The mechanism is called non-firm or flexible interconnection. Under a conventional firm agreement, a utility must be able to serve the load at all times, which means the connection waits until transmission and generation can support the worst-case hour. Under a non-firm agreement, the load accepts curtailment during defined stress conditions and connects against existing capacity instead.
Cornell research published this month frames the arrangement as a tradeable commodity: a large load that can reduce consumption creates value that can be priced and exchanged, rather than simply absorbed as a reliability obligation. That framing matters because it converts flexibility from a concession into an asset with a market.
Why Data Centers Are Unusually Good Candidates
Not all large loads can curtail. A smelter cannot interrupt a pour. A data center, by contrast, runs a workload mix that is partially deferrable. Training runs can pause and checkpoint. Batch inference can shift hours. Only a fraction of the load is genuinely real-time, and that fraction can be carried by on-site generation or storage.
The commercial question is what the flexibility is worth. Years of earlier revenue against some number of curtailed hours is an arithmetic problem, and for most operators facing a five-to-seven-year queue, the arithmetic favours flexibility by a wide margin.
Where This Leads
Parselion's assessment: flexible interconnection is likely to become the default structure for new large loads in constrained regions within two years, and the Google and Constellation agreement has already normalised the load-shaping commitment at hyperscale. Operators should begin instrumenting their workload mix now to understand precisely what fraction of load is curtailable and at what notice period, because that number will shortly determine both queue position and connection cost.