dialogue to keep the transition legitimate.
What a “Just Energy Transition” means in practice
A “just energy transition” combines two ideas:
Energy transition: shifting a power system’s supply, networks, and demand patterns toward low-carbon energy (and often electrification), while maintaining reliability and affordability. In South Africa’s planning documents, this shows up as: modelling supply–demand balance on an hourly basis; explicitly accounting for variability of wind and solar and unexpected outages; and measuring adequacy in terms such as unserved energy and system security.
Just transition: managing the social and economic consequences of that shift so that affected workers, communities, and vulnerable groups are protected and can participate in the new economy. The International Labour Organization and the International Energy Agency summarise this well: the ILO definition (quoted by the IEA) describes a just transition as “greening the economy” in a way that is “fair and inclusive,” creating decent work and “leaving no one behind,” and stresses rights, social dialogue, and labour protections. The ILO’s just transition guidelines further highlight policy coherence, social protection, skills and employment measures, and social dialogue mechanisms.
In South Africa’s own governance lexicon, “justice” is also framed as procedural + redistributive + restorative: involving those most affected in decision-making (procedural), ensuring benefits and burdens are fairly shared (redistributive), and repairing harms and rebuilding local economies (restorative).
The Three Pillars of Energy Justice
| Pillar | Focus | Practical Application |
| Procedural | Decision-making | Involving coal-affected communities in the planning process. |
| Redistributive | Fairness | Ensuring the benefits (jobs/cheaper power) reach the marginalized. |
| Restorative | Healing | Repairing the economic and environmental harms of the fossil fuel era. |
A useful working definition for this blog is:
Just energy transition: the accelerated shift to a low-carbon, reliable electricity system (and supporting industrial changes), designed and governed so that the transition’s benefits and costs are fairly distributed, affected workers and communities are protected and empowered, and historically marginalised groups participate meaningfully in decisions and opportunities.
This definition matters because it changes what “success” means. It requires tracking reliability, affordability, jobs/skills, local economic outcomes, and participation, rather than only tracking megawatts and emissions.
A global finance-and-delivery layer has also shaped the South African transition narrative through the Just Energy Transition Partnership, launched at COP26 and aimed at supporting a shift away from coal while ensuring fairness for workers and communities. One jointly endorsed investment plan described a “just approach” as ensuring coal-affected workers and communities (including women and girls) are not left behind, and it set out large-scale financing requirements and instruments intended to catalyse implementation.
Clean energy options in South Africa and the output question
The clean options that exist today
South Africa’s clean/low-carbon electricity options already span a diversified set of technologies, even though coal still dominates energy production. The “clean” options that are operational and scalable include:
Solar PV (utility-scale and embedded/distributed)
South Africa’s planning baseline (end-2025) includes 3,646 MW of grid-tied solar PV plus 7,065 MW of distributed generation (largely rooftop and C&I self-generation, though “distributed” can be multi-technology). A key planning and operations nuance is that some official renewable datasets exclude embedded generation (behind-the-meter), which matters because embedded PV reduces midday net demand and changes the operational “shape” of the system.
Wind
The same end-2025 baseline includes 5,344 MW of wind.
Concentrated solar power (CSP)
CSP appears in the baseline at 600 MW and is often discussed as a form of solar with dispatchable potential when paired with thermal storage.
Hydro and storage-as-flexibility
Hydro is relatively limited domestically (1,600 MW in the end-2025 baseline), but storage is strategically important. When energy storage is aggregated in the IRP 2025 baseline as a category, it totals 3,444 MW (which, in that document’s accounting, combines pumped storage and battery storage components).
Nuclear (low-carbon, non-renewable)
South Africa’s operational nuclear capacity is 1,860 MW in the end-2025 baseline. (In practice, this is primarily associated with Koeberg, but the key point here is the low-carbon firm generation role.)
Clean energy from IPPs (procured renewables)
Grid-connected renewables procured under the REI4P programme and related channels are significant. One regulator report (status as at end-December 2024) indicates total grid-installed capacity of procured renewables at 6,330 MW, with 6,299 MW in commercial operation at that time.
Is clean output “the same” as non-clean output?
It depends on what “output” means. There are two different “outputs” that matter for power systems:
Energy output (MWh over time)
For the same installed MW, different technologies deliver different annual energy because their average utilisation differs. A widely used proxy is capacity factor / load factor.
A CSIR utility statistics report (data up to December 2024) reported 2024 capacity factors roughly as: coal ~50%, wind ~36.8%, and solar PV ~26.3% (with nuclear ~47.7% in that dataset). A regulator report using REI4P plant performance for 2024 reported similar achieved load factors for REI4P: roughly wind ~37%, solar PV ~25%, and CSP ~30%.
