arXiv study: Data center BYOC mandates beat flexibility under supply-chain strain

A new academic paper from Tong Liu (Northwestern) and Jacob Mays (Cornell) tackles one of the most pressing grid challenges of the AI boom: how capacity markets should handle the explosive growth of data center loads. The study, posted on arXiv as 2608.06528 and submitted to Energy Economics, develops a formal model to compare two policy instruments: bring-your-own-capacity (BYOC) mandates, which require data centers to procure their own generation or storage, and flexibility accreditation, which rewards loads for being able to reduce consumption during peak periods. The authors model supply-chain constraints that limit how much new capacity can be built simultaneously—a realistic bottleneck given transformer lead times and skilled labor shortages.

The results are nuanced. Under efficient (unconstrained) pricing, BYOC has little welfare effect because it simply shifts procurement decisions between grid-built and self-built capacity; flexibility delivers a modest gain by lowering the effective capacity requirement. However, when administrative price caps are in place—as in many real markets—BYOC can force data centers to internalize the full cost of capacity and actually improve static welfare. The ranking flips depending on supply-chain stress: at low or moderate stress, only flexibility raises welfare, but under severe stress with capped prices, BYOC's welfare gain can exceed the gross flexibility benefit. The paper also finds that a unilateral BYOC mandate can crowd out capacity investment in neighboring markets, while flexibility produces essentially no spillover. Finally, the authors show that applying standard non-performance penalties to flexible loads creates incentives too weak to elicit truthful flexibility reporting, suggesting regulators need to redesign penalty schemes. With 37 pages, 4 figures, and 14 tables, this research gives grid operators and policymakers a rigorous framework for balancing data center growth and grid reliability.