Costing a power station that has never existed, NASA's SBSP methodology

What the work claims

This is a methodology report, not a results paper; the authors state explicitly that study findings were still under review and would be published later. What it delivers is the measurement apparatus for deciding whether space-based solar power (SBSP) is feasible as an energy source, defined on three axes at once: technical (does the technology exist to build it), financial (how does its levelized cost compare with alternatives), and environmental (how does its carbon compare). The comparison baseline is deliberately demanding: SBSP must supply 1 GW of baseload power, 24 GWh per day, continuously, to receiving sites in two contrasting US locations, San Antonio, Texas (29.4°N, high solar insolation) and Philadelphia, Pennsylvania (40°N, much less), against coal, natural gas, fission, wind, hydroelectric, terrestrial PV, and solar with storage, using 2050 projections.¹

Two existing public designs were chosen as the costed subjects precisely because they are public and scrutable: SPS-ALPHA Mark III, John Mankins' hyper-modular geostationary concept of millions of loosely structured heliostats feeding a central conversion and transmission module, a structure the paper describes as spanning more than 5 km and requiring in-orbit assembly; and the JAXA SPS-Tethered design, first published in 2006, a roughly 2 km-per-side planar array with nadir-facing microwave emitters stabilized by tension tethers, whose output scales directly with area.¹

How it works

The cost engine is an independent study by the Aerospace Corporation. Every work-breakdown-structure element of the two SBSP designs got a bottoms-up estimate using cost-estimating relationships derived from analogies to other space systems, then a cost-risk treatment: the F-RISK methodology, a mathematical approximation to Monte Carlo, assigning probability distributions to uncertain line items, applying a Design Maturity Factor keyed to how mature each technology is, and defining unallocated future expense as the gap between the 70th-percentile and most-likely estimates. The historical anchors are NASA's CADRe database and Aerospace's own historical database; a 30% development reserve is spread across all elements, and every system is assumed to live 30 years with a built-in maintenance and refurbishment cycle.¹

The learning-curve structure is the methodological heart. The first-of-a-kind (FOAK) estimate starts from what is technologically available today or very nearly, assumes limited learning, and is explicitly labeled a conservative baseline, not a proposed cost. The nth-of-a-kind (NOAK) estimate layers scenarios for learning: longer-life components, deeper learning curves, retirement of technology-development costs, and commercial development practices. The final NOAK is computed by summing the FOAK and 100 final marginal NOAK units and dividing by 101, so the fleet-average number embeds the entire learning history. Where no data exists, the method says so: commercial in-space servicing and tugs were considered and dropped for insufficient historical data, while orbital assembly costs were anchored on Northrop Grumman and Aerojet Rocketdyne analogies, explicitly including the Mission Extension Vehicle's development and marginal costs.¹

On the carbon side, the team used an energy-content method: estimate each input material's mass fraction, convert to the energy required to produce it using published materials data, total the megajoules, and convert to tons of carbon for a lifecycle carbon intensity in tons per MWh. The worked calibration in the paper is a 660 MW coal plant requiring about 1.765 x 10^10 MJ to build, or 571,672.86 metric tons of carbon, which amortized over a 30-year life plus operating emissions yields 2.189 tons of carbon per MWh. Two accounting choices matter for SBSP: launch is included for the space system (because the terrestrial transportation emissions behind launch vehicles are excluded, keeping equivalence), and end-of-life disposal is included, which the paper notes is rarely counted in terrestrial carbon assessments. The authors flag their own undercount: the method excludes the energy cost of manufacturing, so true lifecycle intensity will trend higher.¹

The financial comparison metric is levelized cost of energy (LCOE) per EIA definitions, and the paper devotes care to a quiet scandal of energy economics: no two published LCOE figures are computed with the same assumptions on interest rates, reserves, lifetimes, and financing, so the study enforced a single common source per variable, NREL's 2022 Annual Technology Baseline for renewables-plus-storage and EIA's 2022 Annual Energy Outlook for fossil sources.¹

