Tesla is cutting the memory specifications on its next-generation AI chips to secure enough supply for mass production of the Optimus humanoid robot. CEO Elon Musk announced on X on October 1 that the AI5 chip's memory has been halved to 72GB of LPDDR5, while the AI6 chip drops by roughly one-third to 144GB of LPDDR62,4.
"This was the only way to get enough volume for Optimus production and greatly reduces cost," Musk wrote. He added that the de-spec will have "a negligible effect on Optimus performance" because memory bandwidth remains unchanged.
Memory supply as binding constraint
The AI5 chip was initially expected to carry 144GB of LPDDR5. Trial production for AI5 is already underway at Samsung's Taylor facility in Texas, with volume production expected in 2027. Tesla intends to use both the AI5 and AI6 chips to power its Full Self-Driving compute stack and the Optimus robot.
The decision underscores persistent tightness in the memory market. Tesla thanked Micron Technology in July for providing what it called a substantial memory allocation. Micron Technology CEO Sanjay Mehrotra has said each humanoid robot requires more than 200GB of DRAM and several terabytes of NAND flash, positioning robots as a potential new demand driver for memory1.
ANALYSIS Tesla's revised AI5 spec of 72GB now sits well below Micron Technology's 200GB-plus DRAM estimate per robot, a gap that raises questions about whether Tesla's inference workloads are lighter than the industry benchmark Micron Technology described or whether the company is accepting a tighter memory envelope to hit production targets.
Production targets
Tesla has placed its first large-scale component order for roughly 5,000 Optimus units and is targeting weekly production of more than 1,000 units by year-end. The AI6 and a subsequent AI6.5 chip are expected to retain TSMC and Samsung as fabrication partners.
On October 1, three separate hardware moves treated inference cost and component availability, rather than training, as the binding constraint on AI deployment[1]. ANALYSIS Tesla's chip de-spec fits that pattern: the company is trading peak memory capacity for manufacturability and supply certainty.
Musk framed the memory reduction as both a supply and a cost measure. With Optimus production scaling from pilot orders to a targeted weekly cadence of over 1,000 units, the chip de-spec converts a supply-chain bottleneck into a design concession whose real-world performance cost Tesla claims will be minimal.