STARIXNet: Walmart's new AI slashes cloud costs by 10-50%

STARIXNet is a lightweight deep learning model designed for real-time resource allocation in cloud platforms. Developed by Ahmed Abdulaal and colleagues, it addresses a key limitation of current solutions: most only look at CPU usage (univariate) and treat scaling as a pure forecasting problem. STARIXNet instead captures multivariate spatio-temporal relationships among multiple system metrics—seasonal, temporal, auto-regressive integrated, and exogenous patterns—to make holistic scaling decisions. It prioritizes service stability over raw forecast accuracy, using an aggregation policy that first ensures robustness, then cost-efficiency.

Deployed in critical production microservices at Walmart, STARIXNet delivered tangible cost savings ranging from 10% to 50%, along with intangible benefits like improved service stability and customer experience. The model is computationally lightweight, making it practical for large-scale, real-time deployments. The paper is currently under review and available on arXiv (2606.07565).