Tesla’s Terafab push surfaces again
Social posts highlighted Tesla’s Terafab—a proposed $20B AI chip factory tied to the company’s AI5 ambitions and vertical integration for FSD/Optimus—as a stark contrast to Apple’s TSMC reliance and a signal of how companies are addressing supply shortfalls noted. The discussion frames two supply strategies: build‑your‑own silicon vs. secure external foundry partnerships.
Elon Musk posted “Terafab Project launches in 7 days” on X on March 14, 2026, setting March 21 as the public kickoff for the initiative. (finance.yahoo.com) Industry reporting puts Terafab’s construction cost in the $20–25 billion range and describes an initial target of roughly 100,000 wafer starts per month while aiming for a 2‑nanometer process node. (blockonomi.com) Tesla’s internal roadmap tied to Terafab calls for small‑batch production of its AI5 processor in 2026 and targeted volume production in 2027. (fintechweekly.com) Published technical summaries claim the AI5 design promises roughly 40×–50× more compute and about 9× more on‑chip memory than Tesla’s AI4 architecture, positioning the part as a high‑density inference engine for FSD and Optimus workloads. (teslarati.com) Elon Musk has publicly floated discussions with Intel and warned at Tesla’s 2025 shareholder meeting that the company will “probably” need to build a “gigantic chip fab” because supplier capacity won’t meet Tesla’s projected AI/robotics demand. (datacenterdynamics.com) Analysis of fab economics underscores the technical and capital hurdles: ASML EUV lithography tools cost on the order of $150–$400 million per machine, TSMC’s Arizona campus represents over $65 billion of investment, and modern fabs demand large, stable power footprints and complex cleanroom infrastructure (estimates often cite tens of megawatts to 50–100 MW ranges for advanced fabs). (cnbc.com) Coverage from semiconductor analysts and reporting outlets emphasizes that a March 21 “launch” is most plausibly a project kickoff or public construction/partnership announcement rather than a fully commissioned 2nm production line, with small‑batch outputs and multi‑year ramp risk still front of mind. (basenor.com)
