fuelfleet

A fusion fuel-logistics and deployment-economics model: how much fusion capacity can physically and economically exist by year X, given real fuel constraints?

Existing public tools are plant-level fuel-cycle simulators (TMAP8, FESTIM, FFCSim) or static supply estimates (Kovari 2018, Abdou 2021). Nothing public couples global tritium/He-3 supply dynamics, fleet deployment scheduling, and learning-rate economics. fuelfleet is that fleet/market layer: a 0-D systems-analysis tool at annual/monthly resolution, not a plasma or blanket physics code. Every equation, unit, default, and validation anchor is documented in docs/model.md; the full source is on GitHub under an MIT license.

fuelfleet is a requirements calculator, not a doom calculator. The tritium constraint on D-T fusion is a bet on undemonstrated fuel-cycle performance, and the model quantifies where between two poles any given fuel-cycle demonstration moves the industry. That turns facilities like LIBRTI, H3AT, and UNITY-2 into measurable deployment levers. Neither pole is a forecast: the design-target scenario is not fact, and the demonstrated bound is not destiny.

  1. The tritium question is "which fuel cycle", not "how much tritium". With demonstrated (ITER-class) parameters (1% burn-up where ITER's own projection is 0.36%, 24 h processing, no direct internal recycling, trapping penalty on), the startup inventory is 12.2 kg. The civilian stockpile funds ~3 plant startups, and D-T holds ≈0.8 GWe in 2050: a demonstration technology through mid-century. At published design targets (5% burn-up, 6 h slow loop, 75% DIR, TBR 1.10, trapping on) the startup inventory is 0.75 kg, tritium never blocks a single build, and capacity rides the industrial ramp envelope. No device has ever operated at burn-up ≥2% with DIR ≥50%. SRNL calculates 327 g for a comparable design, and Parisi's tritium-lean concept claims under 0.01 kg (claimed, undemonstrated).
  2. A ~1–3.5 kg startup-inventory threshold separates the poles. Under the 2026 supply outlook with builds from 2027, fleets stay pinned to the ramp-limited envelope (26.4 GWe in 2050) up to I0 ≈ 1.0 kg at TBR 1.02, ≈ 2.1 kg at TBR 1.05, and ≈ 3.5 kg at TBR 1.10. Past those points they fall off it. Published design targets sit inside or below this band; the demonstrated bound sits far outside it. Above ~1.05, raising TBR buys almost nothing once inventories are small. Shrinking I0 buys more capacity than raising TBR.
  3. The supply outlook is flat-to-rising 2–3 kg/yr, not collapsing. Ontario CANDU refurbishments hold ~1.9 kg/yr into the 2040s, and Cernavoda's tritium removal facility (under construction) adds ~0.26 kg/yr plus a stored-moderator drawdown from 2030. With ITER's D-T phase deferred to 2039, the civilian pool grows through the 2030s and peaks near 35 kg with no fusion fleet, inside UKAEA's published range. The pre-2026 declining curve is kept as a legacy toggle below.
  4. The learning-rate prior, not physics, decides "competitive". Identical plants under an empirical megaproject learning prior (2–8%/doubling) give NOAK LCOE P50 ≈ $130/MWh and P(<$100/MWh) ≈ 18%. Under fusion-company guidance (8–20%) the same hardware gives P50 ≈ $97 and P(<$100) ≈ 55%. The SMR comparator lands at P50 ≈ $134; fusion FOAK with no learning at ≈ $153.

The explorer below moves through 9,072 precomputed runs of the model. The demonstrated-technology bound is the default; the design-target and tritium-lean families are one preset away, always shown against the same envelope.