To support modern applications, computer networks perform a plethora of auxiliary functions beyond basic application data forwarding. Obvious examples include serialization and encryption, triggered on most data transfers, but also control and monitoring that analyze and update network states. Unfortunately, the unprecedented increase in application demand and a concurrent slowdown in the scaling of compute capability make it increasingly challenging to maintain these functions performantly and cost-effectively. On the one hand, continued exponential increases in network link speeds have led to the majority of congestion events occurring at microsecond time scales, diminishing the effectiveness of current control and monitoring protocols. Conversely, adding supporting resources (e.g., bandwidth, processing cores, and power budget) incurs expensive costs at scale, entailing not only capital and operating expenditures but also carbon footprint. In this dissertation, we characterize and explore a zero-waste design approach by unlocking the potential of widespread in-network waste and present three case studies for auxiliary functions spanning across data, control, and management planes: (a) OrbWeaver, a weaved stream abstraction that reuses IDLE cycles in Ethernet links at 100s of ns granularity for state-of-the-art in-band control protocols; (b) Mantis, a switch-local reaction framework that recycles switch-local resources and co-designs them with the programmable data planes for user-defined and fine-grained (at 10s of µs granularity) closed-loop control functions; and (c) Beaver, an optimistic gateway marking primitive that reduces the waste of additional servers and instrumentation cost to enable partial snapshots `in-situ' for diagnosing distributed cloud services with near-zero impact to existing service traffic. We also show that it is possible to integrate these functions performantly at near-zero cost. The dissertation concludes with a vision for zero-waste networked systems, where we instantiate zero-waste designs to maximize the utility of residual network capacity despite existing efforts toward high-efficiency designs. More broadly, we posit that a grand challenge of our computing infrastructure is pushing waste to its limits amidst technology scaling slowdowns and increasing environmental concerns. This dissertation invites us to rethink the design patterns for networked system and outlines a spectrum of opportunities to advance this goal.