- Home
- Current Issue
- Article
Is Spare Capacity Worth Its Embodied Carbon? Distributionally Robust Capacity Planning for Carbon-Aware Cloud Data Centers under Grid-Decarbonization Uncertainty
Authors
1 K J Somaiya College of Science and Commerce, Mumbai, Maharashtra, India.
Article Information
- Published In
- Volume 5, Issue 3 (2026)
- Published Online
- September-December 2026
- Pages
- 906-911
- Article Type
- Original Article
Abstract
View PDFCarbon-aware scheduling shifts flexible work into hours when the grid is cleaner, but it can only do so if servers are idle at those hours, so most evaluations implicitly assume that spare capacity exists. Spare servers are not free: their manufacturing (embodied) emissions are incurred at purchase, while the operational savings they enable accrue over a lifetime whose length is uncertain, against grids whose future carbon intensity is uncertain too. We formulate the decision of how much spare capacity to build, and where, as a two-stage distributionally robust program. The first stage chooses slack in each region and pays its embodied carbon; the second stage, for every operating year and decarbonization pathway, solves a linear program that shifts flexible load into low-carbon hours subject to capacity and idle power. Ambiguity over pathways and server lifetimes is a total-variation ball around nominal scenario weights, and the inner supremum is dualized so that the whole problem is one linear program. We show that each scenario’s incremental carbon is convex in the slack vector and give a break-even condition comparing embodied intensity with the worst-case expected marginal operational savings. On a four-region, 100 MW case study, the 1.7× load provisioning assumed in prior carbon-aware scheduling studies adds 68 kt CO₂e over the hardware lifetime, because the embodied carbon of 44.5 MW of extra servers is never recovered. The stochastic optimum builds 4.7 MW of slack in the solar-rich region only, with an expected lifetime saving of just 0.6 kt, and the robust optimum builds none once the ambiguity radius reaches 0.25. Spare capacity pays robustly only for low-embodied hardware (≤ 1,000 t CO₂e per MW), which points to server reuse and burst capacity as the carbon-efficient source of scheduling slack.
Related Articles
Share Article
*Instagram doesn't support direct link sharing from web. Copy the link and share it in your Instagram story or post.