Large Generative AI Models meet Open Networks for 6G: Integration, Platform, and Monetization

TL;DR

This work addresses the challenge of integrating generative AI into open 6G networks from the perspective of mobile network operators. It proposes an API-centric telecoms GAI marketplace that enables in-network deployment, management, and monetization of diverse GAI services, paving the way for a (G)AI-native network. A proof-of-concept in an Open RAN testbed demonstrates latency benefits for edge-local LLM deployment relative to cloud endpoints, while also highlighting trade-offs in sustained throughput and longer-generation tasks. The paper details marketplace design principles, an implementation framework with API-based iPaaS, and a case study that together support in-network GAI adoption and new revenue opportunities for MNOs, with future directions toward collaborative AI and privacy-conscious multi-operator ecosystems.

Abstract

Generative artificial intelligence (GAI) has emerged as a pivotal technology for content generation, reasoning, and decision-making, making it a promising solution on the 6G stage characterized by openness, connected intelligence, and service democratization. This article explores strategies for integrating and monetizing GAI within future open 6G networks, mainly from the perspectives of mobile network operators (MNOs). We propose a novel API-centric telecoms GAI marketplace platform, designed to serve as a central hub for deploying, managing, and monetizing diverse GAI services directly within the network. This platform underpins a flexible and interoperable ecosystem, enhances service delivery, and facilitates seamless integration of GAI capabilities across various network segments, thereby enabling new revenue streams through customer-centric generative services. Results from experimental evaluation in an end-to-end Open RAN testbed, show the latency benefits of this platform for local large language model (LLM) deployment, by comparing token timing for various generated lengths with cloud-based general-purpose LLMs. Lastly, the article discusses key considerations for implementing the GAI marketplace within 6G networks, including monetization strategy, regulatory, management, and service platform aspects.

Figures (6)

• Figure 1: (a) Illustration of our conceptual telecom marketplace platform for GAI model integration and service delivery. It supports model registration, container-based deployment, multi-cloud integration, and open APIs for usage and billing; (b) Flowchart illustrating the monetization process in the telecom GAI marketplace)
• Figure 2: Overall marketplace integration architecture with API invocation. Multiple distributed or collaborative GAI models can be handled seamlessly via centralized API publishing, management, and usage-based billing.
• Figure 3: Illustration of the experimental setup. The local LLM container is deployed on the edge server near the Open RAN CU. Requests are managed via the marketplace's orchestrator.
• Figure 4: Histograms of Input and Output token lengths for the Chatbot arena dataset with the Llama 3.1 8B and 70B models, wherein (a) indicates Log-scaled Histogram of Input Tokens, (b) show Log-scaled Histogram of Output Tokens for Llama 3.1 Models.
• Figure 5: Illustration of the process of GAI content generation over marketplace platform and Open RAN testbed and time measurements, wherein TFT and ITT are the key metrics for comparison.