Samsung Starts Taylor Fab to Fabricate Tesla Chips

Samsung Electronics is initiating operations at its Taylor, Texas foundry and moving to install production equipment at a ceremony scheduled April 24. The company signed a $16.5 billion multiyear contract with Tesla to manufacture next-generation driving and robotics chips, with Samsung projecting mass production of the AI6 in the second half of 2027, and earlier runs of AI5 at the Taylor facility. "The next-generation Tesla chip is scheduled for mass production at the Taylor fabrication plant in the U.S. in the second half of next year, with both its design and manufacturing progressing smoothly," said Jinman Han, head of Samsung's foundry business.

The Taylor plant is slated to run 2-nanometer process technology to produce Tesla's custom accelerators. Samsung has dispatched ultra-fine process engineers to the site and begun installing front-end and back-end equipment after earlier delays tied to order visibility. Achieving stable yields is the immediate technical milestone; industry commentary highlights a yield threshold near 60% as the inflection point for profitable mass production. Samsung also signed a memorandum of understanding with AMD to supply HBM4 memory for the MI455X accelerator and to discuss potential foundry services for future AMD products.

• •Fabrication of Tesla AI5 and AI6 accelerators on 2-nanometer processes
• •Supply of HBM4 high-bandwidth memory for AMD MI455X accelerators
• •Potential expanded foundry collaboration with AMD for next-generation processors

Bringing the Taylor fab online solves Samsung's long-standing need for a domestic U.S. anchor customer and reshapes competitive dynamics with TSMC's Arizona facility, which operates at larger process nodes such as 4-nanometer. For Samsung the deal is strategic: it converts a high-capex U.S. facility into a revenue-generating foundry, improves geopolitical alignment with U.S. customers under CHIPS-era incentives, and accelerates Samsung's push into AI infrastructure supply chains. For Tesla, locking a U.S.-based contract manufacturer for custom AI silicon reduces supply-chain risk for autonomous driving and robotics ambitions. The AMD HBM4 tie-in signals broader ecosystem play beyond a single customer, linking memory supply to rack-scale AI architectures like AMD's Helios.

The ramp profile for yields and throughput at 2-nanometer will determine whether Taylor shifts Samsung from foundry losses to profitability. Track production qualification milestones, initial wafer starts, yield progression toward the 60% threshold, and follow-on orders from other U.S. AI compute customers. Market reactions will hinge on whether Samsung sustains process leadership versus TSMC and can commercialize 2-nanometer volumes on schedule.