AI's forgotten hero

Anthony Laing, General Manager of the Networking Business Unit at NEC XON. (Image: NEC XON)

---

Artificial intelligence dominates almost every technology conversation. Organisations are racing to deploy AI, vendors are competing to launch ever more powerful models and data centres have become symbols of digital ambition. But amid the excitement, one of the technologies enabling the AI revolution has quietly slipped from the spotlight: networking.

It may no longer command the attention it did during the internet boom of the late 1990s, but without robust, intelligent and resilient networks, AI simply doesn't exist. That's a reality the industry is beginning to acknowledge.

After attending HPE Discover Las Vegas 2026, I came away encouraged by a clear message repeated throughout the conference: AI is impossible without the network. From the opening keynote through the networking sessions, speakers reinforced the point that networking has become a critical enabler, and often a defining constraint, of the AI era.

For those of us who have spent our careers building networks, it was a welcome acknowledgement of something we've been saying for years. Networking has been the invisible foundation beneath every major technology wave: first enterprise connectivity, then cloud, mobile, streaming and now AI.

Each innovation captures the headlines while the network evolves behind the scenes, expanding its capacity and intelligence to support growing demands. It rarely receives credit because, when networking works well, people simply expect everything else to work.

AI changes the scale of that challenge dramatically. Every AI interaction generates data. Large language models process billions of parameters, while training requires vast clusters of GPUs exchanging information continuously at extraordinary speeds. Inference workloads must retrieve, analyse and deliver information almost instantaneously. Organisations deploying private AI environments must also move huge datasets securely between users, applications, storage and cloud environments. None of this happens without networks.

AI places greater demands on networks than many previous technology shifts. Inside the data centre, GPU clusters demand lossless fabrics running at 400 and 800 gigabits per second. A congested link or dropped packet can stall an entire training job, leaving millions of rands of computing power idle.

Basic connectivity is no longer enough. Networks must deliver ultra-low latency, massive bandwidth, resilience, security and intelligent traffic management simultaneously. They have become active participants in AI performance rather than passive infrastructure carrying data from one place to another.

Even asking an AI assistant a question illustrates the point. If the connection is unreliable, the AI experience deteriorates. Once AI is embedded in everyday business processes, connectivity becomes business critical. Organisations investing millions in AI platforms while overlooking networking risk creating expensive bottlenecks. The fastest GPUs and most advanced AI models cannot compensate for congested, unreliable or poorly architected networks.

For more than three decades, each major shift in enterprise technology has increased demands on the networks connecting systems, users and data. High-capacity optical networks expanded the scale and speed of information exchange. Cloud shifted workloads beyond the traditional corporate data centre, while streaming and mobile drove further increases in traffic and changed expectations around availability and responsiveness.

The current shift to AI is different mainly in scale. In 2023, NEC XON deployed Africa's first live 400G optical transponder. In 2025, its engineering work received the first TIP Gold Badge awarded for a network operating system. These developments form part of a broader industry trend towards higher-capacity, more programmable and increasingly automated networks.

The lesson is less about any particular technology than infrastructure. New applications attract attention first, while the networks required to make them practical evolve largely out of sight. AI is bringing that dependency into sharper focus because of the volume of data moving between processors, storage, applications and users. As AI workloads become more widespread, network architecture will increasingly influence how effectively organisations can deploy and use them.

For African organisations, the stakes are arguably higher than elsewhere. Much of the AI capacity businesses consume today sits in cloud regions thousands of kilometres away, meaning latency, resilience and connectivity costs shape the AI experience before a model ever runs.

At the same time, data sovereignty requirements and the economics of moving large datasets are driving interest in private and sovereign AI environments hosted on the continent. Both paths lead to the same conclusion: the network is either a constraint on African AI ambitions or the foundation that makes them possible. Getting it right is not simply a technical detail. It's a competitive question for the continent.

As enterprises accelerate AI adoption, networking should no longer be treated as background infrastructure or considered only after strategic decisions have been made. Network architecture directly influences AI performance, scalability, security and user experience. 

The organisations that extract the greatest value from AI won't necessarily be those with the biggest models or most powerful hardware. They'll be those with infrastructure capable of supporting AI reliably, securely and at scale.

Technology trends will continue to evolve. After AI, another breakthrough will capture the industry's imagination, perhaps quantum computing or something we haven't yet anticipated. Whatever comes next will depend on the technology carrying the world's data.

For leadership teams planning their AI journey, the advice is simple: make the network the first conversation, not the last.