The Vacuum Advantage Ends Near a Millipascal: A Reproducible Assessment of Flywheel, Superconducting-Magnetic and Thermal Energy Storage on Earth and in Space

Space is often described as a perfect vacuum, and vacuum as what makes energy storage nearly lossless. This preprint tests both ideas with numerical models of a 24 kWh carbon-fibre flywheel, superconducting magnetic energy storage (SMES) and a hot thermal store. Vacuum removes one loss channel, gas drag on the rotor (windage), which falls from 1.65 MW at atmospheric pressure to 0.28 W at 0.1 mPa. The benefit ends where windage drops below the losses no vacuum removes, here 39.8 W from bearings and motor electronics. A closed-form saturation pressure p*, at which windage is one tenth of those fixed losses, is about 1.6 mPa for the reference rotor. Below it the rotor keeps 96.1% of its energy over a day whether its housing holds 0.1 mPa or the roughly 2 µPa of a 400 km orbit. Because p* scales with the fixed losses, bearings and electronics 100 times better would still saturate at about 14 µPa, a pressure laboratory pumps reach. Vacuum does not raise the ceiling on stored energy either: a spinning rim and a superconducting solenoid are both hoops in tension, limited to σ/2ρ (12–18 Wh/kg for steel, 109–217 Wh/kg for carbon fibre, before housings and cryostats). What space adds is a cold sky, not a better vacuum: an ideal radiator facing deep space rejects 0.15 W/m² at 40 K and 0.009 W/m² at 20 K, enough only for small cryogenic loads, and the same sky makes a 900 K thermal store lose 1.1% more heat. Launching the reference flywheel system to orbit costs at least 373 times the energy it stores. An appendix confirms the tether braking price of the companion paper "Every Extractor Is a Brake" (doi:10.5281/zenodo.23204818) and corrects a figure in the source simulation: a generator tether returns 0.84%, not 2.7%, of minimum launch energy as electricity. The results are negative and confirmatory; no new physics is claimed. Supplementary files: vacuum_storage_ledger.py (Python 3, standard library only) reproduces the paper's tables; physics-lab-code-and-data.zip is the full simulation package (models, nine experiments, results, figures, 40 tests, dashboards).