RICHMOND, Va. — Three of the most active data-centre construction markets in the country — Northern Virginia, central Ohio, and the Hillsboro corridor of northwest Oregon — have begun to push back interconnection timelines for the largest pending projects, citing power-grid constraints that have, in their planning assumptions, swamped the curves the utilities had been working against.

The pushback is, in each market, taking somewhat different operational forms. The underlying constraint is the same: the AI-training-driven escalation in per-facility power demand has, in the past eighteen months, exceeded the utilities' generating-capacity-and-transmission-build curves by margins that none of the parties had previously planned for.

What the constraints look like

In Northern Virginia, the dominant utility has, in recent communications with developers, indicated that interconnection for facilities above a defined size threshold cannot, in some specific substations' service areas, be supported until 2028 at the earliest. The previous communicated timelines had been 2026.

In central Ohio, a parallel pattern has emerged with somewhat different specifics: the utility has indicated it can support continued build-out but at lower per-facility power densities than the developers had been planning for. The mismatch is forcing facility redesigns that affect deal economics across the pending pipeline.

What developers are doing

Developers are responding with a combination of patience and geographic redirection. Several large developers have, in the past two quarters, accelerated due-diligence on alternative markets — including markets that had not been the subject of significant data-centre development in past cycles.

The geographic redirection has visible consequences for several states that have not, until now, been substantial parties to the data-centre conversation. State and local governments in those states are, with varying degrees of formality, weighing the trade-offs between welcoming the development and managing its operational implications.

The grid-investment question

The grid-investment question that the constraints expose is one that the utility-regulator framework has been working through for several quarters. The cost of the grid investments required to support continued build-out at past pace is substantial; the question of who pays the cost — ratepayers, the data-centre developers themselves, or some combination — is the political question of the next several years.

State public-utility commissions have, with notable consistency across the affected markets, signalled a preference for cost-allocation structures that protect ratepayers from the largest portion of the build-out costs. Whether the structures the commissions ultimately approve produce build-out paces that align with developer needs is the more practical question.

The longer view

The longer view of the constraint is that it is unlikely to be permanent. Generation and transmission build-out, on past patterns, eventually catches up with demand — the question is the time path. The current quarter's developments suggest that the time path will, for at least the next twelve to eighteen months, lag the demand profile that the AI training cycle is producing.

That lag has visible consequences for the pace of AI infrastructure expansion in the affected markets. Whether it has consequences for the pace of expansion overall depends on whether the alternative markets that are absorbing redirected investment can sustain the pace they are now growing at.