Musk's Terafab reportedly targets memory output on par with Samsung's DRAM capacity

Terafab is not a fab strategy; it is a bargaining chip. Tesla and SpaceX's real objective is memory sovereignty—autonomous-driving training clusters and Starlink comms silicon are consuming DRAM at exponential rates, while HBM allocation is already locked by NVIDIA and hyperscalers. Yet DRAM remains a brutal experience-curve business: 1α/1β yield ramps demand three to five years, and no new entrant has ever turned a first-gen line profitable. The move to court Japanese and Korean equipment and materials suppliers reveals the absence of in-house process heritage—this is a "buy the stack" play, not a technology play, and it concentrates supply-chain risk in a single geographic corridor. The strategic value is not in wafers per month; it is in the threat itself. A credible "Samsung-scale" alternative forces incumbents to recalibrate HBM pricing and delivery priority. Over the next 12–24 months, expect site selection and tool reservations, not production. The real inflection arrives in 2026, when AI inference workloads create a structural HBM4 shortfall. At that moment, this option either gets exercised—or it becomes the most expensive letter of intent in semiconductor history.