Speaking during a recent RCR webinar, AFL’s Noah Taylor said that as hyperscalers expand AI deployments, the networks linking data centers are becoming increasingly important alongside the facilities themselves
In sum – what to know:
DCI scales rapidly – Taylor said data center interconnect is growing as fast as, or faster than, data center construction as hyperscalers expand AI infrastructure.
Upload traffic shifts – AI inference is driving significantly more upstream traffic, increasing pressure on optical, metro and access networks.
Three pillars matter – Taylor argued that density, deployment speed and ecosystem integration are the key ingredients for building AI-ready interconnect infrastructure.
The rapid growth of AI workloads is increasing demand for both compute capacity and data center interconnect (DCI) infrastructure, according to Noah Taylor, head of market intelligence and growth strategy for broadband at AFL.
Speaking during a recent RCRTech webinar, Taylor argued that as hyperscalers expand AI deployments, the networks linking data centers are becoming increasingly important alongside the facilities themselves.
“Computational demand has skyrocketed,” Taylor said, noting that ChatGPT reached 100 million users within 60 days of its launch and adding that the industry is now seeing AI agents requiring “over 100 to 1,000 times” more compute than before.
According to Taylor, rising compute demand is being matched by growing bandwidth requirements across AI-scale networks. He argued that while much industry attention focuses on new data center construction, data center interconnect is expanding just as rapidly.“What a lot of people don’t realize is that DCI is actually scaling as much or more than what we see in the data center market,” he said.
Taylor explained that this trend is influencing network architecture, with hyperscale operators increasingly seeking mesh topologies that provide multiple interconnections and built-in redundancy between facilities.
The presentation also highlighted changing traffic patterns driven by AI inference. Rather than primarily downloading content from cloud services, users and connected devices are increasingly uploading images, videos and sensing data for processing. “It’s no longer just traffic from a download standpoint,” Taylor said. “Now we’re starting to see a tremendous uplift in the upload.”
Taylor structured much of his presentation around what he called the “interconnect tech stack,” which consists of three elements: density, speed of deployment, and ecosystem integration.
The first pillar is deploying high-density fiber infrastructure capable of supporting AI-scale networks. The second is accelerating installation and splicing so infrastructure can be brought online more quickly. “You can have all the density in the world, but you have to be able to deploy that at scale quickly,” Taylor said.
The AFL executive argued that delays in deployment can leave servers sitting idle, resulting in what he described as “a massive loss in potential revenue.”
The third pillar is building an ecosystem in which cables, closures, splicing equipment, and logistics work together to support efficient deployment. Throughout the presentation, Taylor illustrated these concepts using examples from AFL’s own portfolio, including Spiderweb Ribbon fiber, high-density cable technologies, and intelligent ribbon splicing systems designed for large-scale deployments.
Looking ahead, Taylor also discussed emerging technologies such as hollow-core and multicore fiber. Hollow-core fiber, which transmits light through an air core rather than solid glass, could reduce latency and signal loss, while multicore fiber may increase capacity by incorporating multiple cores within a single fiber.
At the same time, he acknowledged that these technologies face several hurdles before widespread adoption, including standardization, interface compatibility, deployment practices, and cost.
Concluding his presentation, Taylor reiterated that successfully scaling AI infrastructure requires more than simply deploying additional fiber. Instead, he argued that operators and hyperscalers must balance three essential elements: high-density networks, rapid deployment and a supporting ecosystem of compatible technologies and tools.
To access the full webinar, click here.
