Lightpath builds AI-grade fiber for two 1-gigawatt data center campuses

The fiber-optic network that connects AI models to the real world is becoming as critical as the GPUs that run them.

Lightpath on Wednesday announced new fiber builds to support two hyperscale data center campuses now under construction, each planned to exceed one gigawatt of capacity, as surging artificial intelligence demand forces a rethinking of network infrastructure.

"The scale of these campuses demands a network that can handle sustained, high-volume traffic with near-zero latency," said Chris Morley, chief executive officer of Lightpath. "Our AI-grade fiber is built for the throughput requirements of distributed training and real-time inference workloads."

Each campus will require dedicated fiber routes with direct connections to major internet exchanges and cloud on-ramps, Lightpath said. The company did not disclose the locations of the campuses, the expected completion timeline, or the total investment amount. The builds extend Lightpath's existing network of more than 22,000 route miles of fiber across the New York metropolitan area and parts of New England.

The announcement highlights a structural shift in how data center infrastructure is being planned. A single gigawatt campus can power roughly 750,000 homes — a scale that was virtually unheard of in the industry three years ago. The transition from human-driven internet traffic to machine-to-machine AI workloads is changing the economics of fiber deployment, with operators racing to secure long-term contracts with hyperscalers before capacity is spoken for.

The Fiber Bottleneck in the AI Era

The challenge is not simply adding more fiber. AI workloads generate fundamentally different traffic patterns than traditional cloud applications. Training large language models requires moving terabytes of data between GPU clusters across multiple data centers, while inference demands sub-millisecond response times for real-time applications such as autonomous driving and financial fraud detection.

Legacy networks built for best-effort internet routing struggle to maintain consistent latency under the bursty, high-volume traffic that AI generates. Lightpath's AI-grade fiber uses dedicated wavelength paths and direct cloud interconnects to bypass the congestion points that plague public internet routing.

The company faces competition from established players including Zayo Group, Crown Castle, and Lumen Technologies, all of which have announced AI-focused fiber initiatives in the past year. Zayo in March said it would invest $500 million to expand its long-haul fiber network to support AI workloads, while Lumen has signed more than $5 billion in AI-related connectivity deals since mid-2024.

What It Means for Investors

For investors tracking the AI infrastructure buildout, the fiber layer represents a less visible but equally essential component of the ecosystem. While Nvidia's GPU shipments and data center REITs such as Equinix and Digital Realty capture most of the attention, the companies that own the physical fiber connecting these facilities stand to benefit from multiyear, non-discretionary spending cycles.

Lightpath, a privately held company backed by private equity firm Altice USA, does not disclose its financial results publicly. But the broader fiber infrastructure market is projected to grow at a compound annual rate of 12 percent through 2030, according to Grand View Research, driven largely by AI and cloud demand. Publicly traded peers such as Crown Castle and American Tower trade at 18 to 22 times forward funds from operations, reflecting investor expectations of sustained capital spending.

This article is for informational purposes only and does not constitute investment advice.