Inside Tesla’s Terafab: It is a DONE DEAL - TSMC CEO Said Impossible - The Epic Manufacturing Flex

Terrafab is positioned as a chip factory 50 times more ambitious than anything built before, claiming it will slash AI chip costs by 70–90% and power everything from millions of Optimus robots to exascale AI clusters on Earth and in space. Backed by an unlikely coalition of Intel, Samsung, Tesla, SpaceX, and xAI—plus reported billion-dollar interest from SpaceX investor Ron Baron—the project is framed as a done deal despite TSMC’s CEO calling it impossible in under three to five years. The argument leans on Tesla’s history of defying incumbents, especially its homegrown 4680 battery, which became Tesla’s cheapest option even at modest scale by prioritizing manufacturing speed and scaling like no other product. Applying that same playbook, Tesla is building its own chips because Nvidia’s estimated 80% profit margin and TSMC’s refusal to scale to Tesla/SpaceX volumes leave no alternative, with Intel partnering on core manufacturing and Samsung building its AI5 chip in Texas. The Terafab would be ten times larger than the Texas Gigafactory, and Tesla’s manufacturing-first advantage—seen in fully automated Cybercab production and xAI’s record-speed data centers—is cited as proof that industry experts can be defied. Ultimately, the vision extends off-planet, with specialized chips potentially made on the Moon, connected through Starlink laser links, positioning Terafab as the critical bottleneck for Cybercabs, Optimus, and Starship, and a potential path to trillion-dollar revenue.

• ▶ 0:04 Terrafab is framed as a chip factory 50 times more ambitious than anything else, with claims it will slash AI chip costs by 70–90%, power millions of Optimus robots, and run exascale-scale AI clusters on Earth and in space.
• ▶ 0:26 The project is called a "done deal" because of an unlikely coalition—Intel, Samsung, Tesla, SpaceX, and xAI—plus SpaceX's biggest investor Ron Baron reportedly wanting to invest another billion dollars.
• ▶ 0:50 After TSMC's CEO called it "impossible in under 3 to 5 years," the host cites Tesla's 4680 battery and dry electrode tech as precedent for Tesla defining industries, not just entering them.
• ▶ 2:17 Tesla/SpaceX told TSMC they need 50 times as many chips as the entire world currently makes; after TSMC and Samsung declined, Tesla decided to make the chips itself.
• ▶ 2:58 Nvidia sells chips at an estimated 80% profit margin, but Tesla's custom, task-specific chips could cut chip costs by 70–90%.
• ▶ 4:30 TSMC's C.C. Wei warns there are no shortcuts in foundry work—it takes 2–3 years to build a fab and 1–2 more to ramp—but Musk responds that Tesla will always be a major TSMC customer, not a competitor.
• ▶ 6:54 Tesla first partnered with Panasonic because global EV battery capacity couldn't meet demand, but eventually built its own homegrown 4680 cell (dry cathode, no tabs) — designed with manufacturing speed in mind, inspired by bottle production lines.
• ▶ 8:07 Even at relatively low scale, Tesla's own 4680 battery is already the cheapest battery Tesla can get — cheaper than third-party suppliers — and with dry cathode issues solved, Tesla has 40 GWh at Giga Texas and is targeting 80 GWh in Berlin.
• ▶ 9:09 Tesla is applying the same playbook to chips: since TSMC, AMD, Samsung, and Nvidia won't scale to Tesla/SpaceX volumes, Tesla must build its own — with Intel helping and Samsung building its AI5 chip in Texas, as Tesla learns chip-fab production in a two-way partnership.
• ▶ 10:42 The Terafab will be 10 times larger than the Texas Gigafactory, one of the largest buildings in the world.
• ▶ 11:00 Tesla's core advantage is manufacturing, not exotic tech; it focuses on the most affordable battery that can scale like nothing else.
• ▶ 12:07 Tesla's Cybercab factory uses a 100% automated "unbox" production method, producing the most complex consumer product with no humans involved.
• ▶ 14:02 xAI built data centers three times bigger and 20 times faster than experts believed possible, showing Musk can defy industry experts like TSMC’s CEO on the Terafab.
• ▶ 14:52 Musk confirms a $3 billion research fab at Giga Texas, with SpaceX handling the initial Terafab phase, Tesla running the research fab, and Intel partnering on core manufacturing using its state-of-the-art 14A process.
• ▶ 16:23 The research fab will enable thousands of manufacturing iterations, making chip production more efficient and affordable at unprecedented scale, aided by Tesla’s in-house AI system, Tesla 1.
• [18:15-18:30] The Terafab's arrival timeline is uncertain (3-10 years away), but SpaceX and Tesla will build it "unlike anything else on the planet."
• [18:32-18:40] The vision goes off-planet: specialized chips could be made on the Moon for satellites and space data centers, launched via SpaceX's rail gun, with all data centers connected through Starlink's laser technology.
• [19:16-19:23] The Terafab is the critical bottleneck for all major products—Cybercabs, Optimus robots, and Starship—and the host teases it could turn Tesla/SpaceX into a trillion-dollar revenue company.

• ▶ 0:04 Introduction and Stakes (0:04 - 2:06) - Sets up Tesla/SpaceX's Terrafab and why it matters.
• ▶ 2:06 Wall Street and TSMC Reaction (2:06 - 5:09) - Baron's take, Wei's warning, and why TSMC isn't a direct rival.
• ▶ 5:09 Sponsor, Battery History, and Custom-Chip Rationale (5:09 - 10:44) - Sponsor break, Tesla's 4680/battery déjà vu, and the cost edge of custom AI chips.
• ▶ 10:44 Terrafab Scale and Manufacturing Experience (10:44 - 13:59) - Planned scale, Tesla's manufacturing muscle, and why speed beats incumbents.
• ▶ 13:59 xAI, Research Fab, and AI Manufacturing (13:59 - 18:18) - xAI's fast data centers, the $3B Giga Texas research fab, and Tesla 1/AI-driven fabs.
• ▶ 18:18 Terrafab Vision and Future Teaser (18:18 - 20:00) - Moon/space data centers and a teaser of trillion-dollar revenue.