In the case of Anthropic, the revenue has gone as high as $50 million per megawatt. What that now enables them to do is: 'Hey, if I spend 10 bucks on inference capacity, I actually generate 50 bucks of revenue, and then I can turn around and incrementally spend all of that profit on training.'
You might also like
problems that were once constrained by engineering talent can increasingly be attacked with capital and compute
The a16z Show
To poverty I owe the fact that I never went to college. I spent those four years in the school of experience instead of the school of theory.
Claude Hopkins — Founders
Even though it's not a huge share of the world's electricity, almost all of the world's demand is being served by a very few select locations. So you can still have this issue where you're basically powering the entire um world's AI demand from a few centers in the US and and and a few other data centers across the world. And that's a really really key local concern and puts pressure on local grids even if it doesn't consume a huge amount of the world's electricity.
The Prof G Pod