The vocabulary of energy and compute infrastructure

A data center whose generation or storage sits on the customer's side of the utility meter, so its power never crosses the grid. Behind-the-meter siting avoids the interconnection queue and retail tariffs, which is why existing mining sites convert to AI compute faster than greenfield builds.

The backlog of generation and load projects waiting for a grid operator to study and approve their connection. In most US markets the queue runs to multiple years, which is what makes behind-the-meter siting and stranded power attractive to compute developers who need capacity sooner than that.

A long-term contract to buy electricity at an agreed price. Data center developers use PPAs to underwrite new generation and to hedge power-price risk, and the agreement is usually the instrument that makes a compute project financeable in the first place.

Generation capacity that exists but cannot reach load, because it is curtailed or was built without adequate transmission. Because stranded power is cheap and available now rather than after a queue, it is a frequent starting point for new compute sites.

The capital structures behind large AI and HPC data centers: project finance, structured debt, and contracts sized against long-term offtake. Lenders underwrite the power agreement and the counterparty as much as the building, so the financing closes on the energy terms first.

The cost case for direct-to-chip or immersion cooling against air. At high rack densities air cooling becomes impractical, so liquid cooling's higher capital cost is offset by greater achievable density and a better power usage effectiveness.

Reducing a generator's output at the grid operator's instruction, usually because of transmission congestion or oversupply. Flexible loads such as bitcoin mining are paid or credited for curtailing on demand, which is what makes them useful to a grid operator under strain.

Compute that can move its consumption in time or place in response to grid conditions or price. Flexible load lets a data center behave as a grid asset rather than as fixed demand, and it is the basis of most curtailment and demand-response arrangements.