The binding constraint moved from chips to substations

A campus that phases in from 2028 is not waiting on wafers. Nvidia taking equity in an energy developer, rather than in a cloud, is the clearest signal yet that the scarce input in this cycle is interconnection queue position.

Read the Ohio announcement for what it does not say. There is no shortage language about chip supply, no allocation discussion, no fab commitment. What there is: an energy developer, a specific county, a gigawatt figure, and a 2028 start. That is the vocabulary of grid interconnection, not semiconductors.

The $1.5 billion Nvidia is putting into SB Energy is the tell. If capacity were gated on silicon, the rational investment would be upstream — packaging, memory, foundry capacity. Investing in the party that secures land, power purchase agreements and a place in the interconnection queue says the constraint is downstream of the chip entirely.

This reframes a number of arguments that have been conducted as though they were about compute scarcity. Model pricing, capacity allocation between training and inference, the geography of new builds — these look like consequences of a chip market when read from the outside, and increasingly look like consequences of an electricity market when read from the inside.

It also puts a floor under the timeline. Chip supply can respond to demand within a product cycle. Substations, transmission upgrades and generation cannot. A 2028 phase-in for capacity announced in 2026 is not conservatism; it is roughly how long the physical work takes. The financing structure behind it is the part that is novel — the schedule is just physics.

Forbes — Why Nvidia Is Building Data Centers, Not Just The Chips Inside Them → · CNBC — Nvidia backing $105 billion in financing for OpenAI data center in Ohio →