Meta prepares for gigawatt datacentres to power 'superintelligence'

Great. But no-one currently has any idea how to get there. The current approach is 'more power' 'agentic' 'prompt engineering'. It gave us bizarre things but it's not the answer and cracks are starting to show; everyone is bleeding money especially for the hard stuff we exactly WANT to solve. We need a new fundamental model as the transformer is shagged out (yeah yeah it'll make incremental improvements, I know). No one will get AGI with this model and the idea you need gigawatts to get where our brain uses almost nothing is really weird. We will get there but not this iteration.