Elon Musk says Tesla (TSLA) cut RAM for AI5 to 72GB, AI6 to 144GB to enable Optimus volume and lower costs

Product-specification disclosures of this kind from the chief executive, delivered outside formal filings or earnings events, are a recurring feature of coverage of this company and have historically been read as direction-of-travel signals rather than commitments, since stated specifications on future hardware generations have on prior occasions shifted between announcement and production. The substance here is a trade-off: trimming memory on the inference chips is a cost-down and volume-enabling decision, which fits the established pattern of the company de-scoping specifications to hit price points and manufacturing scale, with the open question being whether reduced on-device memory constrains capability and how the deficit is handled, typically through architectural workarounds or reliance on later silicon revisions. The distinction worth drawing is between the nearer-generation part, where changes signal imminent production decisions, and the further-out part, where figures are more aspirational and carry less weight. The relevant watchpoints are whether the company confirms the specifications through formal channels, any supplier or memory-procurement implications for the semiconductor complex that prices such design decisions, and commentary on the cadence of the robotics programme that the cost-down is framed as enabling. As an executive remark rather than a filing, the signal is directional and subject to revision.