Another great example of how every market reaction is always an overreaction. Even more efficient AI will need more vector processors than current global production capacity is providing.
It's how all this nonsense is always played out though, the real science will continue, but the market speculators will continue to lose their shirts. It always had the wiff of a bubble, and so it proved.
They don't even need fewer GCUs (if that even turns out to be true), the demand for AI still outpaces GCU manufacturing. Perhaps now it only outpaces by 10x instead of 1000x.
All the stuff that DeepSeek used was publicly available (in papars and libraries), so I do not understand why they did not crate a version with the alternative approach as well
Comments
I would think xAI, OpenAi, Microsoft, Tesla, Meta, Alphabet etc etc etc are more at risk.