TL;DR

  • Lightpath reports a 32% annualized increase in IP backbone traffic over the past 18 months.
  • The growth reflects accelerating enterprise adoption of AI and large language model platforms.
  • AI inference workloads are shifting network demand from hyperscale data centers to enterprise environments.
  • The trend reinforces the growing importance of dense, high-capacity fiber infrastructure for enterprise AI.

# # #

For much of the AI conversation, attention has focused on the massive data centers used to train large language models. But once those models are built, the real work begins. Every time a hospital queries an AI diagnostic tool, a financial institution scores a transaction, or a university runs an AI-assisted research workload, that request travels across a network.

This stage of AI, known as inference, is becoming one of the clearest indicators of how quickly enterprises are adopting artificial intelligence. Unlike model training, which takes place inside a relatively small number of hyperscale facilities, inference happens wherever organizations are putting AI to work in their day-to-day operations. As adoption grows, the supporting network becomes just as important as the models themselves.

Lightpath’s latest network data provides a real-world view of that shift. The company reported that IP traffic across its backbone has grown at an annualized rate of 32% over the past 18 months, significantly outpacing its historical annual growth rate of 19%. Rather than being driven by a single customer or industry, the increase has been broad-based across its enterprise customer base, reflecting how AI is becoming embedded across multiple sectors.

This is an important distinction because enterprise AI generates a different pattern of network demand than model training. Training workloads are concentrated within large AI campuses, while inference creates millions of distributed requests originating from businesses, healthcare providers, financial institutions, universities and other organizations. Each interaction requires reliable, high-capacity connectivity between enterprise locations and AI platforms, placing increasing demands on metro and regional fiber networks.

As organizations integrate AI into everyday business processes, network performance becomes increasingly mission-critical. Low latency, high-capacity connectivity and resilient backbone infrastructure are no longer simply desirable characteristics—they are becoming foundational requirements for organizations that depend on real-time AI applications to support operations and decision-making.

Lightpath’s network is designed to support this growing demand through 12,100 route miles of AI-grade fiber spanning 11 major U.S. metropolitan markets and direct connections to more than 18,000 service locations. As enterprise AI adoption continues to accelerate, this type of dense metro infrastructure will play an increasingly important role in ensuring organizations can reliably access the AI platforms they depend on.

Perhaps most importantly, the announcement highlights how network traffic itself is becoming an early indicator of AI adoption. While new model launches and hyperscale investments often dominate headlines, increasing IP backbone traffic provides tangible evidence that enterprises are moving beyond experimentation and embedding AI into everyday operations.

As AI continues to evolve from a specialized technology into standard business infrastructure, the networks connecting enterprises to AI platforms will increasingly determine how effectively organizations can realize the technology’s full potential.

Read more in the press release here.