The economics of behind-the-meter AI compute
When you own the power, your cost structure changes. We model out five years of TCO for grid-tied vs. behind-the-meter AI compute at 1 MW, 5 MW, and 50 MW scales.
The pitch for behind-the-meter AI compute is simple: when you own the power, you don't pay demand charges, you avoid grid outages, and you can site in uncongested markets. But the actual economics are nuanced. Here's our model.
Grid-tied baseline
A grid-tied 5 MW compute hall has a straightforward cost structure:
- →Energy: ~$0.07/kWh average industrial rate in our target markets. Annual energy bill: ~$3M at full utilization.
- →Demand charges: ~$10-20/kW-month. Annual demand: ~$600k-1.2M.
- →No battery capex. No battery opex.
Five-year TCO for grid-tied: about $20-25M, of which energy and demand is about $18-21M.
Behind-the-meter with batteries
Adding z1power batteries changes the math. Capex goes up ($4-6M for the battery system), but opex drops:
- →Energy: still pay for charging, but charge from the grid when prices are low. Annual energy: ~$2M.
- →Demand charges: dramatically reduced or eliminated. Battery absorbs the peak. Annual demand: ~$100-300k.
- →Battery opex: ~$200k per year for AURA + cell cycling.
Five-year TCO for behind-the-meter: about $25-30M. Higher capex, lower opex. Payback period: roughly 5-7 years on the battery alone, but the resilience value is on top of that.
Where behind-the-meter wins
The economics flip in favor of behind-the-meter when:
- →Grid prices are volatile (most US ISOs are increasingly this).
- →Outage frequency is high (CAISO, ERCOT, PJM).
- →Demand charges are punitive (urban industrial rates).
- →You need sub-10ms failover for compliance or workload reasons.
For most AI workloads today, at least two of those four apply.