Five-Year EV Charging TCO: The Cost per kWh Delivered

A procurement file usually carries one circled number: the price per unit. Five years later, that number explains very little of what the site actually cost. Two operators can install the same model, bought at the same price, and end up with materially different cost per kilowatt-hour delivered — because most of the cost is created by how the site is used, not by what was bought.

The argument here is not that equipment should be bought cheaply, or that it should be bought expensively. It is that the purchase price belongs in its proper place — one line in a model of six — and that the number deciding the investment is cost per kWh delivered at this site's load factor, not price per kilowatt of nameplate rating.

The Purchase Price Answers a Question Nobody Is Asking

"What does the charger cost" is easy to answer and easy to compare, which is exactly why it dominates tender evaluation. The question that decides whether the site earns its capital back is different: what does a kilowatt-hour delivered from this site cost, across the whole asset life, including everything that is not on the quotation?

Hardware is the only line a supplier quotes. Energy, software, maintenance, downtime and end-of-life are estimated by the buyer, inherited from a previous site, or simply left out. When they are left out, the comparison silently assumes they are identical between options — and they rarely are.

A discount on the machine changes the smallest part of the answer on a low-traffic site, and a meaningful part on a busy one. That asymmetry is the whole subject of this article.

Six Cost Lines That Belong in the Model

The model below is deliberately plain. Each line behaves differently over time, which is why collapsing them into one "total cost" figure hides the decision.

Cost line	What sits inside it	How it behaves
Capital	Hardware, civil works, cabling, protection devices, grid connection or transformer upgrade, commissioning	One-time; the only line a supplier quotes, and often only part of it
Energy	Commodity cost, network charges, and losses through transformer, cabinet, cable and connector	Scales with kWh delivered, plus a separate part driven by the monthly peak
Software and payments	Back-office subscription, connectivity, payment processing charged as a share of transaction value	Partly fixed per charger, partly per transaction; renews annually
Maintenance and spares	Preventive visits, filters and consumables, power module replacement, connector and cable replacement	Steps up with age and with throughput, not smoothly with time
Downtime	Revenue not earned while a unit is unavailable, plus the drivers who do not return after a failed session	Absent from accounting, present in the result; the widest variance between sites
End of life	Removal, disposal, site reinstatement, and whatever residual value the asset still holds	One event at the end, routinely omitted at the start

Two of these six lines are decided before the equipment is chosen: the grid connection and the tariff structure. They are also the two that buyers most often treat as someone else's budget. A site with a constrained connection can spend more on that constraint than on every charger on the plot.

The Formula, and Why the Denominator Decides the Answer

Reduce the six lines to one ratio and the structure of the problem becomes visible:

Every term above the line is a number in currency, and every one of them can be negotiated, benchmarked or squeezed. Every term below the line is a property of the site. Buyers spend their negotiation time on the numerator, but the denominator usually moves the result further.

Load factor here means average delivered power divided by rated power, taken across every hour of the year. It is not the share of busy hours, and it is not the rate observed during a good session. A charger rated at a given power that delivers a fraction of it on average has that fraction built into its cost of energy, whether or not anyone writes it down.

Utilization Sets the Floor, Not the Equipment

Halve the load factor and the cost per kWh roughly doubles, with the hardware unchanged and the tariff unchanged. This is the single largest effect in the model, and it is the reason two quotations cannot be compared at a supplier's reference site.

It also produces a conclusion that sits awkwardly with procurement instincts: on a low-utilization site, the cheapest capable hardware is often the rational choice, because the site cannot earn back a difference in efficiency or build quality that only shows at throughput. On a high-utilization site the same reasoning reverses. The equipment question has no answer until the site's duty cycle is known.

One practical consequence: before comparing options, measure delivered power against requested power on an existing unit, or ask for the derating behaviour of a candidate. A unit whose delivered output collapses in heat, at low state of charge, or when two connectors share a cabinet has a lower effective rating than its nameplate, and the denominator above reflects that. The published curves for a 120 kW and 180 kW DC station are a useful reference for what that curve looks like, whatever brand is finally specified.

