Tesla cuts Optimus chip memory to 72GB to enable mass production scale
- Tesla reduces AI5 chip memory to 72GB LP5 and AI6 to 144GB LP6
- Musk cites volume and cost as primary drivers for memory cuts
- Micron forecasts humanoid robots may need over 200GB memory each
- Tesla orders components for 5,000 Optimus units with 1 million annual target

*this image is generated using AI for illustrative purposes only.
Tesla Inc. CEO Elon Musk announced that the company halved the RAM capacity for its upcoming AI5 chip to 72GB of LP5 memory. This strategic reduction, alongside a one-third cut for the subsequent AI6 chip, is described as the only way to secure sufficient volume for Optimus humanoid robot production.
Musk stated on social media that these adjustments greatly reduce costs. He argued that the cuts would have a negligible effect on Optimus performance because memory bandwidth acts as a bigger limiting factor than total memory storage in real-time processing tasks. This clarification suggests that increasing raw capacity without corresponding bandwidth improvements yields diminishing returns.
Micron highlights memory demand gap
The announcement follows comments from Micron Technology Inc. CEO Sanjay Mehrotra during its fiscal fourth-quarter earnings call. Mehrotra indicated that humanoid robots are expected to require more than 200GB of memory and multiple terabytes of storage per unit, similar to autonomous vehicles. He noted that physical AI could become a significant driver of memory demand by the end of the decade, with several customers already sampling next-generation products.
An X user highlighted the scale of this potential demand by calculating that 10 billion robots at 200GB each would require 2 trillion gigabytes of RAM. This figure far exceeds the roughly 45 billion gigabytes of DRAM produced globally each year, underscoring the supply constraints Tesla aims to navigate through specification cuts.
Production targets and supplier audits
Tesla has reportedly placed its first large-scale component order for approximately 5,000 Optimus units. The company is also auditing Chinese suppliers ahead of production. The long-term goal is to build 1 million Optimus units annually.
| Specification | Previous Status | New Specification | Change |
|---|---|---|---|
| AI5 Chip Memory | Not specified | 72GB LP5 | Halved |
| AI6 Chip Memory | Not specified | 144GB LP6 | Cut by one-third |
| Target Annual Production | N/A | 1 million units | Planned |
On Thursday, Tesla shares closed 0.2% lower at $354.11. Benzinga Edge rankings indicate the stock has a Momentum score in the 13th percentile and a Growth score in the 41st percentile.
How will the significant gap between Micron's projected 200GB memory requirement and Tesla's 72GB AI5 specification impact the long-term scalability of Optimus for complex physical AI tasks?
What specific architectural optimizations or software-level memory management techniques must Tesla implement to ensure the halved RAM capacity does not degrade real-time inference performance?
Could Tesla's strategy to reduce memory specifications accelerate a broader industry shift toward bandwidth-centric rather than capacity-centric chip designs in robotics?

































