Carbon CAPEX v Carbon OPEX

Nobody writes this headline: "This office block emitted four hundred tonnes of carbon before a single person sat down in it, therefore working from home is greener, permanently."

It would be absurd. You would immediately ask the obvious question: fine, but what does the building cost to run, and over how many years?

Yet that is, structurally, the argument made about electric vehicles every few months. Somebody totals up the carbon in a battery, presents it as a shocking number (and it genuinely is a large number), then stops there. The piece never gets to the second column.

This isn't really an argument about the environment. It's an argument about accounting. And the accounting is wrong.

Chapter oneTwo columns, not one

Every asset has two carbon costs.

Carbon CAPEX is the embodied cost. Mining, refining, cells, factory, assembly. Paid once, at the start, then amortised across everything the asset ever does.

Carbon OPEX is the running cost. Every mile, every kilowatt-hour, every inference. Recurring, for as long as you own it.

We run two electric cars, a rear-wheel-drive BYD Seal Design and a Mercedes EQB 250+, both sensible specifications rather than halo models. We are also building a sovereign AI platform backed by battery storage. So rather than argue about it, we opened the ledger on our own assets. Some of what we found supports the case for electrification. One finding embarrassed us. Both are below.

Chapter twoFirst, the part the sceptics get right

Our cars started life in carbon debt. That is true and I am not going to soften it.

Battery cell production runs somewhere around 52 to 69 kgCO₂e per kWh for the Seal's LFP chemistry and 59 to 79 for the EQB's NMC, depending heavily on the grid where the cells were made. On our pack sizes that is a starting debt of roughly 4.3 to 5.8 tonnes of CO₂e per car before either moved a metre. The ICCT puts EV production emissions around 40% above an equivalent combustion car. That gap is real.

Worth noting which variable dominates: it isn't chemistry, it's geography. An LFP cell built in China carries a higher footprint than an NMC cell built in Europe. Where you build matters more than what you build. Hold that thought, because it comes back.

Chapter threeThen the part they leave out

Both cars do 12,000 miles a year. The EQB 250+ returns a real 240 miles from its 70.5 kWh usable pack: 3.4 miles per kilowatt-hour, measured, not brochure. Against a comparable petrol car, and counting the full fuel cycle (combustion plus the extraction, refining and distribution that DEFRA reports separately and most comparisons quietly drop), here is where the debt clears.

• BYD Seal Design14,400 to 19,100 miles · 14 to 20 months
• Mercedes EQB 250+16,400 to 21,900 miles · 17 to 22 months

The wide bands are battery-carbon uncertainty. Anyone quoting you a single precise figure for battery carbon is selling something.

Both cars were in credit before their second service. Over a fifteen-year life the two of them avoid around 90 tonnes of CO₂e against the petrol cars they replaced.

Chapter fourThe finding that embarrassed us

Here is where it stops being a press release.

Both cars charge overnight on a smart tariff, 23:30 to 05:30, which is the thing everyone tells you to do. So we pulled the actual half-hourly grid data for both charging regions across four sample windows spanning a year, and measured what that specific window really costs.

The Seal charges in southern Scotland. Its overnight window averages 12 gCO₂/kWh. Ninety per cent of those half-hours sit in the grid operator's "very low" band. The mix is 47% wind and 44% nuclear, with about 2% gas. A full year of driving that car, 12,000 miles and some 3,700 kilowatt-hours, carries about 45 kg of CO₂. The petrol car it replaced would get through that in four and a half days.

The EQB charges in southern England. Its overnight window averages 240 gCO₂/kWh, and it is 4.5% worse than that same region's round-the-clock average.

Read that again, because I had to. Charging overnight, on a smart tariff, made the English car marginally dirtier than charging at random. The reason is simple once you see it: southern England has enough solar to pull its daytime intensity down, and solar isn't there at two in the morning. What's left is 56% gas and 23% imports. Not one overnight half-hour in our entire sample came in below 96 gCO₂/kWh.

I'm publishing that because it's what the ledger said. It also happens to be the single most useful thing we learned: "charge overnight" is folk wisdom, not analysis. The clean hours move by region, by season, by weather. A timer set at 23:30 is a guess. Reading the grid is a strategy.

And for anyone siting energy-intensive infrastructure, whether data centres, AI capacity or anything with a big continuous load, that twenty-to-one regional spread is not a rounding error. It's the whole decision. Which is exactly why our platform is being placed where the electrons are clean rather than where the property is convenient. Where you build matters more than what you build.

Chapter fiveThe asymmetry nobody prices

Now the part that no analysis of this kind ever seems to capture.

An electric car's operating carbon falls over the life of the asset. UK grid intensity has gone from just under 500 gCO₂/kWh in 2012 to 126 in 2025. The same journey in the Seal will cost less carbon in 2032 than it does today, and we won't touch the car.

A diesel's per-mile carbon is fixed on the day it's built, and drifts the wrong way as the engine ages. We bought a depreciating liability. The alternative was a fixed one.

I'll be straight about the wrinkle: 2025 actually ticked up about 2%, and gas generation rose. The decade-long trend is not in doubt; the line is not smooth, and last year was a poor one. Anyone claiming otherwise is doing the same selective thing I'm complaining about.

Chapter sixThey compare our CAPEX to their OPEX

Petrol has a supply chain too. Unlike a battery, it gets rebuilt with every tank.

DEFRA splits every fuel factor into combustion and well-to-tank: extraction, refining, distribution. For petrol that upstream portion is about 22% of the fuel cycle on top of what comes out of the exhaust. Meaningful, and routinely absent from the comparison.

Chapter sevenNow the strongest argument against me

Here's the objection I'd raise if I were on the other side, and it's a good one.

I've been costing our charging at the average intensity of the grid. But plugging in a car is new demand, and new demand is usually met by turning something up, typically gas. On that basis the honest number is the marginal rate, around 375 gCO₂/kWh, not the average. It's a legitimate criticism and it cuts directly against my figures.

So I ran it. At the marginal rate, break-even moves to 23,100 to 30,700 miles for the Seal and 19,500 to 26,200 for the EQB. Roughly double my headline figure. Two and a half years instead of eighteen months.

And the conclusion doesn't move an inch.

Chapter eightWhat I'm not claiming

Battery manufacturing has a real footprint and I haven't waved it away. Fifteen-year degradation and second-life assumptions carry genuine uncertainty. The published range for battery carbon is wide enough to drive a lorry through. The EQB's consumption is our own observed figure; the Seal's is still a published real-world estimate rather than our meter, and that may move. And the timing argument only works if you charge when you said you would. Claimed behaviour is worthless; metered behaviour isn't.

None of that changes the direction of travel. The case survives every one of those caveats. It just doesn't need them hidden.

In closingWhich column were you adding?

Carbon, like money, has a balance sheet and a profit-and-loss. Judging an asset on its opening entry alone isn't scepticism, it's incompetence. Ask what you'd ask of any capital purchase: what did it cost to acquire, what does it cost to run, and how long until it pays back?

For our two cars, the answer was somewhere between fourteen and twenty-two months. Then fifteen years of profit.

So the next time someone tells you the sums don't add up on an electric vehicle, ask them a simple question: which column were you adding?