Big moves around Tesla and Terafab

Intel’s arrival in Terafab changes the story. When Elon Musk unveiled Terafab on March 21, he described it as a giant Austin project that would eventually support about 1 terawatt of AI compute per year for Tesla, SpaceX, and xAI. The pitch was vertical integration on a scale even Tesla had never attempted. Then on April 7, Intel said it was joining the effort to help “refactor silicon fab technology,” which is the clearest sign yet that this was never going to be a car company casually learning how to run a leading-edge chip foundry on its own (bloomberg.com, electrek.co). That matters because fabs are not just expensive buildings. They are industrial systems with brutal yield targets, deep process knowledge, and supply chains that take years to stabilize. Musk’s original Terafab pitch bundled chip design, fabrication, memory, advanced packaging, and testing into one complex near Giga Texas. Intel’s entry turns that from a fantasy of total in-house control into something more legible: Musk’s companies want dedicated silicon capacity, but they still need a real semiconductor manufacturer in the loop (kut.org, datacenterdynamics.com, finance.yahoo.com). The scale of the demand helps explain why Musk is trying anyway. Tesla needs more compute for self-driving and for Optimus. xAI needs it for training and inference. SpaceX has been tied even more tightly to that stack since Musk combined xAI and X into the broader SpaceX orbit earlier this year, making the case for one shared hardware pipeline stronger than it looked a few months ago (cnbc.com, channelnewsasia.com, (electrek.co)). But Terafab is only half the story. The other half is what Tesla appears to be building all that compute for. A live Tesla job posting for “Technical Program Manager, Optimus Hand” says the company wants someone to lead development of a “high-dexterity, mass-manufacturable hand” for the humanoid robot and to manage the path from design and prototyping through validation, industrialization, and mass production. That is not research language. It is factory language (tesla.com). The hand is also a revealing place to focus. Humanoid robot demos can cheat with carefully staged walking or teleoperation, but hands are where manipulation becomes real and where manufacturing gets hard. Tesla’s listing stresses integration across hardware, firmware, controls, AI manipulation teams, sourcing, and suppliers. In other words, Tesla is not just tuning a lab prototype. It is trying to turn one of the most failure-prone subsystems in robotics into a repeatable product (tesla.com). That push lands at an awkward moment for Tesla’s core car business. The company reported 358,023 deliveries in the first quarter of 2026 while producing 408,386 vehicles, a gap large enough to sharpen the sense that Tesla needs its next act to be something other than more cars. Terafab and the Optimus hand role point in the same direction. Tesla is spending less effort pretending robotics is adjacent to the vehicle business and more effort wiring the company around it, right down to the fingers (cnbc.com, electrek.co, tesla.com)).