The AI-RAN Alliance is a global industry consortium, launched at MWC Barcelona in February 2024 and governed by a Technical Steering Committee (TSC), whose mission is to accelerate the integration of artificial intelligence into Radio Access Networks and to define what an AI-native RAN looks like in practice for 5G Advanced and 6G. The alliance deliberately positions itself as neither a marketing organisation nor a demo factory: it pursues pioneering, pre-competitive work — reference architectures, blueprints, and credible benchmarking — without getting mired in formal standards processes or IP negotiations. Its work spans three complementary objectives — AI-for-RAN (using AI/ML to improve RAN performance and efficiency), AI-and-RAN (co-locating RAN and AI workloads on shared accelerated infrastructure), and AI-on-RAN (hosting tenant-facing AI applications at the network edge for differentiated, monetisable connectivity). Founding members include Ericsson, Nokia, NVIDIA, T-Mobile, SoftBank, Samsung, AWS, Microsoft, and Arm; membership grew from a handful at launch to 130+ organisations by MWC 2026, spanning operators, NEPs, hyperscalers, silicon vendors, universities, and government research bodies across more than 17 countries.
Technical work is organised under the Technical Steering Committee (TSC) in three distinct layers — Working Groups (one per strategic pillar), Task Groups (cross-cutting specialist teams), and alliance-wide programmes — with a deliberate 2026 shift from scattered demos toward a deliverables engine with clearer ownership, timelines, and mandatory benchmarking plans.
Working Groups (three, aligned to the alliance pillars):
| Working Group | Scope |
|---|---|
| AI-for-RAN | AI/ML to improve RAN performance: air interface, channel estimation, radio resource management, energy efficiency, ISAC |
| AI-and-RAN | Shared infrastructure and orchestration: coexistence of RAN and AI workloads, MLOps, compute placement, interaction with O-RAN systems |
| AI-on-RAN | Edge AI applications on RAN infrastructure: differentiated connectivity, monetisation models, SLA-grade assurances |
Within the AI-for-RAN working group, the TSC clusters 20+ work items by RAN process timescale into three functional buckets: Air Interface & Signal Processing (sub-ms PHY); Resource & Mobility Control (10 ms → 1 s); and Network Operations & Automation (seconds to hours).
Task Groups (cross-cutting, span all pillars):
| Task Group | Scope |
|---|---|
| Data-for-AI | Data frameworks and methodologies for AI model training and inference |
| Test Methodology | Testing and validation frameworks through AI-RAN Alliance-Endorsed Labs |
| Agentic AI (ATG) | Agentic AI for autonomous network operations; open-source reference implementations |
| Commercialization | Business models, TAM/SAM analysis, and monetisation pathways for operators |
Programmes: RANPerf is an alliance-wide benchmarking initiative (not a working group) — every result is reported as an (algorithm, platform) pair against 3GPP-anchored test conditions, with mandatory benchmarking plans for WG demos and work items.
Architecturally, AI-RAN publishes reference architectures and industry blueprints, not normative interface specifications. The AI-and-RAN working group designs the AI-RAN platform — including monitoring, MLOps, distributed training, emulation frameworks, and an AI-RAN Workload Placement Function that dispatches RAN, AI-for-RAN, and AI-on-RAN workloads to optimal compute targets across edge, cloud, or data centre based on latency, cost, policy, and data sovereignty. A key architectural shift is decoupling AI-for-RAN from AI-on-RAN: RAN-internal AI optimisation and tenant-facing edge AI no longer share the same compute fate, enabling flexible orchestration by platform orchestrators and LLM routers. Relative positioning with O-RAN: the two alliances are complementary, not competing. O-RAN defines the open, interoperable RAN contract — SMO, Non-RT/Near-RT RIC, O-Cloud, O1/O2/A1/E2 — and publishes normative specifications for multi-vendor assurance. AI-RAN builds on that foundation by addressing what O-RAN does not fully specify: multi-tenant coexistence of generic AI workloads and RAN on shared GPU infrastructure, reproducible AI/ML benchmarking for RAN algorithms, and monetisation models for AI-at-the-edge. Overlap is explicit and intentional — the AI-and-RAN working group carries an active work item on interaction with O-RAN systems for orchestration and coexistence workflows — and spans RIC/xApp AI/ML, SMO-adjacent orchestration, O-Cloud resource management, and automated control loops. The division of labour remains: O-RAN specifies the open RAN contract; AI-RAN produces implementation blueprints, trial evidence, and benchmarked performance data on top of it. Two 2026 focus areas further illustrate the boundary: ISAC (Integrated Sensing and Communications) closes the implementation gap between 3GPP Rel-19 sensing KPIs and deployable AI-native sensing on shared RAN infrastructure, while the AI-RAN Security Framework addresses converged risks across the telecom control plane (RIC/E2/scheduler), AI/ML lifecycle (models, datasets, adversarial inputs), and accelerated cloud infra (Kubernetes, GPU isolation, MIG/MPS) — six vulnerability categories with proposed work items on threat modelling, secure RIC app lifecycle, GPU isolation validation, and telemetry integrity.
Market momentum has been rapid, but the alliance’s 2026 TSC direction emphasises execution discipline over headline demo counts. The RANPerf leaderboard — inspired by MLPerf and structured as mandatory (algorithm, platform) submissions against 3GPP-anchored channels (TR 38.901, TS 38.104), fixed datasets, and classical SoTA baselines — aims to solve the “apples-to-oranges” problem that has plagued AI/ML-in-RAN claims: reporting throughput, P99 slot latency, energy per bit, and compute consumption together so submissions cannot win on one KPI while losing on another. The AI-for-RAN working group mandates a benchmarking plan for every work item and demo; Alliance-endorsed labs provide reproducible test environments. At MWC 2026 the alliance presented 33 innovation demos and four industry blueprints, but the TSC acknowledged a persistent gap between theoretical outputs and real-world field validation — with live network trials (e.g. collaboration with South Korea’s AINA/MSIT) discussed as a flagship next step. The Call for Innovation has been narrowed to six focus areas — ISAC, Security, Agentic AI/Operations, Test & Benchmarking Frameworks, Network Energy Saving, and Physical AI Applications — signalling where the alliance will concentrate pre-competitive investment. Adoption remains earlier-stage than O-RAN’s specification and SCAS ecosystem: AI-RAN outputs are reference designs, blueprints, and leaderboard evidence, not conformance-testable standards. The alliance is best understood as the industry’s fast-moving innovation and benchmarking layer for AI-native RAN — bridging O-RAN’s open architecture, 3GPP’s radio specifications, and the practical deployment of AI across shared RAN infrastructure in the 6G era.