Tesla Cuts Planned Memory Capacity on AI5 and AI6 Chips for Optimus Mass Production

Elon Musk announced that Tesla will cut the planned memory capacity for its upcoming AI5 chip by 50% (from 144GB to 72GB LPDDR5) and its AI6 chip by roughly 33% (from 216GB to 144GB LPDDR6). The decision was made to conserve global DRAM memory supplies to support the mass production of the Optimus humanoid robot. With memory manufacturers like Micron estimating that each humanoid robot requires over 200GB of DRAM, physical memory supply limits present a major bottleneck for scaling robotics. This move demonstrates how global semiconductor supply constraints are directly shaping hardware architecture decisions for next-generation AI and autonomous systems. Musk stated that the performance impact on Optimus will be negligible because memory bandwidth—which remains unchanged—is the main system bottleneck rather than total capacity. The hardware adjustment responds to calculations showing that global annual DRAM output (~45 billion GB) cannot sustain long-term demand if millions of robots are manufactured.