A Causal Inference Approach to Optimising Internet Peering Attractiveness

Autonomous Systems (ASes) rely on interconnection agreements to exchange traffic and ensure global Internet connectivity. Establishing mutually beneficial peering relationships offers significant cost and performance advantages, yet requires agreement based on compatibility and perceived value. While existing research focuses on how ASes select potential peers, the complementary question remains unaddressed: how can an AS strategically modify its own infrastructure to maximize its attractiveness as a peering partner? We present a causal inference framework to answer this question, estimating the impact of infrastructure investments at Internet Exchange Points (IXPs) on an AS’s peering desirability. Using the T-learner meta-algorithm with publicly available data, we quantify treatment effects for three infrastructure decisions: joining IXPs, upgrading port capacities, and leaving IXPs. Our analysis reveals substantial heterogeneity in treatment effects across network types and contexts. Joining an IXP yields the highest returns but with significant variability, while port upgrades provide smaller yet more consistent benefits. Leaving an IXP uniformly degrades peering prospects. These findings demonstrate that effective interconnection strategies must be personalized to network characteristics and local IXP environments rather than following one-size-fits-all recommendations. We demonstrate practical applicability through a case study that shows how operators can evaluate specific investment decisions and predict their impacts on individual peering relationships.