Tesla has cut the RAM specifications on its upcoming AI5 and AI6 chips by half and one-third respectively, with CEO Elon Musk stating the reduction is necessary to secure sufficient component volume for mass-producing Optimus humanoid robots while lowering overall manufacturing costs.
Tesla is putting its next-generation artificial intelligence chips on a memory diet to clear the path for large-scale humanoid robot manufacturing. On October 1, 2026, Elon Musk announced that the in-house AI5 chip will ship with 72 gigabytes of LPDDR5 RAM, down significantly from earlier plans that pointed to allocations above 144 gigabytes. The follow-up AI6 chip is also getting trimmed, dropping by a third to land at 144 gigabytes of LPDDR6 memory.
The engineering pivot is designed to stretch available dynamic random-access memory across a massive production pipeline. Earlier specifications for AI5 had targeted higher capacities, but the sharp reduction lowers the bill of materials for every unit produced while addressing the physical supply constraints involved in building robots at scale.
Elon Musk Explains the Memory Reductions for AI5 and AI6 Chips
The decision to scale back memory capacity stems directly from the sheer volume of components required to manufacture humanoid robots for a global market. Musk addressed the hardware changes on social media, emphasizing that the hardware diet was unavoidable for the company’s robotics roadmap. He added that the cuts should have a negligible effect on Optimus performance, explaining that memory bandwidth serves as a bigger limiting factor than total memory capacity, and noted that memory bandwidth was held constant in the new configurations.
According to Tesla’s chip roadmap, manufacturing duties for these processors are split between industry heavyweights, with Samsung building the chips using a 2-nanometer process alongside production at TSMC. While AI5 was completed ahead of schedule with its tape-out occurring in April 2026, the processor is intended primarily for Optimus robots and training supercomputers, leaving the existing AI4 chip to continue powering Tesla’s electric vehicles.
Micron Technology Warns of Severe Supply Pressures for Physical AI
The hardware reduction highlights a looming crunch in global memory supply as humanoid robots emerge as heavy consumers of advanced silicon. During Micron Technology Inc.’s fiscal fourth-quarter earnings call on Wednesday, CEO Sanjay Mehrotra told investors that humanoid robots are expected to require more than 200 gigabytes of memory and multiple terabytes of storage per unit, placing their hardware demands on par with autonomous vehicles.
Physical AI is fast becoming a major driver of memory demand by the end of the decade, with several customers already sampling Micron’s next-generation products. Observers have calculated that if 10 billion robots require an estimated 200 gigabytes of RAM each, the resulting demand of 2 trillion gigabytes would vastly outstrip the roughly 45 billion gigabytes of DRAM produced worldwide annually. That stark arithmetic illustrates the supply bottleneck Tesla aims to sidestep by halving its memory footprints.
Production Ramps and Market Response as Tesla Targets High-Volume Output
Tesla is already moving to secure its supply chain, having reportedly placed its first large-scale component order for roughly 5,000 Optimus units while auditing Chinese suppliers ahead of production. The long-term goal is to build 1 million Optimus units annually, with high-volume production for the AI5 chip slated for 2027.
On the financial front, Tesla shares closed 0.2% lower at $354.11 on Thursday before ticking up 0.22% to $355.60 in extended trading. As the company prepares for mass manufacturing, any performance penalties tied to the reduced memory pool will face their true test when these processors begin operating inside working robots.