A lot of people are speculating that the AI industry is an economic bubble and will soon burst. And a lot of attention has gone to a figure representing cyclical relationships between chip designers and AI companies. Some some people, notably Hank Green, mistake it for the entire ai industry, while missing Google, Anthropic, Moonshot, Alibaba, all the chinese noname labs.
Now relevant to the discussion of bubble is whether or not AI will improve, and how much compute do we actually need?
In this figure, you have to ask yourself, why is there so much investment in AI compute?
Well here's the checkmate:
* if ai keeps improving, you want to be sure to have the best ai
* if ai doesn't keep improving, it means we rely on scale and that means we need to spend a lot of compute just to get somewhat useful ai systems.
All in all, it seems good to have compute either way, when thinking from perspective of an ai company.
[link] [comments]