Sponsor Editorial

Capital spending on AI infrastructure has hit new highs, with leading technology companies projected to invest over $630 billion in 2026. This boom is driven by large language models, which need far more computing and networking than older deep learning systems. One of the current leading beneficiaries of this boom, NVIDIA, anticipates $1 trillion in GPU orders through 2027. Connectivity remains a key component in the AI data center. Moving from scale-up within the rack to scale-out across racks, and ultimately to scale-across clusters of adjacent data centers, the underlying network shapes the performance of these “AI factories” that consume and produce tokens. The network is no longer one design problem; it is three coupled design problems, each with different physics, economics, and operational constraints.

In this 2026 edition of their data center networking report, AvidThink examines the advances and challenges across three network domains, along with the latest in communication libraries, robustness, and fault tolerance.