Distributed Infra concept
– Getty Images

Networking vendor Dispersive disclosed details of a successful pilot with American Tower to develop a secure means to transport data from edge-based GPU-as-a-service (GPUaaS) environments.

The vendor claims that neocloud and emerging GPUaaS providers rely on public internet transport, which lacks the inherent security needed to properly keep customer workloads safe.

Dispersive’s trials saw its Trusted Cloud Edge (TCE) multipath transport fabric paired with Vertical Data’s high-density GPU compute platform via American Tower’s data center in Raleigh, North Carolina, to send encrypted data across distributed sites.

Results of the “multimonth” pilot saw a 30% throughput improvement compared to conventional secure networking approaches, along with sub-second autonomous failover and apparent enhanced protections against potential data interceptions.

“Enterprises and neocloud providers need low-latency environments that can securely support high-density GPU workloads at the edge,” Jim Poole, VP of product development for edge data centers at American Tower, explained. “By combining our edge data centers with advanced secure transport architectures, we are enabling a new class of AI deployments that extend into regional markets.”

Amid the rising demand for infrastructure capable of powering the ever-growing AI needs, the often-overlooked edge is quickly becoming an area of interest for providing connectivity.

IDC projects global spending on edge computing solutions to increase from nearly $261 billion in 2025 to $380 billion by 2028, while Global Market Insights puts the edge data center market at around $71.9 billion in 2035.

That rise in demand comes from the need for infrastructure to be close by to where the AI workloads are actually being used – especially with powerful models becoming smaller, while the inherent rise of agents running myriad applications meaning latency needs to be low and throughput high.

The Dispersive tests come as American Tower is ramping up its edge play, moving away from its now-discontinued small line of colocation sites under the Access Edge brand name to multimegawatt data centers.

Having recently launched the 1 megawatt facility in Raleigh last year, the firm is looking to add two more sites made using prefabricated designs in Oklahoma City and Tampa Bay.

American Tower’s efforts with Dispersive also look to reduce operational costs. The networking vendor claimed that, based on its tests, edge-based GPUaaS environments can reduce total cost of ownership by up to 40% as it eliminates the need for separate virtual private network (VPN) overlays and dedicated security appliances.

“As AI workloads move to the edge, neoclouds and emerging GPUaaS providers face a critical gap: they don’t have hyperscaler-scale backbones to securely move that traffic,” Dispersive CEO Rajiv Pimplaskar claimed. “Dispersive enables them to embed secure, high-performance transport directly into their infrastructure and unlock scalable AI delivery without the cost and complexity of building private fiber networks.”