If You’re Betting on AGI, Follow the Bottleneck Down the Stack

Models are getting better and cheaper. Compute demand is exploding. The scarce asset may increasingly be something much less glamorous: a megawatt that is already connected to the grid.

If you believe in the AGI thesis, there are several obvious places to put money.

The labs. The chips. Maybe the hyperscalers. But the further this cycle develops, the more interested I become in something considerably less fashionable: electricity. Or, more precisely, not electricity itself.

Available, reliable, grid-connected megawatts in the right place…