Duos Edge AI and Zero Latency Strike 15-Site Edge Deal as Inference Moves Off the Hyperscaler Grid
Duos Edge AI, the infrastructure subsidiary of Duos Technologies Group, has signed a non-binding term sheet with Zero Latency covering all fifteen of its edge data center sites across Texas and Georgia, a portfolio totaling 225 cabinets that the companies intend to lease as the backbone for Zerogrid, Zero Latency's distributed inference network. The agreement, announced August 13, marks a concrete step in the shift of AI workloads from centralized hyperscaler campuses toward smaller, geographically dispersed facilities that sit within miles of end users.
The deal is structured as a 90-day exclusivity period during which the parties will complete due diligence and negotiate a definitive true-lease arrangement. Doug Recker, chief executive of Duos, said the term sheet reflects a strategy of converting already-deployed capital into contracted recurring revenue. Michael Huerta, chief executive of Zero Latency, framed the partnership as a response to a fundamental constraint: inference is moving locally, where context is created and managed, and hyperscalers are not built to serve those use cases.
From Centralized Training to Distributed Inference
The Duos-Zero Latency agreement illustrates a trend that Dell'Oro Group quantified in a July 2026 forecast: the rise of agentic AI and inference workloads is driving a near-40-percent compound annual growth rate in front-end data center network opportunities, with more than half of front-end switch sales growth over the next few years tied to AI-related deployments. Sameh Boujelbene, vice president at Dell'Oro, said the traditional 10-to-1 ratio of accelerator to CPU no longer holds across all deployments, with some environments moving closer to 1-to-1 as CPUs become critical for orchestration and data movement in inference clusters.
That architectural shift favors the kind of distributed, cabinet-scale infrastructure Duos operates. Each of the fifteen sites in the portfolio is designed to deliver 100 kilowatts per cabinet or more, with a 90-day deployment timeline and a target radius of twelve miles from end users. The Texas and Georgia footprint covers corridors where fiber density and power availability support latency-sensitive workloads — telecom radio-access-network processing, physical AI for robotics and autonomous systems, and enterprise inference that cannot tolerate the round-trip to a regional cloud zone.
Telcos and the Edge Mesh Opportunity
The edge-inference wave is not limited to specialist operators. Amdocs, the telecommunications software giant, told analysts this week it is evaluating edge mesh-grid concepts with Nvidia as part of a three-pillar strategy that includes agentic transformation of business and operations support systems and expansion into a second regulated vertical. Anthony Goonetilleke, group president of technology and head of strategy, said early deployments show a European customer's agentic WhatsApp system handling 97 percent of interactions without human escalation, while another network project cut annual operating costs by 60 percent.
For telecom providers, the economics are compelling. Radio access networks generate massive data streams that are expensive to backhaul to centralized clouds. Processing that data at the edge — for real-time traffic optimization, predictive maintenance, or localized AI services — reduces transport costs and unlocks new revenue models. Zero Latency's Zerogrid platform, which the Duos portfolio would underpin, is built explicitly for this distributed dispatch model: workloads route to the nearest node with available capacity, whether that node is owned or sourced from a partner.
Capital Efficiency and the True-Lease Model
The proposed true-lease structure matters. By accounting for the arrangement as a lease rather than a purchase, Zero Latency preserves balance-sheet flexibility while Duos converts sunk infrastructure investment into a predictable revenue stream. The 90-day exclusivity window includes site-level verification of Duos's invested capital across the portfolio, a step that mirrors the due diligence hyperscalers apply to their own buildouts but at a fraction of the scale and timeline.
Duos Technologies Group, which trades on Nasdaq as DUOT, has positioned its edge subsidiary as a manufacturer-agnostic platform — sourcing and fulfilling infrastructure from multiple vendors rather than locking into a single stack. That flexibility aligns with the Dell'Oro finding that Accton, Arista, Celestica, Cisco, HPE/Juniper, H3C, Huawei, and Nvidia are all emerging as beneficiaries of the front-end network expansion. A vendor-neutral edge layer can absorb switching and compute gear from any of those suppliers as workload demands shift.
The Hyperscaler Gap
Huerta's comment that hyperscalers "aren't able to serve" certain inference use cases points to a structural mismatch. AWS, Azure, and Google Cloud optimize for massive, contiguous campuses where training clusters of tens of thousands of accelerators share high-bandwidth interconnects. Inference, especially agentic inference that spawns short-lived, bursty workloads, benefits from proximity to the data source and the user. A 225-cabinet portfolio distributed across two states cannot replace a hyperscaler region, but it can handle the latency-critical fraction of the workload that the region cannot economically serve.
That fraction is growing. Dell'Oro estimates that AI-related front-end network buildouts will drive adoption of 800-gigabit and 1.6-terabit switching in front-end networks, a bandwidth tier that until recently lived almost exclusively in training back-ends. The networking requirements for key-value caching storage racks in inference clusters are creating new rack-level architectures that look more like miniature training pods than traditional web-serving tiers.
What Comes Next
The term sheet is non-binding outside its exclusivity, confidentiality, and expense provisions. Both parties have collaborated on edge data center activities before, which reduces integration risk but does not guarantee a definitive agreement. If the lease closes, it will represent one of the first structured, multi-site edge-inference capacity deals announced in the United States — a template that other infrastructure owners and inference networks may replicate as the economics of distributed AI continue to prove out.
For now, the fifteen sites in Texas and Georgia sit ready, their cabinets powered and connected, waiting for the workloads that the next wave of agentic applications will generate. The grid is not waiting for the hyperscalers to extend their reach; it is being built, cabinet by cabinet, at the edge.
Our cloud and edge computing desk tracks the infrastructure buildout powering distributed AI, and the semiconductors team follows the silicon that fills the racks.
Sources: GlobeNewswire — Duos Edge AI Signs Exclusive Term Sheet with 0Lat LLC, August 13, 2026; Dell'Oro Group — Agentic AI and Inference to Supercharge Front-End Networks Growth, August 13, 2026; Yahoo Finance — Amdocs Targets Telco AI Growth With Agentic Platforms and Edge Computing, August 13, 2026.
Keywords: Duos Edge AI, Zero Latency, edge data center, distributed inference, Zerogrid, AI infrastructure, Texas, Georgia, Dell'Oro Group, Amdocs, Nvidia, edge computing