Making Availability and Economics Physical

Abstract

Fusion economics is often reduced to machine size or projected cost per kilowatt. I instead begin with the physical causes of delivered energy and cash flow. Net electrical output depends on thermal conversion and internal power. Annual generation depends on planned and unplanned availability. Capital cost depends on material quantity, manufacturing complexity, buildings, nuclear systems, construction duration, and financing. Replacement cost depends on neutron life and remote outage design. I develop a transparent model that preserves uncertainty and separates first-of-a-kind demonstration value from mature fleet economics. The most important variables are not isolated plasma gain or field strength. They are net power, component lifetime, replacement duration, construction schedule, and the rate at which factories and operators learn.

Cost begins with delivered megawatt-hours

The levelized cost of electricity is schematically

$$\mathrm{LCOE}=\frac{\sum_t(C_{\mathrm{capital},t}+C_{\mathrm{O\&M},t}+C_{\mathrm{fuel},t}+C_{\mathrm{decom},t})/(1+r)^t}{\sum_tE_t/(1+r)^t}.$$

The formula is familiar. The difficult part is defending each term for a plant that does not yet exist. A precise output from uncertain inputs is not reliable knowledge.

I therefore report ranges and sensitivities. Every assumption carries an evidence class: measured in relevant service, measured in a partial environment, extrapolated by validated model, analogous industry data, or design target.

Net power is the denominator before availability

Gross generator rating can conceal large internal loads. I retain

$$P_{\mathrm{net}}=\eta_{\mathrm{th}}P_{\mathrm{recoverable}}-P_{\mathrm{recirc}}.$$

Current drive, neutral beams, radiofrequency sources, cryogenics, coolant pumps, vacuum, tritium processing, magnets, heat rejection, remote systems, and buildings consume power. Some loads are continuous; others peak during startup or maintenance.

A plant producing $800\ \mathrm{MW_e}$ gross with $300\ \mathrm{MW_e}$ internal load sells $500\ \mathrm{MW_e}$. A ten-percent reduction in gross power is serious, but a $100\ \mathrm{MW_e}$ rise in recirculating load reduces saleable output by twenty percent. Wall-plug efficiency deserves plant-level prominence.

Availability multiplies every capital dollar

Availability can be written as

$$A=\frac{T_{\mathrm{scheduled}}-T_{\mathrm{planned}}-T_{\mathrm{forced}}}{T_{\mathrm{scheduled}}}.$$

Planned outages replace blankets, divertors, pumps, diagnostics, and other life-limited equipment. Forced outages follow faults. Their durations depend on diagnosis, cooldown, access, spares, remote work, testing, and restart.

For repairable equipment, a simple steady estimate is

$$A\approx\frac{\mathrm{MTBF}}{\mathrm{MTBF}+\mathrm{MTTR}}.$$

Fusion is not one component, and this equation omits redundancy, common cause, planned life, and resource contention. It still shows that reducing mean time to repair can matter as much as extending mean time between failures.

Variable	Physical evidence needed	Economic effect
Divertor and blanket life	Integrated heat and neutron exposure	Planned outage frequency and replacement inventory
Sector exchange duration	Full-scale rehearsed remote sequence	Outage length and hot-cell capacity
Heating efficiency	Grid-to-absorbed-power measurement	Net output and equipment size
Magnet reliability	Population, joint, and quench evidence	Forced outage and major capital risk
Construction schedule	Completed design, supply, and installation work	Interest during construction and revenue delay

Component life creates a size trade

Higher neutron wall loading can produce more power from a smaller surface, but it consumes material life faster. A larger machine may cost more initially while reducing load density and extending replacement intervals.

If allowable lifetime fluence is $F_{\max}$ and wall loading is $q_n$, full-power life is approximately

$$t_{\mathrm{life}}=\frac{F_{\max}}{q_n}.$$

Real life also depends on temperature, stress, transmutation, joints, fatigue, and transients. The relation reveals why maximum power density is not automatically minimum cost. Plant systems codes must allow size and maintenance architecture to change together.

Construction time is a technology variable

Large projects incur financing costs before earning revenue. Delay also freezes design, ties up skilled teams, and exposes interfaces to rework. A compact machine with difficult assembly can take longer than a larger modular one.

I separate design completion, licensing, long-lead procurement, civil works, factory production, site assembly, commissioning, and contingency. Schedule logic includes resource limits and rework, not only ideal precedence.

Modular construction can shift work to controlled factories and permit parallelism. It can also multiply transport, alignment, welding, inspection, and configuration interfaces. The economic model credits modularity only after a detailed installation sequence demonstrates its benefit.

First plants and mature plants answer different questions

A first-of-a-kind plant carries engineering, qualification, licensing, supply-chain, tooling, and commissioning costs that later units should not repeat. It may rationally cost more than the energy it produces because its product includes evidence.

Nth-of-a-kind projections use learning curves:

$$C_N=C_1N^b, \qquad b=\frac{\ln(1-L)}{\ln2},$$

where $L$ is the learning rate per cumulative doubling. Learning is not automatic. It requires design stability, repeated suppliers, feedback, workforce continuity, and enough units. A fleet of unique national projects does not experience the same learning as a production line.

Materials and waste belong in the ledger

Fusion fuel mass is small, but the plant contains steel, concrete, copper, superconductors, tungsten, lithium, beryllium or lead in some blankets, cryogenic equipment, turbines, and power electronics. Replacement sectors create activated material streams.

Low-activation material choices can reduce long-term waste burden, yet manufacturing yield, recycling, clearance rules, decontamination, and disposal routes need real cost. Decommissioning funds and component storage begin with plant operation.

The supply of enriched lithium, tritium, specialty alloys, superconducting tape, high-purity tungsten, large forgings, and qualified electronics may constrain build rate before raw resource quantity does. Factory capacity is an economic state variable.

Value depends on the energy system

Fusion may provide firm electricity, process heat, reserve, ramping, hydrogen, or thermal storage. Revenue should be modeled against the grid it joins rather than an undifferentiated energy price.

A high-capital plant benefits from high utilization, but a future grid with abundant variable generation may value flexible output. Thermal storage can keep the fusion core steady while shifting electricity. The added tanks and exchangers compete with improved market value.

OutputMaximize saleable energy

Close thermal efficiency, internal power, duty, and grid service.

UptimeDesign physical availability

Connect component life, remote work, spares, and recovery distributions.

DeliveryControl capital and schedule

Use completed design, factories, modular interfaces, and reference classes.

LearningReproduce a stable plant family

Convert first-unit evidence into manufacturing and operating improvement.

A decision model should be falsifiable

I maintain a ranked list of value-of-information questions. If divertor life dominates cost uncertainty, material and exhaust testing may be worth more than a small confinement improvement. If construction schedule dominates, a full-scale assembly demonstrator may outrank another conceptual optimization.

AssumeDeclare range and evidence class PropagateCalculate output, cost, and uncertainty TestFund evidence with greatest decision value UpdateRevise design and investment gate

The model cannot be validated by analysis alone. Construction and commissioning reveal whether tolerances, supply chains, test boundaries, software, procedures, and organizations actually form an operable plant. The next report designs that transition from completed parts to measured net power.