First-hand information for everyone
This paper looks at the U.S. interconnection queues—the systems that projects use to request access to the electric grid. As of 2025 those queues hold about 8,200 projects totaling roughly 2,061 gigawatts. Yet only 13% of the capacity that entered the queues from 2000–2020 has reached operation. The authors argue that the queue rules can cause self‑reinforcing chains of project withdrawals when projects are re‑studied and costs are reallocated to remaining projects, even if study processing is not slow.
To test that idea the researchers treated the queue as a complex adaptive system and used two methods. First, they ran statistical tests on project withdrawal timing and co‑withdrawal across seven regional grid operators, using project‑level data from Lawrence Berkeley National Laboratory through 2025. Second, they built a simplified network simulation that models how shared upgrade costs can propagate losses from one project to others. They also tested “circuit‑breaker” style interventions that cap how much cost can be shifted to any one neighbor.
The data analysis shows clear clustering of withdrawals in time. The authors report dispersion indices between 5.0 and 102.4 (all with p < 0.001) and identify 39 monthly bursts of withdrawals. The largest burst was 67.4 times the regional monthly mean. Withdrawals were also concentrated within technology groups (a 1,000‑permutation test gave z = 2.58–4.78, p ≤ 0.005). However, co‑withdrawal within cohorts (projects behaving together) was statistically significant in only four of the seven regions and amounted to a change of 0.5–1.8 percentage points.
The simulation results show how outcomes depend on network connectivity and cost rules. At low to moderate connectivity (an average degree k = 3–10) the model produced modest net cascade amplification of about 1.0–1.5 times the initial shock. But at high connectivity (k = 20) the model shows a sharp, phase‑transition‑like move to systemic instability: a 10% initial shock could amplify more than 8.7 times and lead to cascades affecting about 97.7% of projects in the stylized network. That extreme behavior disappears in the model when costs are redistributed pro‑rata (proportionally) rather than passed to a few neighbors.
The authors also find that circuit‑breaker style caps on per‑neighbor cost reallocation can reduce cascades in the model. Under moderate connectivity these caps cut modeled cascade magnitude by up to 20% and reduced the average per‑neighbor cost transfer by 44% compared with the baseline rule in the simulation. From these results the paper frames the interconnection queue as a piece of critical infrastructure that can be vulnerable to cascading withdrawals, and it shows that the severity of that vulnerability depends strongly on how shared costs are allocated.
There are important caveats. The network model is stylized and simplifies many real‑world details; its extreme instability requires specific connectivity and cost assumptions. The statistical tests document clustering but do not prove the exact causal mechanism for withdrawals. Co‑withdrawal effects were not uniform across the seven regions. In short, the paper identifies a plausible and measurable vulnerability and shows that design choices about cost allocation and caps can change outcomes, but further work would be needed to map these model results directly onto every real queue and rule set.