Tesla tapes out its AI5 self-driving and robotics chip

On 15 April 2026 Tesla taped out AI5, its next-generation custom inference chip — meaning the design was finalised and sent to the foundry for fabrication. Tesla designs the silicon in-house specifically for its own neural-network workloads and is dual-sourcing production across Samsung and TSMC to secure supply.

Designing its own AI chips, rather than buying general-purpose GPUs, lets Tesla optimise performance-per-watt and cost for exactly the vision-and-driving models it runs in every car and in Optimus. Tesla says a single AI5 matches an Nvidia H100's inference performance for its workloads at far lower cost and power — the kind of edge that matters when the chip ships in millions of vehicles.

It is also a hedge against the GPU supply crunch squeezing the whole AI industry: by controlling its own inference silicon, Tesla isn't bidding against every other AI company for the same scarce parts. AI5 underpins the unsupervised-autonomy and robotics roadmap, with successor chips already in development.