A plasma source for a pulsed fusion concept is usually built once, by hand, at a cost nobody wants to see written down. A power plant needs many of them, and it needs a fleet of plants. The distance between those two facts is where Wright’s Law applies, and it is the part of fusion costing that gets the least engineering attention. Most of what determines how fast unit cost falls is decided before the second unit exists.
The rule, and what it assumes
In 1936 T. P. Wright observed that each time cumulative aircraft production doubled, the labor needed to build the next airframe fell by about 20% [1]. Written as an equation, unit cost after N units is C(N) = C1 · N^b, with b = log2(1 − L), where L is the learning rate per doubling. A 15% rate gives b = −0.2345, which means the 1,000th unit costs about 20% of the first, and the 10,000th about 12%.
Reported learning rates vary widely by activity. Progress ratios compiled from NASA data put repetitive electrical operations at 15–25% per doubling, purchased parts at 12–15%, repetitive welding at about 10%, and raw materials at 4–7% [2]. Those ranges say something useful before any fusion data exists: the steps in a plasma source that are repeated labor learn quickly, and the steps that are material do not.
Two assumptions sit under the equation. The first is that the same design is being repeated. If every unit is a revision, the cumulative count that matters is the count of the current revision, not of the product line. The second is that a share of cost has nothing to learn. A cost with a fixed floor F follows C(N) = C1 · [F + (1 − F) · N^b]. The floor is where plasma sources differ from airframes.
What the floor is made of
A plasma source in a pulsed power plant has an electrical side and a mechanical side. The electrical side is the energy storage, switching and conductors that deliver the pulse. The mechanical side is the electrodes or coils, insulators, vacuum interface and structure that survive it. In the account structure I use, electrodes and plasma guns sit in Account 22.1.8 and the pulsed power that drives them in Account 22.1.3 [4].
The floor is the physics. Stored energy needs dielectric volume. Conducting a given current for a given pulse needs conductor cross section. Electrode erosion sets a consumable mass per shot. Learning trims the labor and margin around those quantities and does not trim the quantities. If the FOAK source has a floor of half its cost, the 15% curve takes the 1,000th unit to roughly 60% of the first, not 20%. If the floor is 30%, it reaches roughly 44%. These floor fractions are assumptions chosen to show sensitivity, not data on any particular device. The sensitivity is large. A 10% learning rate with a 30% floor and a 20% rate with a 50% floor both leave the 1,000th unit at about 55% of the first, so two quite different sets of assumptions give the same answer, and a single NOAK number cannot tell you which one you are looking at.
What has to be designed in on day one
Four decisions determine how much of the cost sits above the floor and how much of it is repeatable.
The repeating unit. Learning counts repetitions of one design. A source built as a standard module that is replicated many times accumulates experience each time one is built. A source built as a single large assembly accumulates experience once. The same total energy can be delivered either way, and the cost curves differ.
A frozen interface. A module can only be repeated if its mechanical, electrical and control interfaces stay fixed while its insides improve. Interface changes reset the count. I treat interface freeze as the first design review item for any component expected to exist in quantity.
Part count and process capability. Every hand-fitted joint, hand-wound coil and one-off tolerance is a place where labor does not fall with volume until a tool or process replaces it. A design whose critical dimensions can be held by an existing production process starts on the steep part of the curve. A design that needs a new process inherits that process’s own development curve first.
Replaceability. Electrodes and other wear parts are bought repeatedly over the life of the plant. They benefit from the production count of the whole fleet, and the operating cost they carry falls with it. A source designed so that wear parts swap without disturbing the rest of the assembly turns a maintenance item into a high-volume product.
From the source to the plant
The plant-level effect depends on how many sources a plant contains and how many plants are built, and both depend on the concept. A concept that uses a small number of large sources has few repetitions per plant and relies on fleet count. A concept with many small sources reaches high cumulative volume inside a single plant. Neither is better in the abstract. Each shifts where the learning shows up and how soon.
This is why I treat the learning rate as an uncertain input and not a constant. In the extension of the costing standard, materials price, technology maturity and learning curve uncertainty are compounded into cost distributions, not a single deterministic estimate [3]. A plasma source estimate that quotes one NOAK number without a floor fraction and a learning-rate range is hiding the two parameters that decide it.
Practical takeaway
If you are designing a source, write down the repeating unit, the frozen interface and the part that will be bought most often, and ask what each of them does to cumulative count. If you are reviewing a cost estimate, ask for the floor fraction and the learning rate separately, and for the evidence behind each. A NOAK number that cannot be decomposed that way is a FOAK number with a hopeful multiplier.
If you want to work through the floor and learning rate for a specific pulsed power or magnet design, get in touch through woodruffeng.com.
References
[1] T. P. Wright, “Factors Affecting the Cost of Airplanes,” Journal of the Aeronautical Sciences 3(4), 122–128 (1936). https://doi.org/10.2514/8.155
[2] “Experience curve effect,” Wikipedia, summarizing NASA progress ratios and other empirical learning rates. https://en.wikipedia.org/wiki/Experience_curve_effect
[3] S. Woodruff, A. Durham, A. Higginbottom, C. Raastad, “Extension of the fusion power plant costing standard,” arXiv:2602.19389 (2026). https://arxiv.org/abs/2602.19389
[4] S. Woodruff, “A costing framework for fusion power plants,” arXiv:2601.21724 (2026). https://arxiv.org/abs/2601.21724
Also published on LinkedIn: https://www.linkedin.com/pulse/wrights-law-pulsed-power-designing-plasma-source-cost-simon-woodruff-ba9me/
By Dr. Simon Woodruff, CEO — Woodruff Engineering Inc.