Lightpath sees 32% IP traffic growth driven by AI inference
Lightpath, the all-fibre metro connectivity provider jointly owned by Optimum Communications and Morgan Stanley Infrastructure Partners, has reported that IP traffic across its network grew at an annualised rate of 32% over the most recent 18-month period. The company attributes the acceleration to enterprises embedding AI and large language model platforms into day-to-day operations, generating a steady stream of inference requests that must traverse the backbone.
The 32% annualised figure represents a sharp step up from the company's historical growth rate of 19% per year. Lightpath says the increase is broad-based across its customer base in healthcare, financial services and education, rather than concentrated in a single vertical or client.
What is driving the traffic
The mechanism Lightpath highlights is inference rather than training. Model training is centralised; inference is distributed, occurring each time an end-user or automated system queries a deployed model. A hospital querying a diagnostic tool, a bank running real-time transaction scoring, or a university processing research workloads all generate round-trip traffic to an AI platform and back. As organisations move from piloting to embedding these tools in production workflows, that incremental volume compounds across millions of requests per day.
Chris Morley, chief executive of Lightpath, said: "Hospitals, banks and universities are querying AI platforms in real time and acting on what comes back. Every one of those interactions has to cross a network, and increasingly it crosses ours."
Lightpath operates 12,100 route miles of fibre across 11 major US metro markets, with a particularly dense footprint in the New York metropolitan area and direct connections into more than 18,000 service locations. The company positions this metro density as a structural advantage for latency-sensitive inference traffic.
Market context
Lightpath's traffic data adds a carrier-level data point to a wider body of evidence that AI inference is beginning to reshape enterprise network consumption. Hyperscalers and CDN operators have reported similar trends, and a number of dark-fibre and wavelength providers are accelerating metro build programmes in anticipation of continued growth.
The competitive landscape for metro fibre in the north-eastern US is crowded, with Zayo, Crown Castle Fiber and several regional operators vying for enterprise and data-centre interconnect business. Lightpath's differentiation rests on its owned-infrastructure model and its depth of on-net buildings, which reduces the need for last-mile handoffs that can introduce latency variance.
From a regulatory standpoint, US fibre infrastructure investment has been supported by federal broadband funding programmes, though enterprise metro connectivity is largely outside the scope of those schemes. The more pressing policy consideration for Lightpath's AI-driven customers is data residency: enterprises in regulated sectors such as healthcare (HIPAA) and financial services increasingly prefer on-net, in-metro routing to minimise the risk of data crossing jurisdictional boundaries unintentionally.
Lightpath said it will continue tracking IP traffic as an early indicator of AI adoption across the markets it serves, suggesting the company intends to use these figures as a recurring marketing and investor signal as demand evolves.