Ericsson introduced a new portfolio of AI-ready radios, antennas, and AI RAN software ahead of Mobile World Congress 2026 in Barcelona, targeting rising uplink and latency demands from AI-powered and augmented reality devices. The company said its latest RAN hardware incorporates neural network accelerators within Ericsson Silicon to enable real-time, on-site AI inference in Massive MIMO radios. The launch reflects a broader industry shift toward distributed AI architectures embedded directly into radio access networks.
The hardware lineup includes ten AI-ready radios spanning Massive MIMO and remote configurations. New products include high-power FDD Massive MIMO units such as AIR 3286 and AIR 3211, expanded 8-receiver radios including Radio 4891 and Radio 4458, and triple-band radios Radio 4488 and Radio 4464 aimed at network consolidation and RAN sharing. Ericsson also highlighted TDD Massive MIMO updates, including AIR 3267 with 600 MHz instantaneous bandwidth in a 13 kg (28.7 lb) unit, and AIR 6492 delivering 480W output power with 256 antenna elements. The radios leverage programmable matrix cores integrated into Ericsson’s Many-Core Architecture to support AI and machine learning workloads at the edge.
On the software side, Ericsson added AI-managed Beamforming, AI-powered Outdoor Positioning, and an AI model for instant coverage prediction to its RAN stack. The company also introduced a Latency Prioritized Scheduler and Low Latency Mobility features, which it says deliver up to seven times faster response times to support bounded latency and reliable uplink performance for AI and AR applications. The portfolio expands with five new high-performance antennas, including passive trio net designs optimized for uplink performance, carrier aggregation, and spectral efficiency, along with new interleaved AIR configurations for flexible TDD and FDD deployments.
- Ten AI-ready radios with integrated neural network accelerators for distributed AI inference
- AIR 3267: 600 MHz bandwidth, 13 kg (28.7 lb) form factor
- AIR 6492: 480W output power, 256 antenna elements
- Triple-band radios (Radio 4488, Radio 4464) for consolidation and RAN sharing
- AI-managed Beamforming, AI-powered Outdoor Positioning, instant coverage prediction
- Latency Prioritized Scheduler and Low Latency Mobility for up to 7x faster response times
- Five new antennas including energy-efficient passive trio net designs
Mårten Lerner, Head of Networks Strategy and Product Management at Ericsson, said: “As AI transforms traffic patterns and raises consumer expectations, networks must provide precise performance where and when it’s needed most. At Ericsson, we’re committed to delivering that needed performance. We are also embedding AI solutions across our full portfolio, introducing AI RAN software solutions that deliver revolutionary improvements in spectral efficiency. We are now taking the final step to full AI enablement across our portfolio by introducing neural network accelerators in our leading Massive MIMO portfolio.”
🌐 Analysis: Ericsson’s move aligns with operator efforts to shift AI workloads closer to the radio edge, reducing latency and improving uplink determinism for emerging multimodal AI applications. The announcement comes as competitors including Nokia and Huawei expand AI-native RAN capabilities, signaling that embedded accelerators and distributed AI inference will become a standard feature in next-generation 5G Advanced and future 6G platforms.
AI RAN is emerging as a structural shift in mobile network architecture rather than a feature upgrade. Operators are moving from rule-based optimization and static parameter tuning toward closed-loop, data-driven control systems that continuously adapt spectrum, power, and scheduling decisions in real time. By embedding neural network accelerators directly into radio units, Ericsson is pushing inference closer to the air interface, reducing backhaul dependency and enabling distributed intelligence across thousands of cell sites. This architecture supports bounded latency, uplink-heavy traffic patterns, and fine-grained beam management required by multimodal AI services.
The broader AI RAN movement also reflects a competitive realignment among infrastructure vendors and hyperscalers. Open RAN frameworks, cloud-native DU/CU deployments, and edge compute integration create new opportunities for AI models to run across Layer 1 through Layer 3 functions. Vendors are now differentiating on silicon integration, accelerator density, and AI model efficiency rather than solely on spectral efficiency metrics. As 5G Advanced evolves toward early 6G research programs, embedded AI inference in radios and baseband units is likely to become a baseline capability, particularly for operators seeking energy optimization and autonomous network operations at scale
