Every Claude Code user has hit a rate limit this month.
Anthropic's S-1 kind of explains why. They say they've "consistently operated in a compute-constrained environment where customer demand exceeds available compute."
And they flag power as "a significant constraint on data center development" in the US.
We pay per token, but behind every token there's a GPU sitting in a building that needs a lot of electricity. The chips keep improving fast. Getting new power connected takes years.
So honestly I think the bottleneck for our agents is more about megawatts than prompts at this point.
That's the side of AI I follow, companies like $IREN that already have the power.
I'm digging into who actually has the megawatts to back all of this. That's probably my next post.