Particle.news
Download on the App Store

Samsung Prepares to Produce Tesla’s AI5 on 2nm at Taylor, Texas Fab

If sustained, Samsung’s move would signal its 2nm process can meet big-customer demands and push the AI5 design from prototype toward customer qualification and volume ramp.

Overview

  • Multiple industry reports on July 13 said a Samsung foundry engineer posted that the Tesla-Samsung AI5 design reached tape-out and was slated for manufacture at Samsung’s Taylor, Texas plant; the post was later deleted and Samsung declined to comment.
  • Tesla has already said it taped out AI5 to both Samsung and TSMC and showed engineering samples in April, meaning two foundries will make slightly different physical versions of the same chip design.
  • The AI5 module shown by Tesla combines a compact accelerator die with 12 SK hynix memory packages that look like GDDR6/GDDR7, a design that aims to deliver very high memory bandwidth for cars, robots and data-center use.
  • Engineering samples exist but customer qualification and long-run yield stabilization remain outstanding, and reports differ on volume timing with some outlets citing late 2026 or early 2027 while Samsung has publicly targeted mass production in the second half of 2027.
  • If Taylor can sustain acceptable yields, landing AI5 on 2nm would be a major proof point for Samsung against TSMC, potentially anchoring large orders at the U.S. fab and widening foundry options for AI chip customers.