When Faster Power Stops Being Expensive

Someone offers you power two years early and wants to be paid for it. The usual advice is to run the net present value and pick the winner.

The winner moves.

What you gain from arriving early is the months you buy times what a megawatt-hour of your load earns before electricity. What you pay for those months is a fixed figure that does not change when your margin does. So the ranking is not a property of the projects. It is a property of your economics.

Three regimes

On the illustrative case in the model below, the answer changes twice.

Below about $49 per MWh, nothing clears. Not the cheap options, all four. The right answer is to wait for the grid, and negotiating the price will not fix it, because the problem is that the load does not earn enough to justify any price.

Between about $49 and $114, the capital-heavy path wins. An accelerated network upgrade arrives eight months later than bridge power and still ranks first. Bridge power clears comfortably here, at $51mm on the default inputs, but it trails, because it pays per megawatt-hour delivered early and puts capital in. It buys the same speed twice.

Above about $114, that reverses. Bridge power takes first place and keeps it, and the gap widens with every dollar of margin.

Most people carry a rule that a recurring charge is worse than one-time capital, since it compounds with volume and duration. That rule holds in the middle band and breaks in the top one.

Why bridge power ever wins

At the default inputs a phased ramp beats it outright: same eighteen months earlier, better net present value, lower cost per month. So why does bridge power take over at the top?

Because a phased ramp gives up early megawatts to save money. Bridge power buys all 250 of them from month eighteen. When margin is thin, the saving matters more. When margin is fat, the volume does, and it compounds faster than the charge it carries.

Find where your own answer changes

The number to take into the room

Net present value ranks the options. It does not tell you the rate you are paying for speed.

Divide each path's total cost, capital plus every recurring dollar, by the months it buys. Phased $4.6mm a month, accelerated upgrade $6.5mm, bridge $11.3mm, co-located $14.0mm.

Bridge power buys the most months at the worst rate. That sentence survives an argument about discount rates.

What to do

Find your crossovers before the meeting. Knowing the level at which you walk away entirely is worth more than price discovery.

Above the top crossover, stop optimizing the charge. The calendar decides the outcome there, not the rate. Time spent grinding a supplier is time not spent pulling the date forward.

Below the bottom one, say so. Sometimes the finding is that the project should energize on the grid's schedule and the money should go elsewhere. That is unpopular and often right.

This is where an industrial process and a data center part company. Electricity is a large share of conversion cost for an electro-intensive process, so its margin per megawatt-hour sits near the bottom crossover. Revenue per MW of IT load is far higher, which puts a data center near the top. The two should follow opposite rules when offered faster power, and usually do not.

Methodology

Each path is valued against the grid-only baseline over 60 months at 730 hours a month, capital charged at month zero, discounting at (1 + r) to the power t over twelve rather than the nominal rate divided by twelve. Those differ by about a million dollars on the bridge case.

The load is valued before electricity cost, and each path's charge is the increment above the baseline grid price rather than the whole bill. That is what stops the same megawatt-hour being counted twice when a temporary path is compared against the eventual grid source. Availability derates reduce effective energized megawatts.

What generalizes and what does not. The $49 and $114 belong to this scenario set and move when the inputs do. What generalizes is that the preferred path can change as load value rises. How many crossovers there are, and where they sit, depends on the cost, timing and ramp structure of the alternatives actually available to you. The model computes them on whatever you enter, which is the only honest way to publish a number like this.

One limit to note. Cost per month treats a small load brought forward the same as a large one. Cost per accelerated MW-month is the better denominator and is not implemented here.

Defaults are illustrative and drawn from no specific project. Nothing entered is transmitted or stored.

Context: the Department of Energy launched Speed to Power in September 2025, FERC acted in June 2026 to speed large load integration including co-located and electrically proximate generation, Berkeley Lab counted about 8,200 generation and storage projects seeking interconnection at the end of 2025, and EIA put the industrial average retail price at 8.71 cents per kWh in May 2026.

All analysis by Tafel Power from public sources.