The future is unknown, but I think it's clear that that there is a huge imbalance right now between supply and demand on compute. It has been this way since the introduction of Chat GPT-3.5 earlier this year. The scrum has been intense. Tesla's CEO, its prior CFO, NVIDIA's financials, and Oracle's CEO are all telling us this plain as day.The presumption that compute resources are scarce is in no way supported. Abundant resources are just as likely.
Particularly in light of Tesla's statements such as those regarding the potential to sell time on Dojo at some point.
Optimus and FSD most significant constraints are more likely to be found in the wetware trying to determine how to accomplish the task most efficiently.
You can be sure that Tesla paid a king's ransom for its 10,000 GPU H100 cluster that it just brought on line. Tesla readily paid the price, publicly thanked NVIDIA for their kindness in selling it to them, and undoubtedly wished for more. This despite the fact that Tesla's a penny-pinching miser relatively speaking.
It is said that Microsoft is investing $50 billion per annum on compute infrastructure. I wouldn't doubt it, although I have not confirmed that number. We soon may need to be thinking in those terms.
Last edited: