As AI workloads push closer to end users, the limits of centralized cloud infrastructure are becoming harder to ignore. Latency, bandwidth costs, power constraints, and uneven data center availability—especially outside mature markets—are driving renewed interest in edge computing. Zettabyte and LiteOn are betting that the next phase of AI infrastructure won’t live in hyperscale facilities, but at the base of cell towers.
The two companies today announced a joint R&D collaboration to evaluate a distributed edge AI inferencing platform, known as the Ultra Edge Pod, designed for deployment at cell towers and tower-adjacent locations. The initiative focuses on validating whether GPU-based AI inference can operate reliably, efficiently, and at scale in telecom-proximate environments that were never designed for traditional data center workloads.
Turning Cell Towers Into AI Infrastructure
At the center of the collaboration is a specialized Mobile Edge Compute (MEC) platform that brings AI inference closer to users, radio access networks, and data sources. Instead of routing requests back to distant data centers, workloads are processed locally—dramatically reducing latency and improving responsiveness for real-time and location-aware applications.
The Ultra Edge Pod is intended for deployment not only at cell towers, but also at tower-adjacent facilities and other network-proximate infrastructure. This approach reflects a broader industry shift: treating telecom infrastructure as a distributed computing fabric rather than a simple connectivity layer.
For regions without mature data center ecosystems, the implications are significant. MEC-based edge AI platforms could provide a practical path to digital and AI readiness, enabling local enterprises and public-sector use cases without requiring massive centralized infrastructure investments.
Clear Lines Between Hardware and Software
A defining aspect of the Zettabyte–LiteOn collaboration is its strict separation of responsibilities—an architectural choice meant to improve scalability and operational clarity.
LiteOn will supply the physical foundation of the deployment, including:
- Power systems
- Cooling and thermal management
- Physical infrastructure optimized for edge environments
Zettabyte, meanwhile, delivers the software control plane, responsible for:
- GPU scheduling and orchestration
- Observability and monitoring
- Remote operations across distributed sites
Together, the two layers aim to make it possible to operate GPU-powered AI inference as a single logical platform, even though it is physically spread across hundreds—or potentially thousands—of cell towers. In effect, the system aggregates the compute capacity of multiple towers into a unified MEC fabric.
Designing for Real-World Edge Constraints
Unlike purpose-built data centers, telecom-adjacent sites impose strict limits on power, space, cooling, and operational access. These constraints have historically made them challenging environments for AI hardware, particularly GPUs.
The Ultra Edge Pod deployment is designed explicitly to test those limits. By tightly integrating infrastructure and software, the collaboration aims to demonstrate that low-cost, low-latency AI inference is feasible even under real-world conditions.
This is a critical validation step for the broader edge AI market. Many edge concepts work well in lab environments, but struggle when faced with inconsistent power availability, thermal constraints, and the operational realities of distributed telecom sites.
Enabling New Classes of AI Workloads
The Edge AI Deployment is intended to support low-latency, location-aware AI inference workloads operating close to mobile users and RAN infrastructure. These workloads could span multiple sectors, including:
- Real-time video analytics
- Network optimization and monitoring
- Smart city and transportation applications
- Industrial and enterprise edge use cases
By running inference closer to the point of data generation, the platform reduces backhaul traffic and improves performance—an increasingly important consideration as AI models grow more complex and data-intensive.
Just as importantly, the project validates whether GPU-based AI computing can be economically and operationally viable in scenarios managed by tower companies and telecom operators, rather than hyperscale cloud providers.
A Modular Model for Edge AI Scale
According to Zettabyte, the collaboration is as much about operational design as it is about technology.
“This deployment allows both teams to validate a practical and scalable model for edge AI deployment, emphasizing repeatability, resilience, and operational efficiency through a clear separation of infrastructure and software responsibilities,” said Kenneth Tai, Chairman of Zettabyte.
That emphasis on repeatability matters. For edge AI to scale beyond pilots, deployments must be standardized enough to roll out across geographies while remaining resilient to local variations in power, climate, and network conditions.
LiteOn brings decades of experience in the physical side of that equation.
“LiteOn’s experience in power systems, thermal management, and physical infrastructure positions the company to support emerging edge AI use cases through disciplined, deployment-driven collaboration,” said Jason Tsao, Associate Vice President and Head of Direct Current Microgrid at LiteOn.
Context: Why Edge AI Is Having a Moment
The collaboration arrives amid renewed momentum for edge AI across telecom and enterprise markets. As AI moves from batch processing to real-time inference, latency becomes a competitive factor rather than a technical detail.
At the same time, geopolitical and economic realities are forcing organizations to rethink centralized infrastructure strategies. Distributed edge platforms offer resilience, data sovereignty, and cost advantages—particularly in markets where building hyperscale data centers is impractical.
Telecom operators and tower companies are increasingly exploring how their physical assets can support this shift. By hosting AI workloads at or near cell towers, they can unlock new revenue streams while enhancing network intelligence and service quality.
What Success Would Look Like
If the Ultra Edge Pod proves viable, it could serve as a reference architecture for telecom-adjacent AI infrastructure, combining:
- GPU-based inference at the edge
- Software-defined orchestration across distributed sites
- Infrastructure designed specifically for constrained environments
That combination could accelerate adoption of edge AI in both developed and emerging markets, providing a blueprint for how AI compute can be embedded directly into network infrastructure.
For Zettabyte and LiteOn, the collaboration is less about a single deployment and more about validating a model—one that could be replicated wherever connectivity exists, turning cell towers into active participants in the AI economy.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
