NVIDIA just joined BT Group, Deutsche Telekom, Ericsson, Nokia, SK Telecom, SoftBank, and T-Mobile to build 6G on open, AI-native platforms. Announced at Mobile World Congress 2026, the initiative embeds AI across radio access networks, edge computing, and core infrastructure. The NVIDIA AI Aerial platform — to be released as open source — provides the software stack for training, simulating, and deploying AI-native wireless networks. T-Mobile has demonstrated concurrent AI and RAN processing. SoftBank achieved industry-first 16-layer massive MIMO. The first user-to-user call over an AI-native 6G stack has already been completed.

AI-native means the network doesn't just carry data — it reasons about spectrum allocation, beam management, interference mitigation, and resource scheduling in real-time. Instead of static configurations tuned by engineers, AI models continuously optimize network parameters based on current conditions. This software-defined approach enables 6G networks to evolve through updates, delivering real-time intelligence at the infrastructure level.

But real-time optimization and long-term learning are different: AI-native networks optimize brilliantly for current conditions while having no memory of the patterns they've observed across days, weeks, and seasons.

Why Network Intelligence Needs Memory

Wireless networks exhibit strong temporal patterns. Rush hour creates predictable demand spikes. Stadium events concentrate traffic geographically. Seasonal weather affects signal propagation. Building construction changes interference patterns over months. An AI-native network that optimizes for current conditions handles each situation effectively in isolation. A memory-enabled network anticipates these patterns, pre-allocating resources before demand materializes.

The scale of 6G networks amplifies the memory opportunity. Millions of cells, billions of devices, and continuous optimization cycles generate enormous operational intelligence. Without memory, each optimization cycle starts fresh, re-deriving patterns from sensor data rather than building on accumulated understanding.

How MemU Adds Network Memory to 6G

MemU provides persistent network intelligence that spans optimization cycles. Traffic patterns, interference signatures, and resource allocation strategies are stored as structured memories. Before each optimization cycle, the AI retrieves relevant historical patterns, enabling predictive allocation that anticipates demand rather than merely reacting to it.

For multi-operator deployments, MemU enables shared network intelligence that improves spectrum efficiency across the ecosystem. The AI-native network that learns from experience optimizes more effectively than one that only reacts to the present.

NVIDIA built the AI-native network platform. MemU gives networks the memory to learn from every optimization cycle.

Get Started

Add persistent intelligence to your AI-native networks. Explore MemU at memu.pro and on GitHub.