Tesla cuts RAM on AI5 and AI6 chips to speed Optimus production

Tesla is putting its next-generation AI chips on a memory diet. On October 1, 2026, Elon Musk said the AI5 chip will now ship with 72GB of LPDDR5 RAM, while the AI6 chip’s memory is being cut by one-third to 144GB of LPDDR6.

The goal is volume and cost, specifically for the Optimus humanoid robot. Musk’s pitch is that the robots won’t notice, because capacity was never the real bottleneck.

What Tesla changed, and why Musk says it won’t hurt

Earlier specifications for AI5 had pointed to RAM allocations above 144GB. The new figure of 72GB is a sharp step down from those expectations.

AI6, the follow-up chip, is getting a smaller trim. Its memory drops by a third, landing at 144GB of LPDDR6.

Musk’s central argument is that memory bandwidth, not total memory capacity, is what actually limits performance. He expects no significant impact on how the robots perform.

Per the research findings, the redesign kept critical bandwidth performance in place while trimming capacity.

Where AI5 and AI6 fit in Tesla’s chip roadmap

AI5 was completed ahead of schedule, with its tape-out happening in April 2026. Tape-out is the point where a chip design is finalized and sent off for manufacturing.

Notably, AI5 is not headed primarily for cars. It is intended mainly for Optimus robots and training supercomputers, while the existing AI4 chip will continue to power Tesla’s vehicles.

Tesla plans high-volume production of AI5 for 2027. AI6 is expected to deliver about twice the performance of AI5.

Manufacturing is split between two of the biggest names in chipmaking. Samsung is producing the chips using a 2nm process, and TSMC is also building them.

What this means for Tesla and the Optimus push

The most direct read is that Tesla is optimizing for scale over spec-sheet bragging rights. Cutting RAM reduces the bill of materials for every chip, and those savings multiply quickly when the plan involves mass production of humanoid robots.

The research findings frame the move as one that could improve margins through cost savings.

Tesla says the reductions should have a negligible impact on operational capability, but the real test comes when these chips are running inside working robots.

AI5 is slated for high-volume production in 2027, so any performance gaps tied to the smaller memory pool may not show up publicly until then.

For AI6, a chip that is expected to roughly double AI5’s performance while carrying 144GB of faster LPDDR6 memory still leaves meaningful headroom, even after the one-third cut.