In order to satisfy the high bandwidth and performance demands of applications, host servers are built with multiple network interfaces, and a data center network consists of multiple redundant links. It is important to make efficient use of all the available network capacity, using multiple physical paths whenever possible, but traditional forwarding mechanisms using a single path are not able to take advantages of available multiple physical paths. The state-of-the-art MPTCP (Multipath Transmission Control Protocol) solution uses multiple randomly selected paths, but cannot give total aggregated capacity. Moreover, it works as a TCP process, and so does not support other protocols like UDP. This paper presents an alternative solution using adaptive multipath routing in a Layer-2 network with static (capacity and latency) metrics, which adapts link and path failures. This solution provides in-network aggregated path capacity to individual flows, as well as scalability and multitenancy, by separating end-station services from the provider’s network. The results of deploying a proof-of-concept prototype on a data center testbed, which show the aggregated path capacity per flow, demonstrate an improvement of 14% in the worst bisection bandwidth utilization, compared to the MPTCP with 5 subflows.