Marginal Emissions and the Real Grid

The previous lesson introduced the difference between counting the average emissions of a grid and counting the clean attributes you purchased. This lesson goes deeper into a distinction that is subtle, technical, and quietly decisive: the difference between average emissions and marginal emissions. Understanding it changes how you evaluate a site's true carbon impact, and it explains why two campuses on the same grid can have genuinely different carbon footprints depending on when and how they consume.

Start with average emissions. A grid's average emission rate is the total emissions of all its generation divided by all the electricity it produces, over some period. It is a blunt, aggregate figure: this grid is, on average, so many grams of carbon per unit of electricity. It is easy to calculate and easy to report, and it underlies most location-based accounting. But it answers the wrong question. When your campus consumes an additional unit of electricity, the grid does not respond by producing a little more of its average mix. It responds by ramping up whatever generator is on the margin — the next unit of supply, the one that switches on to meet incremental demand. On most grids at most times, that marginal generator is a fossil-fuel plant, because clean generators like wind and solar run whenever they can and are rarely the ones held in reserve. So the carbon consequence of your consumption is the marginal emission rate, which is frequently much higher than the average.

This distinction has real bite for a data center, whose defining feature is a large, constant load. Because the campus runs around the clock, it draws power in the dirty marginal hours as well as the clean ones, and its true carbon impact is better described by marginal rates than by comfortable averages. A campus that could shift some of its flexible load — certain computing tasks that need not run in a specific hour — toward the times when clean generation is on the margin would genuinely reduce emissions, while a campus that ignores timing entirely is quietly responsible for more marginal fossil generation than its average-based report admits. This is the sophisticated frontier of carbon accounting: not just what mix the grid has, but what your incremental demand actually causes the grid to do.

Now connect marginal thinking to siting, because it does bend the map, if more subtly than water or raw power. Grids differ enormously in what sits on their margin. On a grid where the marginal generator is usually gas, every incremental unit your campus draws is relatively dirty, no matter how green the grid's average looks. On a grid with abundant clean generation frequently on the margin — lots of hydro, for instance, or so much wind and solar that clean power is often the marginal unit — your incremental demand causes far less emission. A carbon-serious developer, looking past average figures to marginal ones, will prefer the second kind of grid, and this preference reorders the map in ways that average-based screening would miss. It also interacts with the hourly-matching discussion from the previous module: a site where clean power is frequently on the margin is a site where hourly matching is achievable and where marginal emissions are low, and these are two views of the same underlying virtue.

There is a further consequence for how campuses should behave once built, which loops back into their carbon story and their social license. A large flexible load is not merely a burden on the grid; it can be an asset if it is willing to modulate. A campus that ramps its flexible computing down when the grid is straining and dirty, and up when clean power is plentiful, reduces both its marginal emissions and the strain it places on the community's grid. This is sometimes framed as being a good grid citizen, and it is increasingly part of how regulators and communities judge whether to welcome a campus. The carbon accounting and the social license reinforce each other: the same flexibility that lowers your marginal emissions also lowers the friction of getting permitted. A campus that treats its load as immovable forfeits both advantages.

Step back and see the shape of the whole module. Carbon accounting began as a matter of boundaries — which scopes, which method, whether embedded carbon counts. It ends as a matter of realism about how grids actually respond to demand. The through-line is honesty: the discipline exists to close the distance between the clean story a campus tells and the emissions it actually causes. Averages, annual matching, and narrow scopes all widen that distance and flatter the teller. Marginal thinking, hourly matching, and wide scopes all narrow it and expose the truth. The reason this matters for siting is that honest accounting reveals the real differences between places that flattering accounting conceals. A developer who accounts honestly will find that the map is more sharply drawn than the averages suggested — that some grids are genuinely far better than others for a round-the-clock load, and that the best sites are the ones where the honest numbers, not just the convenient ones, come out clean. Carbon accounting, in the end, is not the conscience of the project checking its work. It is the instrument that tells you, before you commit, which sites will still look clean once someone insists on counting properly.