Tesla cuts AI5 RAM to 72GB of LPDDR5, Musk says

Tesla's decision sits against a backdrop of tight memory supply across the industry, driven by ballooning AI demand, a squeeze the article refers to as “RAMageddon.” By shaving gigabytes off each chip, Tesla aims to build more of them and lower per-unit cost — which Musk frames as the only way to reach the volumes Optimus production needs.

The move is also a bet on where the real constraint lies. Musk argues memory bandwidth limits these systems more than total capacity, and since bandwidth has not changed, he expects a negligible effect on Optimus performance. If that reasoning holds, Tesla gets more chips at lower cost when volume production starts next year; if it does not, Optimus will have to run on less than half the memory Micron estimates a humanoid robot may need.

The comparison to today's hardware helps frame the scale: AI4 computers ship with 32GB of GDDR6, and Hardware 3 had just 8GB of LPDDR4 per chip, with bandwidth as the real bottleneck at roughly 15% of AI4. A mid-generation AI4+ upgrade is also in the works, expected to double AI4's memory to 32GB per chip and 64GB total, while AI5, AI6 and later generations are expected to use a single chip per board.