Why Tesla Cut Memory in Its AI5 and AI6 Chips

Tesla has reduced the memory planned for its AI5 and AI6 chips, saying the change is needed to secure enough supply for Optimus production and lower costs. The new configurations use less RAM, but Tesla has not provided a guaranteed production target or independently verified how the cuts will affect real-world performance.14

The new AI5 and AI6 memory specifications

Chip	Previous memory plan	New memory specification	Change
AI5	144GB	72GB LPDDR5	Down 50%
AI6	216GB	144GB LPDDR6	Down one-third

The revised figures were reported after Elon Musk said Tesla had halved AI5 memory and cut AI6 memory by a third.49

Why Tesla says it made the cuts

Musk described the reductions as necessary to get enough memory supply for Optimus production, and said they would substantially reduce costs.14 Using less memory per chip could allow a given supply of memory to support more chips, all else being equal. But Tesla has not published a per-chip savings figure, a production-volume commitment, or evidence that memory supply is the only factor limiting robot output.

The decision also fits a wider concern about memory availability for AI hardware. One report relaying a Micron estimate says a humanoid robot could require about 200GB of DRAM; that estimate is not a specification for Optimus or a direct comparison with the memory configured on Tesla’s chips.9

What the change could mean for Optimus performance

Musk said he expects the reduced memory capacity to have a negligible effect on Optimus, arguing that memory bandwidth matters more than total capacity and that bandwidth is unchanged.9 That is Tesla’s stated expectation, not a published benchmark or independent performance result.

Capacity still matters for how much data and how many workloads can fit in memory at once. The available reporting does not establish whether the smaller configurations will constrain particular robot tasks or future software. The practical effect will depend on Tesla’s systems and workloads, not just the headline memory total.

What is known about AI5’s deployment timeline

The memory announcement does not give a deployment date. Reporting says AI5 is expected to enter mass production in 2027 and describes the chips as planned for Tesla’s vehicle-computing and Optimus efforts, but it does not establish when AI5 will appear in either product.8 A production target is not the same as a confirmed rollout schedule, so specific vehicle or robot deployment dates remain uncertain.

The clearest takeaway is that Tesla is prioritizing memory availability and cost alongside chip capacity. The revised specifications may help the company use scarce memory more efficiently, while the performance trade-off and resulting production impact remain to be tested.