3 ms·

Two years ago Elon actually sounded a warning about Dojo. I recall him mentioning this in an earnings call: “And we’re pursuing the dual path of Nvidia and Doj

by ivape 1y ago

Two years ago Elon actually sounded a warning about Dojo. I recall him mentioning this in an earnings call: “And we’re pursuing the dual path of Nvidia and Dojo. But I would think of Dojo as a long shot. It’s a long shot worth taking because the payoff is potentially very high. But it’s not something that is a high probability. It’s not like a sure thing at all. It’s a high‑risk, high‑payoff program. Dojo is working, and it is doing training jobs, and we are scaling it up, and we have plans for Dojo 1.5, Dojo 2, Dojo 3, and whatnot. So I think it’s got potential, but I can’t emphasize enough: high risk, high payoff.” https://thetranscript.net/transcript/5550/tesla-q4-2023-earnings-call-transcript https://thetranscript.net/transcript/5550/tesla-q4-2023-earn... In other words, he admitted it’s a long shot to replace Nvidia with their own chip awhile ago. It’s interesting because Google hasn’t given up on their TPUs, but Elon being a hardware guy mostly, has.

kranke155 1y ago
I doubt Elon had the time or the inclination to build yet another hardware stack, and he probably saw something that showed him he couldn’t quite catch up to Nvidia? The scale of Nvidia manufacturing and RnD should be staggering atm. Google built the entire software and hardware stack for TPUs before the AI boom. I can’t even imagine how hard it would be to hire GPU hardware talent right now.
- bubblethink 1y ago
  Really hard to do training hardware as the software stack is quite complex and needs to be general and flexible. Inference on the other hand is easier, which is what a lot of new hardware will be. I doubt if it makes sense for Tesla to build hardware for datacenter inference. Xai on the other hand could benefit from it.
mrandish 1y ago
> it’s a long shot to replace Nvidia with their own chip awhile ago It's also worth mentioning that having their own ambitious effort in secret development was a valuable BATNA (negotiating tactic) for long-term contracts with outside suppliers. The value of that recent >$16B deal with Samsung for AI chips thru 2033 may have benefited from the prospect of Dojo as a potential in-house alternative.

by ivape 1y ago

kranke155 1y ago

I doubt Elon had the time or the inclination to build yet another hardware stack, and he probably saw something that showed him he couldn’t quite catch up to Nvidia? The scale of Nvidia manufacturing and RnD should be staggering atm. Google built the entire software and hardware stack for TPUs before the AI boom. I can’t even imagine how hard it would be to hire GPU hardware talent right now.

bubblethink 1y ago
Really hard to do training hardware as the software stack is quite complex and needs to be general and flexible. Inference on the other hand is easier, which is what a lot of new hardware will be. I doubt if it makes sense for Tesla to build hardware for datacenter inference. Xai on the other hand could benefit from it.

mrandish 1y ago

> it’s a long shot to replace Nvidia with their own chip awhile ago It's also worth mentioning that having their own ambitious effort in secret development was a valuable BATNA (negotiating tactic) for long-term contracts with outside suppliers. The value of that recent >$16B deal with Samsung for AI chips thru 2033 may have benefited from the prospect of Dojo as a potential in-house alternative.