Demand Charges Follow the Peak, Not the Volume

Most commercial tariffs price energy in two parts: a rate per kWh, and a charge based on the highest demand recorded in the billing period. The second part does not care how many sessions the site served. It cares how much power was drawn at the worst moment.

That distinction changes the hardware decision. A site can deliver more energy across a day without raising its peak, if power is shared between connectors rather than summed:

Design choice	Effect on the site peak	Effect on energy delivered
Fixed split between connectors	Adds up to the sum of the connectors when both are busy	Each connector capped by its own allocation, busy or not
Dynamic sharing across a cabinet	Held near the cabinet rating, not the sum of outlets	Largely preserved across the day; each session slower when both are active
Buffer storage at the site	Clipped, with the shortfall drawn from the battery	Throughput preserved, with round-trip losses added

A 480 kW charging hub host is one way to hold the peak while still offering high power per session; a PV and storage integrated system attacks the same charge from the supply side. Both add capital, and both reduce a recurring cost — which is why they belong in the same model rather than in separate budgets.

The Break-Even Load Factor Belongs in the Investment Memo

Payback period is the usual headline, and it is a poor one: it depends on a revenue assumption that is rarely tested. A more honest figure is the load factor at which the site stops losing money — the point where revenue per kWh equals cost per kWh.

The output is a single number that can be checked against reality. If the site's measured load factor sits below it, the project does not fail in year five; it fails in year one, and no equipment choice rescues it — the constraint is demand, not hardware. If the measured figure sits comfortably above it, the remaining question is how much capital the margin supports.

This framing also settles an argument that runs through most tender reviews: whether to spend more per unit for efficiency, redundancy or service. That extra capital moves the numerator by a known amount; it only pays if the site's load factor is high enough for the savings to accumulate. On a quiet site the arithmetic says no, and saying so early is more useful than defending a premium on principle.

Four Inputs to Measure, Not Assume

The model is only as good as the numbers fed into it, and four of them are routinely guessed:

• The tariff structure, not the average rate. Energy rate, demand charge, and time bands all matter, and one year of bills gives the real curve rather than a quoted headline figure.
• The arrival pattern by hour and by weekday, taken from a network log rather than from a traffic count. What matters is simultaneous occupancy, not the number of cars passing the site.
• Energy per session and dwell time. These two set how long each bay is occupied, and therefore how many sessions the site can serve at its busy hour.
• Delivered power against requested power, from charger logs. A datasheet states the ceiling; the log states what the site actually receives.

For a depot or a return-to-base operation, the arrival pattern is largely fixed by the shift schedule, which makes the model far more reliable than at a public site. The fleet hub charging layout is a useful reference for how overnight dwell time and morning departure compress the required power into a short window.

Where Five-Year Models Usually Break

Six assumptions account for most of the error in these models. Each is easy to correct once it is named:

Assumption	What it does to the result
Nameplate power used as delivered power	Overstates energy in the denominator, so cost per kWh looks better than the site can achieve
Year-one tariffs and software fees held flat	Understates the recurring lines; renewal terms in the contract, not the introductory price, belong here
Module and connector replacement treated as warranty	Moves a predictable cost out of the model and into an unbudgeted event
Downtime counted as zero	Removes the line with the widest variance between good and poor sites
Grid upgrade parked in another budget	Hides the largest single capital item on constrained sites
Residual value left unstated	Either zero or optimistic, and neither is wrong — but the choice belongs on the page, not in a footnote

The same discipline applies to the documents behind the numbers. Efficiency figures, derating curves and module counts sit in the technical documentation, and the product documentation library is a reasonable place to see what a complete data set looks like before asking suppliers for theirs.

What the Five-Year Number Actually Tells You

It is not a property of the charger. It is a property of the pairing between a charger and a site — which is why the useful question in a tender is not "what is the price per kilowatt" but "at what load factor does this design stop making sense".

Two columns can be built from that question. In the first, the site parameters against the measured curve. In the second, the same parameters against a cheaper unit with a lower effective rating. If the two columns converge, the saving is real. If they diverge, the difference is being paid for monthly, in energy and in downtime, long after the invoice has been filed.

Whichever direction the numbers point, the model has done its job: it has turned a price into a decision, and moved the argument from preference to arithmetic.