Xebec
Well-known member
A few interesting "Semi relevant" details from the Q1 2026 Tesla Earnings call today:
- Tesla is spending $3B towards TeraFab this year - to build a Research Fab, to "try out new physics", "will probably produce 1,000 wafers/month or so" (acknowledging very small scale initially). He indicated they have some long shot ideas for improving semi fabrication they want to try.
- TeraFab full scale plan still in the works, board members and others between SpaceX and Tesla must be consulted first before plan is finalized.
- Musk: "Terafab is NOT to get leverage over our chip suppliers, it is to ensure that we have enough chips down the road" -- "the rate that the industry is growing, both in logic and memory [is not enough to meet our needs unless we also make chips ourselves]"
- Regarding the Intel partnership - ~ [still being figured out], but "Tesla will use Intel 14A in the future, [by the time it is mature]". He sung praise on LBT and a few others at Intel.
- AI Hardware 4 is getting a minor upgrade later this year - 10% more bandwidth and ASIC clock speed, 2X the memory (32GB per AI4 chip). Samsung is doing the engineering for Tesla for this, [we are beholden to Samsung for the date]. (Note: probably some vagueness here to avoid the Osborne effect).
- AI5 not coming for Tesla Self driving right away even after availability, AI4 will be around a long time. HW3 not able to meet FSD unsupervised, we will have to build large depots for AI4/HW3 retrofits due to the amount of work (cameras, AI hardware, etc) and for the opportunity to convert all HW3 cars to the Robotaxi fleet if desired.
Webcast: https://ir.tesla.com/#quarterly-disclosure
P.S. These are my semi-paraphrased human-made notes.
- Tesla is spending $3B towards TeraFab this year - to build a Research Fab, to "try out new physics", "will probably produce 1,000 wafers/month or so" (acknowledging very small scale initially). He indicated they have some long shot ideas for improving semi fabrication they want to try.
- TeraFab full scale plan still in the works, board members and others between SpaceX and Tesla must be consulted first before plan is finalized.
- Musk: "Terafab is NOT to get leverage over our chip suppliers, it is to ensure that we have enough chips down the road" -- "the rate that the industry is growing, both in logic and memory [is not enough to meet our needs unless we also make chips ourselves]"
- Regarding the Intel partnership - ~ [still being figured out], but "Tesla will use Intel 14A in the future, [by the time it is mature]". He sung praise on LBT and a few others at Intel.
- AI Hardware 4 is getting a minor upgrade later this year - 10% more bandwidth and ASIC clock speed, 2X the memory (32GB per AI4 chip). Samsung is doing the engineering for Tesla for this, [we are beholden to Samsung for the date]. (Note: probably some vagueness here to avoid the Osborne effect).
- AI5 not coming for Tesla Self driving right away even after availability, AI4 will be around a long time. HW3 not able to meet FSD unsupervised, we will have to build large depots for AI4/HW3 retrofits due to the amount of work (cameras, AI hardware, etc) and for the opportunity to convert all HW3 cars to the Robotaxi fleet if desired.
Webcast: https://ir.tesla.com/#quarterly-disclosure
P.S. These are my semi-paraphrased human-made notes.