Tesla is ramping up its own chip production, with Elon Musk confirming that the Dojo 3 supercomputer project is back after a pause last year. The company paused it to focus on stabilizing their AI5 chip design, which now looks solid enough to power the next big push. This move aims to cut reliance on Nvidia gear for training AI models needed for self-driving tech and robots like Optimus.
Key Chip Details
AI5 packs performance like Nvidia's Hopper in a single chip, and matches Blackwell levels when pairing two together. It runs way more efficiently too, around 250W versus Nvidia's 700W+ for similar output, which helps with power-hungry data centers. Tesla plans to make these chips in huge volumes using TSMC and Samsung fabs, keeping costs low at pennies compared to rivals.
Dojo 3 Plans
Dojo 3 will leverage AI5, AI6, or even AI7 chips in a pure in-house setup, skipping Nvidia mixes from earlier versions. Musk eyes it for massive AI training, possibly even space-based compute to dodge Earth's energy limits. They're hiring engineers now for what could be the world's highest-volume AI silicon production, with new gens every nine months up to AI9.
Production Timeline
Samples of AI5 hit late 2026, scaling up in 2027 for cars like Cybercab, Optimus bots, and data centers. Tesla's going all-in on fabs and design to match software perfectly with hardware. Challenges remain in scaling complex chipmaking, but the strategy balances cost savings against Nvidia dependence.
