Korva is a scroll of the internet’s best ideas.

Spread data centers like sand
The rock and the web
Picture a rock suspended in a spider's web. With one big rock, it is hard to find a spot on the web where it won't break the whole thing. With a lot of sand, the load spreads across the web, and the same web can hold much more.
Spreading the load
A large centralized data center is the rock. Put the load in places where the grid has headroom instead, and no single spot has to take hundreds of megawatts, while the network as a whole still takes hundreds of megawatts. More compute connects, and it connects faster. Batteries beside the compute can condition the power at a very high rate, because Base controls the hardware, so the grid doesn't take a bad hit.
Base's grid data
Base's firmware and software already pull telemetry from the batteries in more than 17,400 customers' homes and map the grid at a hyperlocal level, in more detail than utilities themselves have. Zach Dell's plan to point that capability at AI has two parts: bringing the power to the compute and bringing the compute to the power.
Why do grids limit how much a data center adds?
Each connection point has a fixed capacity in its lines, transformers, and substations. A new load of hundreds of megawatts forces the utility to study whether the local equipment can carry it, and upgrades can take years to build. A single megawatt can power roughly 750 to 1,000 average US homes, so a campus of a few hundred megawatts equals a city's worth of demand. That mismatch is why data center developers often wait in line for a grid connection.
How can batteries smooth out AI computing spikes?
Large AI training clusters can swing their power draw quickly as jobs start, pause, and synchronize. Power electronics in a battery can respond in milliseconds, absorbing the swings before they reach the grid. The grid reliability body NERC has documented a 2024 incident in Virginia in which about 1,500 megawatts of data center load dropped off the grid at once. Events like that have made large sudden changes in demand a reliability concern for operators.
How does a spider web actually hold weight?
An orb web has radial threads that run from the center to the frame and spiral threads that wrap around them. A load on any one point pulls on several radial threads at once, which spreads the force across the whole structure. Spider silk can stretch by roughly a third before breaking, which lets it absorb the energy of an impact. A web that catches a flying insect at full speed relies on that stretch to avoid snapping.
More from Packy McCormick