These figures imply that 1 MW of wind or solar does not produce the same annual energy as 1 MW of coal. As a simple illustration, if solar PV averages ~25–26% utilisation, you need close to ~4 MW of solar PV nameplate to get ~1 MW of average power over the year; for wind at ~37%, you need ~2.7 MW. (This is an energy arithmetic point, not a cost or feasibility conclusion.)
Capacity output (firm/dispatchable MW at the right time)
Even more important than energy is whether capacity is available during peak periods and under contingency events. Variable renewables are not dispatchable in the same way coal, gas turbines, hydro, or nuclear are; their output follows weather and daylight patterns. The IRP 2025 explicitly defines residual load as demand remaining after renewable production (like wind and solar) is accounted for, and uses hourly-stochastic methods to test adequacy under demand variability, renewable variability, and unplanned outages.
This is why “matching coal’s output” cannot be framed purely as “replace X MW of coal with X MW of solar.” The system must replace both:
• coal’s contribution to annual energy, and
• coal’s contribution to firm capacity at critical hours (or compensate via storage, flexible generation, demand response, and transmission).
The Math of Clean Energy: Why 1 MW 1 MW
One of the biggest hurdles in the public discourse is the “output question.” If we decommission a 1,000 MW coal plant, we cannot simply build 1,000 MW of solar and call it a day.
To understand why, we look at the Capacity Factor (CF), the ratio of actual energy produced to the maximum possible output. Based on 2024 performance data, the math looks like this:
- Coal: CF
- Wind: CF
- Solar PV: CF
The Energy Arithmetic: > To match the annual energy output of 1 MW of Coal, you would need approximately:
This is why the IRP 2025 calls for a massive 105 GW of new capacity by 2039. We aren’t just replacing energy; we are replacing firmness. The system must be able to meet demand even when the sun sets and the wind dies down, requiring a massive rollout of storage and transmission lines ( by 2034).
What production capacity could be exploited to satisfy demand?
A pragmatic answer should reference both (a) national demand magnitude and (b) the scale of build-out implied by national planning.
Demand magnitude (energy and peak)
South Africa’s annual peak demand in 2024 was reported at about 33.5 GW in a CSIR report. On the energy side, a regulator report notes total system energy for resources shown of about 220 TWh in 2024 and a CSIR analysis reported annual energy demand at ~219.6 TWh in 2024.
Current dominance of fossil output
Even with growing renewables, coal remains the main energy contributor. The regulator report attributes about 82% of energy in 2024 to coal, with renewables at about 8.3% and imports around 4.6% (in that accounting), while the IRP 2025 states the system is dominated by coal and references coal’s large share of production.
Capacity potential and planning scale
The IRP 2025 states that, based on a grid survey of renewable developer interest, there is “up to 134 GW of potential” wind and solar, and it uses such nodes in modelling (while acknowledging grid constraints).
Crucially, the IRP 2025 frames the required build as massive: it describes adding 105 GW of new capacity between now and 2039 and shifting the mix “decisively towards cleaner sources.” Its “Proposed Balanced Plan” table implies (new build totals over the plan horizon shown in that table) approximately:
• ~43,041 MW new wind
• ~28,713 MW new solar PV
• ~9,000 MW distributed generation
• ~8,223 MW energy storage
and additional flexible supply such as gas and some nuclear build in certain scenarios.
A back-of-the-envelope “energy sufficiency” check (not an hourly adequacy proof) shows why these magnitudes are in the right order. Using 2024 utilisation proxies:
• 43.0 GW wind at ~36.8% yields ≈ 139 TWh/year
• 28.7 GW solar PV at ~26.3% yields ≈ 66 TWh/year
Together that is ≈ 205 TWh/year, before adding hydro, existing renewables, distributed generation output, nuclear, and imports.
This is close to the ~220 TWh/year scale reported for 2024 system energy, which suggests that tens of GW of wind and solar are indeed the right scale to replace a large portion of coal’s annual energy.
What blocks “capacity exploitation” is less about theoretical resource availability and more about delivery bottlenecks: transmission, connection capacity, and flexibility. The IRP 2025 points directly to transmission requirements: a cited TDP result in the IRP indicates 14,494 km of new lines required by 2034 (and 5,044 km by 2029) to accommodate increased generation and demand requirements.
So, the exploitable capacity pathway is:
• Scale wind + solar toward the 70–100+ GW level over time (depending on capacity factors, electrification demand growth, and demand-side changes), supported by large-scale storage and flexible resources.
• Build transmission fast enough that high-yield resource zones can connect to load centres.
• Treat “net load” (demand minus variable renewables and embedded PV) as the operational forecasting object that dictates how much flexibility is required.
