Climate & Energy AI Analyst

An analyst presents a proposal claiming a software optimisation will save a country-scale quantity of electricity. Nobody in the room can immediately tell whether the figure is plausible, so it goes unchallenged for two weeks. The error was three orders of magnitude, and it was detectable in seconds by anyone carrying a rough sense of the scales involved. Numeracy about energy is not background knowledge in this field, it is the primary defence against wasted work.

The units, and how they nest

Energy is measured in watt-hours, and the prefixes climb quickly. A kilowatt-hour runs a typical appliance for a few hours. A megawatt-hour is a thousand of those. A gigawatt-hour is a million, and a terawatt-hour is a billion — the unit in which national consumption is discussed.

Useful anchors: a European household uses very roughly 3,000 to 5,000 kWh per year. A mid-sized country consumes on the order of hundreds of TWh annually; large industrial economies run into the thousands. Global electricity consumption sits in the tens of thousands of TWh.

The distinction that trips people constantly is power versus energy. A watt is a rate; a watt-hour is a quantity. A gigawatt is a capacity figure — how much a plant can deliver at an instant — while a gigawatt-hour is what flows over time. Conflating them produces errors of exactly the magnitude in the opening example.

Capacity factor is where intuitions break

Capacity factor is actual output divided by theoretical maximum output over a period. It is not efficiency, which describes how much input energy becomes useful output.

The numbers matter because they are widely misjudged. Nuclear plants run at very high capacity factors, typically above 85 percent. Wind is commonly in the 25 to 50 percent range depending on siting, and solar photovoltaic often lands between 10 and 25 percent depending on latitude and weather.

So a gigawatt of solar capacity and a gigawatt of nuclear capacity are not comparable quantities of energy — they may differ by a factor of four or more in annual output. Any comparison stated in capacity rather than generation is likely to be misleading, and spotting that substitution is one of the more useful habits in this field.

Baseload, peak, and why timing dominates

Demand varies by hour and season. Baseload is the persistent minimum; peak is the daily or seasonal maximum. Grids must match supply to demand continuously, and peaking capacity is typically both the most expensive and the most carbon-intensive.

The consequence that shapes most climate-AI work: when energy is used matters as much as how much. Shifting a flexible load from a peak hour to an off-peak one can reduce emissions substantially without reducing consumption at all, because the marginal generator differs by hour.

Failure modes

Power and energy conflated — capacity figures treated as generation. Signal: a claim in gigawatts where gigawatt-hours are required.

Capacity compared across technologies — installed capacity compared without capacity factors. Signal: solar and nuclear compared in megawatts.

Average emissions factors used for marginal decisions — grid average applied to a load-shifting question. Signal: an analysis with no hourly resolution.

Order of magnitude unchecked — a result not sanity-checked against a known anchor. Signal: a saving claim comparable to national consumption.

Explaining this to stakeholders

On power versus energy: a gigawatt is how fast, a gigawatt-hour is how much. Confusing them is like confusing speed with distance.

On capacity factors: a gigawatt of solar and a gigawatt of nuclear do not produce the same electricity in a year. The sun does not shine at night, and the ratio is often four to one or worse.

On timing: moving a flexible workload to a different hour can cut emissions without cutting usage, because the last plant switched on to meet demand varies by hour and it is usually the dirtiest one.