presentofai

Tesla 2nm A15 AI chip prototypes run at Samsung Texas fab

TL;DR

Samsung's Texas fab is running 2nm prototypes of Tesla's A15 AI chip, a generational leap that will power Full Self-Driving, the Optimus robot, and the Cybercab robotaxi, with volume supply targeted for mid-2027.

What happened

  • Samsung's Taylor, Texas fab has begun prototype runs of Tesla's next-generation A15 chip on a 2nm process, per the Seoul Economic Daily.
  • A15 performance claims are dramatic: 40x faster than A14, 8x the compute, 9x more memory, and 3x better performance per watt.
  • Final production verification is targeted for end of 2025, with production supply starting in 2027 and volume ramping mid-2027, per Business Korea.
  • The Taylor fab is equipped for 50,000 wafers per month and is reported fully booked with Tesla, Broadcom, and Arm.
  • Tesla holds a $16.5 billion supply deal with Samsung, which includes Elon Musk's right to walk the fab production floor. Musk also holds a separate production cooperation agreement with Intel.

Why it matters

  • A15 is the compute backbone for three of Tesla's highest-stakes bets simultaneously: Full Self-Driving software, the Optimus humanoid robot, and the Cybercab robotaxi.
  • A 40x speed jump over A14 is not incremental. It signals Tesla is designing for inference workloads that current hardware cannot handle at the required scale or cost.
  • Full Taylor capacity booked by Tesla, Broadcom, and Arm means Samsung's U.S. advanced node is now a strategic chokepoint. Any yield or schedule slip ripples across all three customers.
  • Samsung's $16.5 billion Tesla anchor deal validates the Taylor fab's commercial viability and strengthens Samsung Foundry's position against TSMC in the race for U.S.-based advanced node production.
  • Samsung is already planning Taylor 2 on a 1.4nm process with mass production targeted for 2030, suggesting Tesla's roadmap extends well beyond A15.

What to watch next

  • End-of-year production verification at Taylor: a pass clears the path to 2027 volume ramp; a delay compresses Tesla's Cybercab and Optimus launch timelines.
  • Yield rates at 2nm in Texas will determine whether Samsung can close the gap on TSMC's N2 node, which currently leads on process maturity.
  • Intel's role under the separate cooperation agreement is undefined in current reporting. Watch for whether Intel serves as a backup fab or a complementary node for different Tesla chip tiers.

Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.