Dimitra Simeonidou leads UK project to develop AI-native network architectures

Embedding AI into the network fabric represents a shift from treating intelligence as an external application to making it a core component of the hardware stack. For streaming providers, this transition promises more efficient edge processing and improved quality of service through real-time resource optimization. This development aligns with the broader industry move toward AI-RAN, where traditional telecom assets are repurposed as distributed compute grids capable of handling high-bandwidth video traffic. As these open platforms democratize experimentation, the ecosystem will likely see faster commercialization of software-defined networking tools. Watch for the second part of Simeonidou’s findings regarding the specific deployment hurdles facing network architects in late 2026.

Ericsson has moved aggressively to commercialize AI-native network capabilities, directly paralleling the academic work led by Dimitra Simeonidou at the University of Bristol. In March 2026, Ericsson launched a suite of AI-ready radios featuring its custom Ericsson Silicon with integrated neural network accelerators, designed to boost on-site AI inference in Massive MIMO radios for real-time optimization and fully distributed AI. The company simultaneously introduced AI RAN software enhancements including AI-managed beamforming, AI-powered outdoor positioning, and a coverage prediction model that complements its existing AI-native Link Adaptation feature. The commercial validation of these approaches is accelerating through operator partnerships. In March 2026, AT&T and Ericsson demonstrated AI-native Link Adaptation on a Cloud RAN stack powered by Intel Xeon 6 SoC, achieving up to 20 percent throughput gains compared to legacy rule-based link adaptation. The test used AT&T's frequency bands and propagation characteristics, marking the first call with portable Ericsson software on commercial off-the-shelf hardware and establishing a collaborative benchmark for AI in RAN deployments. Ericsson's head of networks, Per Narvinger, noted that AI RAN embedded in link adaptation together with RAN neural accelerators is already boosting spectrum efficiency at customers by around 10 percent, with Bell Canada running field tests in April 2025 that were described as the first anywhere. The architectural debate around AI-native networking also centers on whether GPU acceleration is necessary. Ericsson has insisted that AI-RAN is achievable without Nvidia GPUs, arguing that its purpose-built baseband silicon and custom radio chips can handle the required AI inference workloads. This positions Ericsson's approach differently from Nvidia-backed AI-RAN initiatives, and aligns with Simeonidou's research philosophy of embedding intelligence directly into programmable hardware fabric rather than relying on external compute layers. The vendor confirmed that its AI-RAN offerings target customers of its most recent purpose-built 5G products, with the model running on existing baseband units without requiring hardware upgrades. As these evolve, vendors are increasingly prioritizing to drive software-driven efficiency. Recent to support these deployments. For operators seeking to scale these deployments, provide a proven model for managing high-capacity infrastructure.