Marginal vs Average Grid Intensity
Two people ask what the carbon of grid electricity is, and both are right to get different answers. One is doing the company’s carbon accounts and wants to know the emissions behind the power it drew last year. The other is deciding whether to run a heavy compute job now or at 3am, and wants to know what turning on more load right now would actually cause. These are different questions, and they have different numbers.
The two numbers have names. Average grid intensity is the carbon of the electricity you consume; marginal grid intensity is the carbon of using more.
Average grid intensity is a grid’s total emissions divided by total generation — the basis of location-based Scope 2. Marginal intensity is the emissions of the generation that responds when demand changes. Average for accounting, marginal for decisions.
Definition — Two Ways to Price a Kilowatt-Hour
Both average and marginal grid intensity express the same thing — the carbon content of electricity, in grams of CO₂e per kilowatt-hour — but they answer opposite questions, and confusing them is one of the most consequential errors in energy carbon accounting.
Average grid intensity is attributional: it takes all the emissions of everything generating on the grid and divides by all the electricity generated, giving the mean carbon of a kilowatt-hour drawn from the mix. It answers “what was the carbon of the electricity I consumed?” Marginal grid intensity is consequential: it looks only at the generation that actually changes in response to a change in demand — the plant that ramps up when you plug in, or ramps down when you switch off. It answers “what is the carbon of using one more kilowatt-hour, right now?”
The distinction is the electricity-grid version of the same average-versus-marginal split that runs through digital carbon — the energy intensity of data transfer and the per-transaction figure for Proof-of-Work — and through attributional versus consequential life-cycle assessment. Getting it right means matching the number to the question.
Definition at a glance
| Average intensity | Marginal intensity | |
|---|---|---|
| Answers | Carbon of electricity consumed | Carbon of using one more kWh |
| How it’s found | Total emissions ÷ total generation | Emissions of the plant on the margin |
| Type | Attributional | Consequential |
| Used for | Accounting, Scope 2 reporting | Decisions — load-shifting, timing |
| Behaviour | Smooth, published as factors | Volatile, time- and place-specific |
Average Intensity: The Accounting Number
Average grid intensity is the familiar one — the number in national emission factor sets like DEFRA’s, currently 0.131 kg CO₂e per kWh [GreenCalculus grid.gbr.electricity.location_based · DEFRA 2026 'UK electricity'!E25 · v2026.284] for the UK. It is calculated by dividing the total emissions of all generation on the grid over a period by the total electricity produced, so it blends the coal, gas, nuclear, wind, solar, and hydro on the system into a single representative figure.
Its natural home is attributional accounting: reporting the carbon of the electricity an organisation consumed. This is exactly what location-based Scope 2 uses — your consumption multiplied by the average grid factor for your region — under the GHG Protocol Scope 2 Guidance. Average intensity is stable, published, and auditable, which makes it right for inventories and disclosure. What it is not built for is telling you the effect of changing your consumption.
Marginal Intensity: The Decision Number
Marginal grid intensity asks a sharper question: if demand rises or falls by a little, which generator responds, and what are its emissions? On most grids, the plants that sit idle waiting to ramp up for extra demand — the “marginal” units — are dispatchable fossil plants, typically gas, because wind, solar, and nuclear are already running whenever they can. So the marginal kilowatt-hour is frequently a fossil one even on a grid that looks clean on average.
Because it reflects what actually changes, marginal intensity is the right signal for decisions: when to charge a battery or an EV, when to run a flexible compute job, whether an efficiency measure is worth it. It is also highly variable — at a windy, low-demand hour the marginal plant might be one that would otherwise be curtailed, so extra demand is nearly carbon-free; at a still, high-demand hour it might be a peaking gas or even coal plant, so extra demand is very dirty. This volatility is the entire basis of carbon-aware computing and demand response: shifting flexible load to low-marginal-intensity hours cuts real emissions in a way that a smooth average would never reveal. It is why the SCI metric asks for a marginal, location-based factor rather than an average one.
Marginal intensity is harder to obtain — it must be modelled from grid dispatch data, as services such as WattTime and the electricity-map providers do, and as datasets like the EPA’s eGRID non-baseload rates approximate — but it is the only number that correctly values a change in when or whether power is used.
Why They Differ — and Which Is Higher
The two can diverge sharply, and the direction is often counterintuitive. On a grid with substantial clean baseload — nuclear plus must-run renewables — the average is dragged down by all that low-carbon generation. But that baseload is not what responds to a change in demand; the marginal plant is the flexible fossil unit at the top of the stack. So on a moderately clean grid, marginal intensity is frequently higher than average.
Higher is the common case, but not universal. When renewables are so abundant that some would otherwise be curtailed, the marginal generator is that surplus wind or solar, and marginal intensity drops toward zero — well below the average. The point is that marginal intensity is dynamic and can sit either side of the average depending on the hour, the weather, and the demand, whereas the average is a smoothed, slow-moving figure. Treating the average as if it captured this variation — for instance, assuming load-shifting saves the average intensity times the kWh moved — misjudges the benefit, usually by understating it.
