Musk says Tesla nears completing AI5 chip, now developing AI6

Tesla is close to completing its AI5 chip design and has started developing the AI6 chip for use in its vehicles and data centers, CEO Elon Musk said on November 23, in a post on X.

Tesla currently uses the AI4 chip in its cars.

Musk said the company aims to bring a new AI chip design to volume production every year.

In July, Samsung said it secured a US$16.5 billion deal to manufacture AI semiconductors for Tesla, with a new facility in Texas dedicated to producing the AI6 chip.

Musk also said Tesla expects to eventually manufacture more AI chips than all other AI chips combined.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Dual-foundry plan lowers risk though tight High Bandwidth Memory supply could slow Tesla’s yearly chip pace

Elon Musk confirmed Tesla’s AI5 and AI6 will be built at Samsung’s Taylor, Texas site, with a second line at Taiwan Semiconductor Manufacturing Company (TSMC)’s Arizona fab 1. Samsung’s foundry has struggled at leading nodes, with share sliding from 19.1% in Q1 2019 to 7.7% in Q1 2025 2.
AI6 needs High Bandwidth Memory (HBM) for Full Self-Driving (FSD) and robotics. Micron sold out its entire 2025 HBM book 3, with a 3.5% supply gap lifting prices 8–12% in 2025 4.
An annual chip pace depends on successful increases at both U.S. fabs and on locked HBM deals. The HBM market sits near $34–35 billion in 2025 5 4, so Musk’s line about “more AI chips than all others combined” reads ambitious without visible ramp timing or contracts.

Packaging plus nearby vendors see openings through 2027 near Samsung Taylor and TSMC Arizona

AI5 plus AI6 split between Texas, with Arizona too. Both sites will need advanced packaging with HBM integration. Substrate supply links the die to the package. HBM work involves high routing density, Through-Silicon Via (TSV) steps, and complex assembly 1 6.
Micron is investing $2.5 billion in Singapore post-fabrication capacity targeting 2027 4. U.S.-based OSAT or substrate firms near Taylor or Phoenix could win share as domestic sourcing expands 1.
Power and cooling vendors have an opening as AI6-class parts paired with HBM3E or HBM4 raise heat. Recent HBM samples reach over 2 TB/s and up to 2.8 TB/s per stack, pushing next-gen cooling across fabs, plus Tesla data centers 7.