Network traffic foundation models: A systematic review

Network Traffic Foundation Models (NT-FMs) aim to enable a “train-once, adapt-anywhere” paradigm for automated traffic management, yet current research remains fragmented. This systematic literature review analyses 51 primary studies published between 2017 and 2025. To evaluate the feasibility of developing NT-FMs, we address 4 research questions: (RQ1) we map prevailing AI architectures and their training regimes, including self-supervised pretraining objectives; (RQ2) we assess how network traffic is represented for input to NT-FMs; (RQ3) we examine the network environments and tasks to which NT-FMs have been applied; and (RQ4) we survey publicly available datasets suitable for training NT-FMs. Our content and bibliometric analyses indicate a rapidly expanding body of work, yet cross-task transfer remains limited; terabyte-scale corpora are scarce; and efficiency techniques are only beginning to emerge. Taken together, these findings indicate that NT-FM development is feasible but currently constrained by data scale and transfer limitations. We distil open challenges and propose a research agenda aimed at scalable, robust, and reliable NT-FMs.