Part 7/12:
The engineering behind Optimus is also evolving, with neural networks for humanoid robots being substantially larger than those for Tesla’s vehicles. Tesla’s new supercomputer platform, Dojo 2, scheduled for release in 2026, will focus on balancing cost and scalability rather than raw compute power—an approach motivated by Tesla’s desire to produce more cost-effective AI solutions tailored specifically for FSD and Optimus.
Further, Tesla is developing dedicated ASIC chips for training AI models, expecting that multiplying compute capacity by ten times and leveraging cheaper chips could be a significant breakthrough—a potential unlock for widespread robotic deployment.