The ramp: why year one looks nothing like year three

Most charging pro formas open the doors at full volume. Month one looks like month sixty: the same kilowatt-hours, the same revenue, the same margin, copied down a column. Real sites do not do that. They open to a trickle, build as drivers learn the stop exists, and settle somewhere a year or two later.

The gap between those two pictures is the single largest thing a model can get wrong, and it is wrong in a specific direction. This post walks through what the ramp input is, what the three calibrated curves the engine ships actually look like, and what each one does to the same eight-stall site over fifteen years. Nothing else changes between the three runs.

Three curves, one input#

The ramp is a twelve-month path of utilization, in kilowatt-hours dispensed per stall per day. The engine ships three presets and a custom option:

Month one runs for fifteen operating days rather than thirty, since sites rarely open on the first of the month. From year two the engine grows the month-twelve level by 5% a year through year seven and 3% a year after that, capped at 490 kWh per stall per day, which is the most the hardware can physically deliver.

Everything below is the default eight-stall site: PG&E's BEV-2 rate, $0.45 per kWh retail, $1.21M of capex with 20% down and the rest on a ten-year loan at 8%. The only input that varies is the curve.

What the curve does to year one#

Net operating cash here is everything that leaves the account: revenue less electricity, Tesla's network fee, rent, insurance, entity fees and the loan payment. All three curves lose money in year one on this site. That is expected for a leveraged asset in its ramp year, and it is the reason the model carries a peak-cash-deficit line rather than assuming the site funds itself from day one.

Notice the electricity row. The site pays $27,521 a year for its demand subscription regardless of how many cars arrive, so the effective cost per kWh falls as volume rises: 31.3 cents on the slow curve, 27.8 cents on the premium one. The tariff did not change. The denominator did. The loss-factor post and what a flat blended rate hides cover the same mechanism from the other side.

Year one, month by month#

The Standard curve, in dollars:

Month one carries the one-time costs and half a month of sales. Months three, six, nine and twelve are lighter because that is when the quarterly LCFS cash arrives. By December the site is dispensing five times what it did in its opening fortnight and is close to covering its monthly outgoings, but it has absorbed $158,515 getting there. That figure is the working capital the ramp consumes, and a day-one budget that stops at capex has not funded it.

The ramp is a level, not a delay#

Here is the part that surprises people. A ramp is usually described as a delay: the site gets to the same place, just later. In the model it is not a delay. Years two onward grow from wherever year one finished, so the three curves never converge.

The Premium site reaches the 490 kWh hardware cap in year fifteen. The Slow-ramp site ends the horizon dispensing less than the Premium site did in its first year. Year three is not year one with the kinks worked out. It is year one, grown.

That is a modelling choice, and it is stated as one. The engine assumes the first-year level reflects the site's catchment, which is the thing that persists, rather than a temporary awareness gap. A site that genuinely opens slowly and then catches up to a Premium level is a Custom curve, and the wizard takes one.

What fifteen years of that looks like#

One input. Payback moves from under twelve years to never. NPV swings by $1.26M between the top and bottom curves. The spread between the Premium and Standard sites alone, $630,943 of NPV, is larger than the effect of most of the inputs people spend their diligence on.

The returns pillar, is hosting a Supercharger profitable, sets out how the headline metrics are built, and payback vs NPV vs IRR explains why the three columns above can rank sites differently.

Choosing a curve#

The curves are calibrated shapes, not forecasts for any particular address. What they encode is a first-year level and a build pattern, and the thing to check is whether that level is plausible for the catchment: the traffic passing the site, the dwell time the surrounding use gives drivers, the nearest existing Supercharger and how full it runs. Tesla publishes network average utilization quarterly, which gives a reference for where a mature site sits.

The engine also reports the reverse: for the Standard site at $0.45, year two utilization has to reach 281 kWh per stall per day before the engine's internal viability test, a composite of payback, cash position and IRR, stops flagging the site, against the 227 the curve assumes. That gap is the honest summary of the site, and it is the number a ramp assumption either closes or does not.

To see what the three curves do to a real address and tariff, the scenario wizard runs all three. The methodology page documents the growth and cap rules used here.

Engine outputs are model projections under the stated assumptions, not advice or guarantees. The three curves are the engine's shipped presets; the site, tariff and prices are its documented defaults.