Cummins' 5-MWh battery masks the real scandal: who pays when data centers skip the grid bill

Cummins announced in August 2026 that its Power Generation business would supply a 5-megawatt-hour lithium iron phosphate battery energy storage system (BESS) to a large U.S. data center project, the company's largest deployment to date.^[1] The system uses liquid cooling and flexible DC architecture to rapidly charge and discharge, smoothing AI workload spikes that would otherwise create power swings at the grid connection point.^[1] The announcement frames batteries as the answer to a real technical problem: AI training loads can cause electricity demand to swing wildly, destabilizing the interconnection and making utilities uncomfortable.^[1] But the news release omits everything that matters to ratepayers. What data center? Which utility? What are the terms of the grid connection contract, and is it confidential?

The battery itself is not the scandal; the scandal is the contract structure hidden behind it. When a data center operator opts for BESS, one of two things happens. Either the operator brings its own generation (the battery, solar, or contracted clean supply) and keeps the cost and performance risk, or the utility builds firm generation and transmission to serve the facility, assigns those assets to a general customer class, and spreads the cost across all ratepayers if the load underperforms or the facility shuts down early. The Cummins announcement does not disclose which regime applies. No term sheet is public. No minimum-take clause, no collateral requirement, no cost-isolation language has been filed for review. The silence itself is the mechanism: utilities routinely negotiate special contracts with hyperscalers and file them with state commissions under partial confidentiality redactions, keeping the price and often the customer identity sealed while the underlying capex and its cost allocation remain ratepayer-funded.^[2]

What matters is whether the data center's volatile load is being used to justify new utility-owned generation and transmission that will sit underutilized or stranded if the facility moves, scales down, or brings its own supply. The battery announcement suggests the operator understands the grid constraint, and has chosen to solve it without asking the utility to build firm capacity. That is the healthier outcome. But if the utility has already filed a rate case claiming that this data center (or a pipeline of them) requires $1 billion in new gas plants or transmission backbone, and that case used an inflated forecast of announced (not contracted) load to justify the capex, then the battery news is actually a marker that the load forecast was wrong, the capex may be stranded, and ratepayers are on the hook. The question to pose into every data-center docket: Is the load served by the operator's own generation and storage, or is the utility building firm capacity and socializing the cost?^[3]

The protective tariff response is straightforward. A large-load tariff (now adopted in roughly 23 states) establishes a standing customer class for facilities at or above 20, 25 megawatts, with standardized rules: long minimum-take commitments (10, 14 years or more), high demand ratchets (the customer pays for 85% of contracted transmission and 60% of generation capacity whether it uses it or not), collateral ($1.5 million per megawatt or more), and cost isolation so the data-center class, not residential customers, absorbs the loss if the load doesn't materialize.^[4] Virginia's GS-5 and Ohio's AEP precedents show the structure works; it lets utilities offer firm service with guaranteed returns while shifting stranded-cost risk from ratepayers to the customer who created it. The Cummins battery announcement does not prove such a tariff is in place here. A concrete intervention is available now: demand into your state's pending or current data-center rate case that any special contract or large-load tariff include a high minimum-take ratchet, collateral, and cost isolation. If the utility claims the load requires new generation but the operator is deploying its own BESS, that contradiction belongs in the record.