Which to Use, and When
The choice follows directly from whether you are accounting or deciding:
| Task | Use |
|---|---|
| Corporate GHG inventory, location-based Scope 2 | Average |
| Disclosure & audit of consumed electricity | Average |
| Timing a flexible workload or EV charge | Marginal |
| Valuing an efficiency or demand-response measure | Marginal |
| Carbon-aware computing / SCI | Marginal |
In short: use average when the question is “how much carbon do I report for the electricity I used,” and marginal when the question is “what changes if I use electricity differently.” A third case, market-based Scope 2, sidesteps both by reflecting the specific contracts and certificates a buyer holds — a separate accounting choice, not a physical grid measure. The one thing you should never do is use a single number for both jobs and assume it fits.
Common Confusions
- Using average intensity for decisions. The impact of using more or less electricity, or shifting a load, is a marginal question — the average smooths away exactly the variation that makes timing worthwhile.
- Using marginal intensity for accounting. A GHG inventory reports the carbon of electricity consumed, which is an average, location-based figure — not the marginal plant’s emissions.
- Assuming marginal is always higher. It usually exceeds the average on a moderately clean grid, but at times of renewable surplus it can fall near zero. It sits either side of the average.
- Confusing marginal with market-based. Marginal is a physical grid concept; market-based Scope 2 is a contractual accounting choice. They are unrelated.
- Ignoring time. Both vary through the day, but marginal far more so — an annual marginal figure hides the very swings that make it useful.
Dataset — the full factor set behind this term, versioned with source provenance and downloadable as CSV with a citable Zenodo DOI, is published as the Google Cloud region carbon-intensity dataset.
Frequently Asked Questions
Both express the carbon of electricity in grams of CO₂e per kilowatt-hour, but they answer different questions. Average grid intensity is the total emissions of everything generating on the grid divided by the total electricity generated — the mean carbon of a kilowatt-hour from the mix, and the number used to report the carbon of electricity you consumed. Marginal grid intensity is the emissions of the specific generation that responds when demand changes — the plant that ramps up or down as you use more or less power. Average is attributional and answers “what was the carbon of what I consumed”; marginal is consequential and answers “what is the carbon of using one more kilowatt-hour now.” They frequently differ, because a grid’s average is lowered by clean baseload while the plant on the margin is often fossil-fuelled.
Use average intensity for accounting and marginal intensity for decisions. Reporting a corporate greenhouse-gas inventory, or location-based Scope 2 emissions, calls for the average grid factor, because you are attributing the carbon of the electricity you consumed — this is what the GHG Protocol Scope 2 Guidance requires, using stable, published, auditable factors. Deciding when to run a flexible workload, charge an EV or battery, or whether a demand-response or efficiency measure is worthwhile calls for the marginal intensity, because you are asking what changes when you change your consumption. This is also why carbon-aware computing and the Software Carbon Intensity metric use a marginal, location-based factor. The rule of thumb: average for “how much do I report,” marginal for “what changes if I act.”
Usually, but not always. On a grid with significant clean baseload — nuclear and must-run renewables — the average is pulled down by all that low-carbon generation, while the plant that responds to a change in demand is typically a flexible fossil unit such as gas. So marginal intensity is frequently higher than average on a moderately clean grid. But when renewables are so abundant that some would otherwise be curtailed, the marginal generator is that surplus wind or solar, and marginal intensity can fall toward zero, below the average. Marginal intensity is therefore dynamic and can sit on either side of the average depending on the hour, the weather, and the level of demand — which is precisely why it is the right signal for timing flexible electricity use, and why a smooth annual average cannot substitute for it.
Carbon-aware computing is the practice of shifting flexible computing work in time or place to run when or where the electricity is cleaner — for example, delaying a large batch job to a windy overnight hour, or routing it to a region with a cleaner grid. It relies on marginal grid intensity, not average, because the whole point is to value the change in emissions caused by moving the load, which is a marginal question. Using an average figure would hide the hour-to-hour variation that makes the shifting worthwhile, since the average barely moves while the marginal intensity can swing from near zero at times of renewable surplus to high during a fossil-fuelled peak. This is why the Software Carbon Intensity specification calls for a marginal, location-based intensity: it ensures that running work at cleaner times actually shows up as a lower score.
Location-based Scope 2 uses average grid intensity: your electricity consumption multiplied by the average emission factor for your grid region, reflecting the physical carbon of the local mix. Market-based Scope 2 is different again — it reflects the specific contracts, certificates, and tariffs a buyer holds, so a company on a certified renewable contract can report lower emissions than its physical grid would imply. Neither location-based nor market-based Scope 2 uses marginal intensity, because both are attributional accounting methods concerned with the carbon of electricity consumed, not the consequence of changing consumption. Marginal intensity sits outside the Scope 2 inventory framework entirely; it is a decision tool for timing and demand, not a reporting basis.
Need someone who does this? 6 carbon accounting & inventory providers in our directory →