BT Highlights the Crucial Role of Networks in AI at Mobile World Congress

The 2025 Mobile World Congress was not just about flashy devices and AI demos. BT ensured the spotlight was also on something less conspicuous but equally vital: networks. As the industry advances with generative models, autonomous systems, and AI-infused services, BT posed a critical question: How do we ensure these systems run reliably at scale?

Their answer is straightforward yet complex: the network should be treated as an integral part, not just a utility. At their booth, in panels, and through private briefings, BT emphasized that AI implementation without strong, smart networks is like racing a Formula One car with flat tires.

The Network: No Longer in the Background

For decades, networks remained in the backdrop, merely channels for data transmission. But today’s AI systems are real-time, data-intensive, and distributed. Generative AI needs to access data from multiple sources, process it in real-time, and deliver results within milliseconds. Reliable, high-throughput networks are essential for this.

BT showcased this shift through live demos and network simulations, illustrating how even minor jitter or latency dropouts can cripple AI decision-making in logistics or smart manufacturing scenarios.

BT’s CTO compared the network to the nervous system in the human body—AI might be the brain, but without fast reflexes and clean signals, the system underperforms. For instance, network delays impacting an autonomous vehicle fleet’s route planning aren’t just technical hiccups—they’re real-world disruptions. AI systems depend on time-sensitive inputs, and a five-second delay can lead to wrong outcomes. BT argues that recalibrating the entire stack to treat the network as dynamic and intelligent is crucial to mitigating these risks.

Developing AI-Aware Networks

A major takeaway from BT’s presence at the Mobile World Congress was its advocacy for AI-aware networks. This goes beyond using AI to optimize network traffic; it involves building infrastructure that adapts in real-time to AI workload demands. Such networks can reroute traffic before congestion occurs, isolate unstable nodes autonomously, and adjust bandwidth on the fly depending on whether a model is being trained, fine-tuned, or just queried.

BT demonstrates its adaptive network systems capable of real-time adjustments.

BT showcased prototypes of these adaptive systems through partnerships with vendors using edge and cloud co-processing. In one instance, an edge site detected a local spike in demand for a vision model tied to city surveillance feeds. Instead of sending all that raw data to the cloud, the system compressed, filtered, and partially processed it locally, then passed only the meaningful data upstream. This flexibility cuts latency and improves reliability, especially in regions with patchy connectivity.

AI-aware networking doesn’t just enhance performance—it reduces cost and increases predictability. Training large models, particularly in federated environments, can quickly exhaust budgets if the network isn’t smart enough to prioritize traffic effectively. BT’s emphasis on “workload-driven optimization” resonated with enterprise visitors, many of whom manage ballooning AI costs. Their demos showed how running the same AI job can be done faster and more cost-effectively by empowering the network with more responsibility.

Embedding AI in Network Layers

BT proposed embedding AI directly into the network layers, a concept not entirely new but now required at larger scales and higher speeds. BT demonstrated how AI models, both small and large, can be integrated within routers and edge nodes. These models can detect problems, predict usage spikes, and even anticipate failures before they happen.

This distributed intelligence means the network is not just reactive but also predictive, crucial in sectors where milliseconds matter. In an automated factory, preempting a network slowdown could prevent hours of machine downtime. In healthcare, an AI-aware network can dynamically prioritize patient monitoring traffic over system logs.

BT also highlighted how AI-in-network systems could support trusted AI governance. As the central point for all data flow, the network can serve as a checkpoint for compliance, anomaly detection, and even ethical controls. Think of it as an always-on traffic cop—monitoring not just what moves, but why and how it moves.

Scaling AI Without Scaling Chaos

Beyond technical specifics, BT presented a business narrative that resonated with enterprise leaders. Everyone wants to scale AI, but not the headaches of ballooning cloud bills, regulatory risks, and unexpected downtime. BT’s message was clear: a more intelligent, integrated network can mitigate these issues.

BT’s network telemetry optimizes AI decision paths in real-world applications.

BT cited examples where network upgrades enabled companies to deploy AI tools to thousands of edge locations without rewriting code or changing software stacks. In one case, a logistics firm reduced failed deliveries by 12% after optimizing AI decision paths using BT’s network telemetry. These were real firms with tangible results, moving the conversation beyond buzzwords.

BT also acknowledged limitations, particularly regarding interoperability with legacy hardware and vendor lock-in. However, their argument was clear: Even if you’re not yet engaging in advanced AI, prepare your network now. Waiting until workloads arrive is like upgrading your internet mid-video call—it’s already too late.

Networks: The Foundation Layer for AI

At Mobile World Congress, BT underscored an often-overlooked aspect of the AI conversation: the network. AI depends on the seamless flow of data, decisions, and outcomes, all reliant on networks built to handle the load. BT compared this relationship to a city and its roads—one cannot function without the other. Their focus was on smarter routing, edge processing, and distributed intelligence, creating infrastructure ready for AI. While the event showcased impressive models and demos, BT’s point stood out: no matter how advanced the AI, it can only perform as well as the network beneath it.