Deep-learning projects that directly support a just transition
A just transition-oriented AI portfolio should be judged on two axes at once:
- (1) grid value (reliability, cost, emissions), and
- (2) justice value (affordability, fairness, livelihoods, visibility of impacts).The projects below are framed to connect demand, capacity, and roadmap feasibility.
Net load and adequacy intelligence
Net load forecasting with embedded PV inference
Because some renewables datasets exclude embedded generation, a high-impact project is a model that infers behind-the-meter solar contribution and forecasts residual/net demand under different rooftop adoption trajectories. This matters for municipal revenue and tariff design (justice), while also improving dispatch and maintenance planning (grid value).
Probabilistic adequacy forecasting (unserved energy risk)
The IRP 2025 explicitly uses stochastic techniques and treats adequacy in terms such as unserved energy. A deep-learning project can learn a fast surrogate for “risk of shortfall” given weather, unit availability signals, and demand forecasts, supporting day-ahead and week-ahead public risk communication.
Coal unit availability and planned maintenance forecasting
Where coal remains dominant (as it did in 2024 energy shares), predicting forced outages and availability at unit and station level allows better sequencing of retirements and maintenance, reducing the probability that transition steps trigger reliability crises.
Flexibility, storage and dispatch optimisation
Storage dispatch with reinforcement learning under fairness constraints
Storage is repeatedly highlighted as essential (by the President and in planning). RL can learn charge/discharge policies that minimise unserved energy and diesel/gas peaking, while including constraints such as: protect low-income feeder areas (priority load), cap tariff shock proxy costs, or maintain minimum reserve margins. The “justice layer” is implemented as constraints and multi-objective rewards, rather than a separate narrative.
Curtailment and congestion forecasting to support equitable grid access
As renewables scale, grid constraints create curtailment and connection queues. Forecasting congestion risk and curtailment windows supports fairer connection decisions and targeted grid upgrades. The IRP 2025’s emphasis on transmission bottlenecks and build-out need makes this directly relevant.
Roadmap accountability and feasibility modelling
Scenario-to-feasibility “digital twin” of planned build-out
The IRP 2025 provides a structured procurement/build plan and explicitly models variability and outages in assessing adequacy. A deep-learning project can build a fast surrogate that maps a candidate build schedule (wind/solar/storage/gas/transmission milestones) into probabilities of meeting constraints: reliability, emissions trajectory, and affordability proxies, enabling transparent dashboards for “are we on track?” discussions.
Socio-economic benefit tracking from procurement performance
Regulator reporting already includes socio-economic development/ownership reporting for IPPs and performance stats. A project can turn these into an interpretable “justness scorecard” attached to capacity additions (e.g., location, local community participation, jobs, skills programmes), and compare alternative build sequences for fairness outcomes.
Why TimesFM-style models are a good fit
Time-series foundation models are designed for cross-domain forecasting and can materially improve the “data scarcity” problem that appears when you need many forecasts (multiple regions, multiple net load definitions, multiple scenarios).
Google’s TimesFM is explicitly positioned as a decoder-only transformer foundation model for forecasting, pre-trained at very large scale (Google describes a corpus on the order of 100 billion real-world time-points) and demonstrating strong zero-shot performance in the associated publication. The open-source repository supports PyTorch installation and checkpoints, and it notes a 2025 update that adds back covariate support (XReg) for TimesFM 2.5, useful when you want to inject weather and operational covariates.
Flagship Net Load Platform (Dockerised Demo)
GitHub Link: flagship-netload-platform-
A repo-first, multi-container demo system for:
- Probabilistic net-load forecasting (quantiles / P50–P90 bands)
- Storage dispatch (baseline rule policy; RL-ready interface)
- Roadmap feasibility (scorecard baseline; surrogate-ready interface)
- Observability hooks (DogStatsD-compatible metrics emission)
Core principle: everything revolves around net (residual) load — the operational object that drives dispatchable needs and adequacy risk.
South Africa’s electricity system is constrained by a combination of demand growth, variable renewables, unit availability, and infrastructure delivery risk (transmission and build schedules). In operational terms, the system is governed by a single variable that constantly answers:
How much dispatchable power do we still need after renewables and embedded generation?
That quantity is net (residual) load.
This repo gives you a credible flagship engineering artefact you can run on a laptop:
- Forecast net load with uncertainty bands (quantiles).
- Dispatch storage against that uncertainty (risk-aware schedule today; RL-ready tomorrow).
- Score roadmap feasibility under scenarios (is the plan on track, and what bottleneck binds first?).
Where it’s useful
- Policy & delivery accountability: a transparent monitor for “Are we building enough (and fast enough)?”
- System operations / SRE-style decision support: forecast uncertainty + dispatch outcomes you can track.
- Energy + data-centre planning: stress the system with a data-centre load proxy and quantify the adequacy risk.
- Research & prototyping: fast iteration on models and contracts without needing a full power-flow model.