Elon Musk Slashes Tesla AI Chips Memory Amid Global Shortage, Claims Optimus Performance Unaffected

Tesla Reduces AI Chip Memory Specifications Amid Supply Constraints

Elon Musk revealed Tesla is cutting memory capacity for its AI5 chip and AI6 chip as the company prepares for large-scale production of processors powering its Optimus robot and Full Self-Driving systems1

1

2

2

. The AI5 chip memory was initially slashed to 72GB of LPDDR5—half the originally planned 144GB—before Musk adjusted it to 96GB to avoid being "the only company using the min RAM version of LP5"2

2

. The AI6 chip memory now carries 144GB of LPDDR6, representing a one-third reduction from initial specifications4

4

. Musk described the memory reduction as "the only way to get enough volume for Optimus production," emphasizing it would "greatly" reduce costs while having a "negligible effect on Optimus performance"3

3

. He explained that memory bandwidth matters more than total memory capacity for the humanoid robot's performance3

3

.

AI5 and AI6 Architecture Relies on SRAM for Performance

The AI5 uses a half-reticle design featuring a large central compute die surrounded by 12 DRAM modules1

1

. Manufacturing will occur at TSMC and Samsung using 2nm process technology, with the chip expected to deliver between 2,000 and 2,500 TOPS of performance1

1

. The single-chip version aims to match performance with Nvidia H100, while a dual-chip configuration targets competition with Blackwell-based systems1

1

4

4

. Trial production is already underway at Samsung's Taylor facility in Texas, with volume production expected in 20271

1

4

4

. Tesla plans to use TSMC and Samsung for AI6 and AI6.5 production as well1

1

. These designs rely heavily on SRAM, with Musk stating that roughly half of their TRIP AI compute accelerators are dedicated to SRAM1

1

4

4

. This architectural choice allows calculations held in SRAM cache to access effective memory bandwidth about an order of magnitude higher than DRAM1

1

.

Micron Forecasts Massive Memory Demand for Humanoid Robots

The memory supply constraint driving Tesla's specification changes reflects broader industry challenges. Micron Technology CEO Sanjay Mehrotra estimated during the company's fiscal fourth-quarter earnings call that each humanoid robot could eventually require around 200GB of DRAM and several terabytes of NAND storage1

1

3

3

. Micron believes physical AI could become a major source of memory and storage demand by the end of the decade1

1

4

4

. An X user calculated that 10 billion robots would require 2 trillion gigabytes of RAM, far exceeding the roughly 45 billion gigabytes of DRAM the world produces annually3

3

. Tesla has reportedly placed its first large-scale component order for roughly 5,000 Optimus units and is auditing Chinese suppliers ahead of production3

3

. The company aims to eventually build 1 million Optimus units per year3

3

. The AI5 chip taped out in April, according to Musk, who believes it will be three times as efficient as Nvidia's chips on a performance-per-Watt basis and achieve comparable performance at just 10% of the total cost2

2

.