XAI and Tesla are buying 1 million Nvidia B200/B300 chips over the 14 months from February, 2025 to April 2026. This is about 20% of Nvidia's production. If Tesla can make the Dojo 2 chip at scale and volume production in the second half of 2025 this is huge. Tesla has made and installed at least 11,000 Dojo 1 chips but they only have 362 Teraflops of compute. The Dojo 2 chips have ten times the compute.

If Tesla has 25 Dojo 2 chips per wafer and orders 20,000 wafers from TSMC then this would be 500,000 Dojo 2 chips. This would surpass AMD in AI GPU chip production at 400,000 to 500,000 and put Tesla in second place behind Nvidia. Nvidia will make about 5 million B200-B300 chips in 2025.

If Tesla goes to Dojo 3 chips in late 2026 and buys from itself and XAI buys from Tesla this could be multi-million chip order for 2026-2027.

TSMC has described the Tesla Dojo chip process enabling over 40 times the compute performance by 2027. This would put in at about 15 Petaflops or near the performance of an Nvidia B200 at 20 petaflops.

IN April 2024, at TSMC's North American Technology Symposium, the company says Dojo's next-generation training tile — the D2, which puts the entire Dojo tile onto a single silicon wafer, rather than connecting 25 chips to make one tile — was already in production.

In July , 2024 AI5 is ~18 months away from high-volume production. This would be end of 2025.

In March 2025, Elon said that the Tesla Dojo 2 chip will be in volume production in a few months. The Tesla Dojo 2 chip has ten times the performance of the Dojo 1. If Tesla Dojo 2 is ten times better than the Dojo 1 training chip then it would be 3.6 petaflops for fp8. This would be several time better than the Nvidia H100 and maybe 40% of the B200.