2 Comments
User's avatar
Neural Foundry's avatar

Fascinating data on that 3.3x annual growth rate outpacing Moore's Law. The shift from algorithms to capital and power as the primary constraint is exactly what Meta's nuclear bet signals. Ive seen datacenters in my area already causing grid strain, so the Prometheus cluster needing dedicated nuclear makes sense. The 7-month doubling window for available compute realy changes how fast experiments can scale once funding is secured.

Yaro Celis's avatar

As models get smaller and more efficient, I wonder if all this computer power will be needed?