Abstract
Between February and May 2026, Elon Musk publicly committed SpaceX/Tesla strategic intent to lunar mass driver development as part of "Project TERAFAB" — a declared strategic direction (not yet funded engineering program) to relocate AI compute infrastructure to the lunar surface. The verbatim public record: an xAI all-hands description in February 2026 of a "cannon-like device using magnetic power"; a Giga Texas presentation on March 22-23 2026 with the quote "I want to live long enough to see the mass driver on the moon" and a tweet caption "Mass drivers on the Moon!"; a May 3 2026 tweet "Path to Petawatts is Mass drivers on Moon." Musk's framing claims 500-1000 TW/yr of AI satellite deployment via mass driver and projects ~$500/kg cost (a 90% reduction from current launch). No specific budget, timeline, or engineering program has been announced.
Key claims
- xai-all-hands-feb-2026: Describes the device as "a cannon-like device using magnetic power."
- giga-texas-march-2026-quote-1: "I want to live long enough to see the mass driver on the moon."
- giga-texas-march-2026-quote-2: "Mass drivers on the Moon!" (tweet caption, March 22, 2026)
- may-2026-tweet: "Path to Petawatts is Mass drivers on Moon" (X, May 3, 2026, 9 words)
- capacity-claim: "500 to 1000 TW/year of AI satellites into deep space" (Giga Texas, March 2026)
- cost-claim-attributed: "Per-kilogram costs could drop to approximately $500 — a 90% reduction versus current methods" (article paraphrase from Musk presentation).
- not-funded: "Lunar mass driver remains a declared strategic direction, not yet a funded engineering program." (Basenor article paraphrase)
- ambition-context: "We are going to push the limits of physics in compute and do some wild and crazy things" (Giga Texas, March 2026).
Reviewer notes
Tier B as public-figure quotes from one of the most influential principals in space launch. The numbers Musk has stated (500-1000 TW/yr of AI satellites, $500/kg, 90% cost reduction) are aspirational targets, not engineering commitments. The 500-1000 TW/yr deployment scale is roughly consistent with Handmer's 100 TW/yr economic-threshold framing — but to actually deliver 1000 TW/yr of orbital infrastructure would require ~10^9 kg/yr to LEO equivalent (assuming ~1 kg of structure per kW of orbital solar/data-center), which is ~100× the Handmer 10 Mt/yr nameplate throughput. Either Musk's TW figure refers to peak rather than continuous capacity, or the mass intensity of "AI satellites" is much lower than typical orbital infrastructure. The strategic commitment from SpaceX is significant for q7 capital-mobilization analysis: if SpaceX seriously pursues this, capital flow and engineering acceleration are substantially different from the 1979-2011 baseline. But the "I want to live long enough to see it" framing suggests Musk himself is not predicting near-term operational deployment.