Advancing AI Opportunities at the Telecom Edge
The growth in AI adoption around the world is generating widespread deployment of foundational AI models and training infrastructures. Suddenly, demand is increasing for localized infrastructure to support AI inference, often bundled with a need for security, high bandwidth, and low latency connectivity.
To meet this need and generate new revenues, telecom operators are actively exploring the deployment of AI at the edge with the use of graphics processing unit (GPU)-powered infrastructure and specialized hardware accelerators. Edge AI enables operators to boost operational efficiency, improve customer experience, and offer differentiated services, especially as data traffic and demand for real-time applications grow.
With their distributed data centers, operators are perfectly positioned to create inference engines and sovereign AI factories close to end users, reducing latency and improving performance for AI applications. Dense footprints of central offices, metro points-of-presence, and thousands of radio sites could double as low-latency mini-data centers for inference workloads.
McKinsey estimates hosting/colocation and revenue-share models around AI infrastructure could unlock a USD $30–50 billion opportunity this decade — but only if operators move fast.
Telco AI opportunities at the edge
Operators have a variety of AI edge opportunities that are mapped in the graphic below to a range of addressable markets and their requirements.
Operators are well positioned to be “trusted sovereign AI hosts” with governments and industries that require AI localization and regulatory adherence. Nationally available infrastructures, regulatory compliances, trusted relationships with enterprises, government, and citizens, and complementary capabilities for security, connectivity, and resilience are important differentiators.
Government agencies want data sovereignty and control on a secure and resilient national AI infrastructure, as well as regulatory compliance.
Enterprises want to process data closer to its source to achieve the low latency needed for applications like AI computer vision, smart city analytics, and threat detection, while ensuring deterministic latency, privacy, and regulatory compliance.
Small and medium-sized businesses want out-of-the-box plug-and-play AI solutions that are easy to integrate into their business processes and may include security and connectivity.
Edge hosting AI opportunities
Beyond using the edge for AI inferencing, applications, optimized connectivity, and enhanced security, operators are pursuing two hosting opportunities: GPU as a Service (GPUaaS) and Inferencing as a Service (IaaS) — in this context, referring to AI model inferencing rather than the more common Infrastructure as a Service.
GPU as a Service
Leading operators are piloting and launching GPUaaS commercial models that enable enterprises to access GPU and accelerator resources on demand. GPUaaS allows customers to rent GPUs rather than buy, set up, and maintain their own physical hardware. Operators host the hardware, but their customers have control over which frameworks, models, and workloads to deploy.
GPUaaS targets customers who need raw GPU computing resources for training large AI models, running complex simulations, and analyzing large datasets efficiently. These would typically be large enterprises and governments that would install their own models on operator GPU resources to meet their localized sovereignty and low latency connectivity requirements.
Inferencing as a Service
IaaS allows operators to offer plug-and-play access to pre-deployed AI models, including bespoke industry-specific models, for inference at scale. Operators host training models on their infrastructure and provide model and infrastructure management. They may partner with third parties for applications, such as fraud and AI vision. Customers send their data to the pre-deployed models.
IaaS targets small and medium enterprise customers who want a fully managed platform that allows quick integration of AI capabilities into their business.
Leading telecom operators are pursuing edge AI opportunities
A leading North American operator packages a private 5G slice plus GPUs in its multi-access edge computing (MEC) nodes so factories or stadiums can spin up computer-vision or Generative AI (GenAI) agents on-premises with sub-microsecond latency.
Several leading Asian operators have officially launched GPUaaS for enterprise and government customers showcasing edge inferencing capabilities.
Several leading European operators are launching GPUaaS and IaaS in Europe, focusing on delivering nationally sovereign AI services from national data centers.
The big three operators in China have all launched a combination of centralized AI data centers with hundreds of localized AI Edge nodes, with the national cell tower provider in the process of upgrading two million base station shelters to become AI Edge data centers hosting accelerator processing units (xPUs).
AI opportunities at the edge are available for operators to capitalize on today. Technical tradeoffs should be assessed regarding which edge is best suited for situations that span user equipment far edge, network edge, aggregation points, and metro or centralized data centers. Factors such as energy efficiency, physical space, network performance, and comprehensive security need to be considered.
In our next blog, we will address these tradeoffs and impacts on AI and edge strategies, including considerations for testing and assurance.
Explore Keysight's edge AI testing and assurance solutions, and read our eBook Testing Networks Enhanced with AI.