Steelman: the best case for this methodology

This is how a serious institution prices an unbuilt system without fooling itself. Every assumption chain is anchored to a named historical database, every uncertain number carries a probability distribution instead of a point value, the FOAK/NOAK split prevents both the optimist's mistake (pricing a mature fleet on day one) and the pessimist's (never crediting learning), and the carbon ledger charges SBSP for launches and disposal that terrestrial accounting usually ignores, which is the honest direction of error if anything. The paper's framing line deserves weight: NASA is not proposing to build an SBSP; it is building a baseline against which any external SBSP proposal can be checked for reasonableness.¹

The design selection is also defensible. SPS-ALPHA and the tethered SPS are public, technically documented, architecturally opposite (concentrating hyper-modular versus planar area-scaled), and mature enough to cost but generic enough to generalize. That makes the study's sensitivity analysis, varying line items stochastically to find which subsystems dominate the LCOE, genuinely informative for the whole SBSP field rather than for one vendor's proposal. And the historical framing is pointed: the paper notes that between 1996 and 2016 the proposed specific power of SBSP systems, kilowatts per kilogram, barely moved even as other space systems improved dramatically, which is exactly the stagnation a transparent costing method exists to test.¹

Where a skeptic should push

The most load-bearing assumption is that cost-estimating relationships (CERs) built on heritage space systems transfer to a system the paper itself describes as having very little heritage. Its own discussion concedes the point in one sentence: SBSP is unique among space systems in size and construction, it stresses all the assumptions that make existing cost models work, and many space systems have useful heritage whereas SBSP has very little. A Design Maturity Factor applied to a machine with no antecedent is an educated discount, not a measurement; the F-RISK distributions inherit the analogies' blind spots.¹

Second, the 5-to-10-year data horizon. The paper warns that historical data may mislead for work performed more than 5 to 10 years out, which is fatal for a fleet whose FOAK is decades away and whose LCOE lives or dies on launch-cost trajectories no one can bound. Third, the learning-curve mechanics: dividing by 101 assumes a hundred-unit learning path that no policy environment guarantees; a single-FOAK world looks nothing like the NOAK number, and SBSP economics are precisely the case where the fleet never reaches unit 100. Fourth, on carbon, the authors exclude manufacturing energy and say so, and the equivalence argument for including launches is reasonable but debatable; the resulting intensity figures are indices for comparison, not audited footprints. All of this is methodology, published without results, so no cost or carbon number in this paper should be quoted as a finding.¹

What honest SBSP cost math means for space power

Non-obvious implication: the energy-and-space sector boundary, not the technology, is what such studies actually measure. The paper keeps returning to the question of which sector should shoulder SBSP's costs and leaves NASA's own role to a follow-on analysis, because a system that exists to sell electricity but requires space-sector maturation to build has no natural owner for its learning curve. Any in-space power scheme with terrestrial revenue, beaming, orbital data centers, or power-for-propellant, inherits this same orphan-learning-curve problem, and this methodology is currently the only public template for pricing it.¹

Genuine threat: the study's central historical observation is that SBSP's specific power stalled for two decades while everything around it improved. If the 2022 costing exercise reproduces that stall in LCOE terms, the honest conclusion is not that SBSP is one technology away, but that the architecture class itself has been price-stable against rapidly improving terrestrial alternatives, which sets a very high bar for any new proposal claiming the old numbers no longer apply.¹

Opportunity: the method is reusable, and the paper says so in its conclusions. The FOAK/NOAK learning machinery, the F-RISK cost-risk treatment, and the launch-inclusive carbon ledger can price any speculative space power architecture, from lunar grids to beamed-power constellations, with assumptions stated variable by variable. For space power advocates, adopting this template is strategically wise: a proposal costed in NASA's own method is far harder to dismiss than one priced in a pitch deck, and the sensitivity apparatus shows exactly which subsystem claims would move the answer most.¹

The bottom line

Established by this paper: a complete, internally disciplined methodology for costing and carbon-accounting SBSP exists, with named data sources, explicit learning-curve mechanics, and an honest statement of its own exclusions. Not established: any cost, LCOE, or carbon result, because findings were explicitly unpublished at the time of writing; nor the transferability of heritage cost models to a heritage-free system, which the authors themselves flag as the open question. Confidence is high that the methodology is a fair measuring stick, and low that any stick can measure a machine whose components, assembly industry, and demand context do not yet exist. What would confirm its value is the follow-on results report reproducing credible uncertainty bands; what would break it is those results collapsing into a single-point estimate, which would mean the machinery described here was decoration on a number chosen in advance.