Tesla has reduced the LPDDR6 memory on AI6, the next-generation chip for its Optimus humanoid robot, from 216GB to 144GB — a one-third cut driven by tightening low-power DRAM supply and the need to scale production1,3.
Elon Musk disclosed the specification change on X last month, saying it was the only way to get enough volume for Optimus production and that it greatly reduces cost. Musk also said memory bandwidth is a bigger limiting factor than total memory storage, and Tesla has kept bandwidth at the originally planned level.
The cut is not limited to AI6. Tesla also lowered the LPDDR5 capacity on its current-generation AI5 chip from 144GB to 96GB.
The backdrop is a widening battle for low-power DRAM. LPDDR was originally designed for smartphones and tablets, but adoption is growing in battery-powered robots, electric vehicles, and — critically — AI servers, intensifying competition for supply.
Samsung Electronics led the DRAM market with a 38% share, and the broader industry brought in nearly $200 billion in DRAM revenue in Q3. Counterpoint Research projects global DRAM demand for humanoid-robot computing units will grow from roughly 8,300 terabits this year to 172,000 terabits in 2030, more than a 20-fold increase.
Micron Technology said at a memory roadmap briefing that LPDDR6 is expected to cut power consumption by about 35% versus LPDDR5. Even after the reduction, AI6's 144GB is roughly 9 to 12 times the 12GB or 16GB in a Galaxy S26 Ultra, and a single robot requires about 10 times more DRAM than a smartphone.
ANALYSIS Tesla's decision to trade memory capacity for production volume reflects a concrete constraint: as LPDDR migrates into data-center and edge-AI workloads, robotics programs that once competed mainly with mobile devices for supply now face server-class demand on the same memory lines. Preserving bandwidth while cutting capacity suggests Tesla is optimizing for inference throughput per unit rather than raw model size.