Artificial intelligence (AI) is dominating boardroom discussions, keynotes, and investment pitches. Everyone talks about the efficiency gains, the competitive edge, the transformation it promises.
However, from my perspective as a seasoned network professional, AI only delivers its full potential if the network underneath it is robust. And increasingly, the network itself runs better when AI helps manage it.
The realities of running AI in SA
AI might look straightforward, but in practice, it can stress networks in ways many teams do not expect. Training models in a data centre drives huge volumes of sideways traffic between GPUs, the digital equivalent of gridlock on a major highway.
At the edge, the story is similar. Whether it is a Durban retailer that relies on AI-powered checkout scanners, or a warehouse outside Upington streaming video analytics to the cloud, when older uplinks or tired switches cannot cope, the AI gets blamed for “failing”.
That distinction matters in South Africa. IT budgets are tight, teams are stretched, and there is no room for costly missteps. A failed AI project does more than waste money. It also erodes confidence and hinders other transformation plans.
Where AI pays the network back
The benefits become clearer when you flip the equation. Most IT teams in the country are small and often responsible for dozens of sites. They cannot be everywhere at once.
Troubleshooting the old way, for example, combing through logs after a midnight outage, drains energy and time. AI-driven telemetry can change that. It does not replace the engineer, but it points them directly to the real problem, such as a failing cable, an overworked access point, or a misconfigured VLAN.
AI only delivers its full potential if the network underneath it is robust.
We have already seen examples in action. Predictive analytics have warned of downtime risks at Johannesburg campuses before classes were disrupted. Bandwidth optimisation has kept remote schools connected to e-learning platforms. Automated alerts have saved hours of guesswork in mining operations where a trip to the site is expensive.
For CIOs, the shift is significant: less firefighting, more proactive management, and more space for strategic improvements.
Building an AI-ready network
Making a network “AI-ready” is not about buying everything new. It is about getting the fundamentals right. The starting point is visibility. Without real-time awareness of device health, traffic patterns and anomalies, companies are working blind.
Once that foundation is there, AI tools can do something useful with the data, like flagging patterns and giving teams a chance to act before users notice.
Capacity is the next piece of the puzzle. Too many environments still run gigabit uplinks, while expecting WiFi 6, WiFi 7 and heavy analytics to perform flawlessly. Upgrading chokepoints, planning PoE budgets properly, and tightening cabling standards yield faster gains than many big-ticket projects.
Segmentation is equally essential. AI workloads mix tills, sensors, cameras and laptops. Left on a flat network, that combination magnifies risk. Proper segmentation contains failures, makes audits easier and strengthens security.
And finally, networks must reflect the specific conditions of South Africa. Rooftops in Johannesburg are crowded with competing signals. Fibre builds are often delayed by wayleaves or strikes. Warehouses and factories deal with dust and heat. Wireless backhauls, rugged radios, and designs that anticipate these obstacles are not luxuries; they are survival tactics.
Why this matters now
When networks are designed with these realities in mind, AI stops being a buzzword and starts delivering. Retailers keep analytics running at the till, campuses keep exams online even if a switch fails, and mines keep safety systems connected through dust and distance.
For executives weighing AI investments, the right questions are simple: can our network carry the load, and can AI make our network easier to carry?
When the answer is yes to both, projects move from theory to value, and they keep delivering long after the launch day is over.
The opportunity is not to chase AI for its own sake, or to rip out infrastructure nobody can afford to replace. It is to build networks that are honest about local conditions, supported by intelligence that eases the pressure, and able to keep people connected without unnecessary drama.
That is how AI moves from hype to business impact in South Africa.
Share