Data centers can sell grid services instead of begging for power

Nigel Jakeman, engineering and business development director at Turbo Power Systems, argues in Data Center Dynamics that the speed-to-power wall is a negotiating problem first, even if it usually reads as an engineering one — too much load, too little wire. The two clocks that make it are hard to reconcile: venture-backed AI workloads scale in months, while transmission line upgrades take five to ten years, and between them sit operators holding billions in unexecuted capex, waiting in the utility queue or asking an already strained network for an outsized favor.

The usual bridge is a diesel generation block, and it is also the reason the door stays shut. Utilities block those, Jakeman writes, because they add volatility and emissions while bringing little systemic value to the local network; that regulatory judgment about what a large load is worth to the grid explains why a fast-deploy solution can be technically adequate and still dead on arrival at the utility.

Jakeman points to work from the Electric Power Research Institute, including its Powering Intelligence and Flex MOSAIC initiatives, which frames the data center load challenge as increasingly solvable through flexibility at the grid edge rather than through new transmission — localized infrastructure that can stabilize, shape, and support the grid in real time. The utility, in that framing, is no longer being asked for a favor but is being offered a resource.

The technical shape of that offer is where the negotiating position meets the underwriting model. HVDC backbones at 800 volts DC cut the power path from four or more conversion stages to two, reducing energy loss and infrastructural complexity, and Jakeman writes that even a modest hyperscale facility can realize efficiency improvements of several percentage points, savings he describes as compounding into material financial benefit. Conversion stages are usually treated as an engineering preference, but at hyperscale volumes a few points of loss is a line item that recurs every hour the campus runs, putting the architecture in the underwriting model next to the power contract rather than downstream of it.

Grid access, as this publication has argued, is being priced in state dockets rather than in queues, and Jakeman's proposal is a docket argument in engineering clothes: bring the utility a battery and an efficient DC backbone, and the interconnection case changes character from a drain to be managed into a service to be scheduled. The corollary is less comfortable for anyone financing diesel blocks as a bridge to a permanent connection, because on Jakeman's account of how utilities respond, those operators are carrying an approval risk that the returns on a backup generator do not obviously cover. Load-class rules in state dockets are the next variable: the first docket to credit edge flexibility will separate operators holding storage and HVDC as financed grid assets from operators holding a diesel bridge that remains dead on arrival at the utility.