Tesla is changing the configuration of future AI5 and AI6 chips due to a shortage of RAM. Elon Musk said that the company has reduced the amount of LPDDR5 in AI5 from 144 GB to 72 GB, and AI6 will now get 144 GB instead of the previously expected 216 GB.
Both processors will be used in Tesla’s fully autonomous driving computing platform and Optimus humanoid robot. AI5 has been trial-produced at Samsung’s Texas factory and is expected to be mass-produced in 2027. The chip is manufactured using TSMC and Samsung’s 2nm process technology and should provide 2000 to 2500 TOPS computing performance. Depending on configuration, the Tesla is expected to be on par with the NVIDIA H100 or Blackwell.
Musk explained that the massive reduction in memory was to ensure sufficient production. At the same time, Tesla leaders claimed that cutting DRAM will not affect the functionality of Optimus. On-chip SRAM should play a big role here: According to Musk, about half of the AI6 and AI6.5 compute accelerators will be dedicated to SRAM, which provides significantly higher throughput. This creates a rather unusual situation: future chips will require less memory, precisely because memory itself is now in short supply.
