Nokia's 6G bet: AI-native networks that learn, predict, and optimize
Pallavi Mahajan, Nokia's Chief Technology and AI Officer, is clear on where this is going: 6G isn't just faster radios. It's the network gaining the ability to judge and learn on its own.
Nokia is building that future with AI-RAN-an AI-based radio access network that runs, tunes, and heals itself in real time. Nvidia backed the vision with a $1 billion investment in October, acquiring roughly 166.39 million new shares at $6.01 each for a 2.9% stake, making it Nokia's second-largest shareholder. The plan: co-develop AI-RAN and the AI compute stack needed to run it at scale.
Why AI-RAN matters for engineers
- Dynamic resource allocation: real-time AI models allocate spectrum, power, and processing per cell and per user session.
- Predictive ops: failures and performance dips get flagged and resolved before users feel them.
- Per-user QoE: traffic patterns drive per-subscriber policies without static rules.
- Efficiency: internal Nokia cases report up to 30% lower energy use plus lower latency and faster response times.
What Nokia is building into 6G
- THz spectrum exploration for extreme capacity and sensing-aware use cases.
- Network slicing with tighter per-slice guarantees and AI-driven SLA control loops.
- Edge-cloud integration to place RAN and inference where it makes the most sense.
- Post-quantum security with quantum-resistant encryption primitives.
- Energy-first design across radio, transport, core, and orchestration.
Roadmap stages: research and standardization → prototypes and trials → commercialization around 2030.
Where Nokia claims an edge
Nokia leans on 5G rollout experience and an end-to-end stack: baseband, radio, transport, core, and management systems as one architecture. It's active in standards and R&D with partners across industry, academia, and governments. Sustainability targets are built into product strategy, which matters as operators chase energy reductions per bit.
Early AI-RAN results and what's different from 5G
The AI-RAN platform brings a GPU-based, software-defined RAN so operators can evolve site-by-site. Unlike 5G, where AI sat beside operations, 6G aims for a cognitive network: models continuously learn from live traffic, services, and device states to approach autonomous operation.
Reported outcomes so far: up to 30% energy savings, lower latency, and faster response-direct levers for OPEX and ESG targets.
The Nvidia factor
Core RAN functions-beamforming, channel estimation, scheduling-benefit from heavy AI compute. Nvidia GPUs run these models in real time to lift spectrum efficiency and QoS. Paired with Nokia's anyRAN software approach, the stack can run across varied hardware and cloud footprints without giving up performance.
- Joint workstreams: AI and edge research, GPU integration into RAN/core, and pilots beyond telecom.
- AI-RAN Centre with T-Mobile US and Nvidia to validate in production-like environments and shape commercialization models.
Hard problems still on the table
- Unified edge-cloud fabric that meets strict latency and jitter budgets.
- Reliable, transparent AI models with lifecycle controls and auditability.
- Security posture that includes quantum-resistant encryption.
- Spectrum access, standards maturity, and the cost profile of high-performance infrastructure.
- Smooth coexistence and interop with 4G/5G during long transition periods.
What builders and operators can do now
- Instrument everything: high-fidelity RAN counters, radio/edge telemetry, and clean data pathways for model training.
- Plan the GPU footprint at the edge: size for inference latency, power budgets, and cooling constraints.
- Stand up MLOps for networks: feature stores for RAN, drift monitoring, rollout/rollback for models, and shadow modes.
- Adopt post-quantum crypto pilots for control/user plane where feasible; track standardization progress from NIST.
- Build digital twins for RF and transport to test AI policies safely before live rollout.
- Enforce zero-trust and SBOM policies across the RIC, xApp/rApp ecosystem, and third-party integrations.
- Run interop testbeds mixing vendor gear, private/public cloud, and different accelerators to avoid lock-in.
- Upskill teams on AI for networks, GPU programming, and edge orchestration.
Industry impact
As AI-based networks take on repetitive ops, teams can shift to policy and strategy. Reliability climbs through predictive maintenance; cost and energy drop through smarter allocation. Manufacturing, logistics, automotive, healthcare, and smart city deployments stand to benefit first from real-time optimization and tighter SLAs.
Standards and ecosystem
Nokia is pushing open collaboration through public-private programs, international standards, and industry-academia consortia. It participates as a Platinum member of the Open Compute Project, part of a broader effort to build an open, intelligent network stack.
Resources
- Open Compute Project (OCP)
- NIST Post-Quantum Cryptography
- AI upskilling paths by job role (Complete AI Training)
Your membership also unlocks:
