Ericsson Exec: Mobile Networks, Not GPUs, Will Decide AI Winners
An Ericsson executive argued that mobile networks will be the ultimate determinant of who wins in AI, not just AI models or GPUs. The head of thought leadership stated that superior connectivity for edge computing and monetization will be the key differentiator, shifting the focus from pure processing power to data delivery infrastructure.
The argument from Ericsson's Head of Thought Leadership, Peter Linder, suggests the AI battleground is shifting from data centers to the network's edge. This move means AI inference—using trained models to make real-time decisions—will happen closer to users and devices, making the performance of the underlying mobile network critical. The focus is moving from raw processing power for training models to the infrastructure that delivers AI-powered applications. This shift is driven by the rise of real-time, data-intensive AI applications that cannot tolerate the delay of sending data to a centralized cloud. Use cases like autonomous vehicles, augmented reality, and smart factory robotics require the sub-10-millisecond latency that 5G Standalone (SA) networks provide. Ericsson's CEO, Börje Ekholm, has emphasized that without high-performing, programmable connectivity, AI and cloud technologies cannot scale effectively. The debate pits the network against the current AI hardware king, the GPU. NVIDIA currently dominates the AI hardware market, holding over 80% of the market for GPUs used in AI training. However, Ericsson is betting on its own custom silicon, embedding AI-accelerating capabilities directly into its network hardware, like AI-ready radios with neural network accelerators. This strategy aims to avoid a "GPU tax" for carriers by using optimized, purpose-built chips instead of expensive, power-hungry GPUs at cell sites. Ericsson is actively preparing for this shift by focusing on "AI for networks and networks for AI." The company is developing AI-native networks, even ahead of 6G, and has achieved Level 4 network autonomy, allowing networks to make operational decisions with minimal human intervention. This includes using AI for network optimization, such as predicting coverage to ensure smoother connections for users on the move. The evolution to AI-native applications will transform network traffic. Historically designed for downlink-heavy smartphone use, future networks must handle a massive increase in uplink traffic from a growing number of AI-centric devices like smart glasses and robotics. This makes low latency and uplink performance key commercial differentiators, positioning advanced 5G and future 6G networks as the essential infrastructure for the next wave of AI.
