By Mandy Duncan, Nation Supervisor of HPE Networking South Africa
Attempting to scale AI on outdated digital foundations is like taking part in Jenga on a wobbly desk: early wins stack up rapidly, however as ambition grows, each underlying weak point is magnified. South African organisations are reaching this level with AI. Adoption is accelerating, experimentation is widespread, and productiveness positive aspects are seen, but many try to construct bold AI capabilities on foundations that had been by no means designed for the dimensions and complexity of the AI period.
Whereas a number of elements form AI success, from information and abilities to governance and safety, the community, usually handled as background infrastructure, is rising as some of the essential enablers of sustainable AI success.
Momentum is actual, however uneven
AI’s financial potential is effectively established. IDC analysis exhibits organisations reaching a median return of $3.7 for each $1 invested in generative AI, with main adopters seeing returns exceeding $10.
South Africa is clearly a part of this shift. PwC’s native modelling suggests AI may contribute 1.2 proportion factors to nationwide GDP over the subsequent decade, even at right now’s adoption ranges.
But a niche is rising between adoption and execution. Whereas a KPMG survey exhibits that 71% of African CEOs are investing in AI, respondents additionally cite integrating AI into core operations as their prime problem. Actually, a lot of right now’s adoption is bottom-up and tactical — groups experimenting with instruments and not using a coordinated plan for scale, safety or long-term sustainability.
That method delivers fast wins, however it additionally creates threat. With out the proper foundations, early productiveness positive aspects can plateau; technical debt accumulates and confidence in AI erodes, not as a result of the know-how fails, however as a result of the surroundings can not help it.
AI exposes weaknesses within the community
AI workloads behave very in another way from conventional enterprise purposes. Coaching fashions generate large east-west site visitors throughout information centres and cloud environments, whereas inference calls for ultra-low latency and constant efficiency to ship real-time predictions and selections.
On the excessive finish of the spectrum, the quickest supercomputer on this planet – hosted on the Lawrence Livermore Nationwide Laboratory – can carry out quintillion calculations per second. This degree of high-performance computing is barely attainable as a result of the community can transfer huge volumes of knowledge predictably, securely and at pace, underscoring simply how intensive AI workloads are in comparison with typical enterprise purposes.
Conventional networks, designed for predictable north-south site visitors, weren’t constructed for this scale or volatility. Immediately, networks should securely join infrastructure, purposes, customers and information, whereas supporting compute-intensive workloads and more and more advanced hybrid environments. When networks fail to maintain up, the results are tangible: congestion slows down fashions, compute is wasted, downtime will increase, and the return on AI funding erodes.
South Africa’s constraints elevate the stakes
These challenges are amplified domestically. Organisations face persistent abilities shortages, infrastructure constraints, and rising regulatory and compliance necessities. Community transformation is capital-intensive, and few can afford to switch legacy environments wholesale, forcing many to modernise in a phased, pragmatic method.
Because of this, organisations throughout sectors are starting to rethink not simply how they improve their networks, however how these networks are conceived within the first place. Fairly than bolting AI onto legacy environments, frontrunners are shifting towards AI-native programs, designed from the bottom up with AI as a core element.
In observe, this implies embedding intelligence straight into the community administration layer. AI-native networks simplify operations, enhance productiveness, and ship extra dependable efficiency at scale by constantly analysing community behaviour and predicting points earlier than they affect customers. Groups acquire deeper visibility into efficiency throughout purposes, infrastructure and third-party providers, permitting them to rapidly pinpoint the supply of issues and resolve incidents in hours moderately than days.
The result’s distinctive consumer and operator experiences. A number of of HPE’s hospitality companions, for instance, are utilizing AI-enabled networks to recognise returning company as quickly as they join, personalise digital interactions in actual time, and securely help high-density convention venues with a number of distributors shifting on and off the community. At large-scale occasions such because the Nedbank Golf Problem, AI-native networking has enabled hundreds of attendees to attach seamlessly whereas receiving real-time, location-aware info on their units, demonstrating how community design straight shapes expertise and operational efficiency.
This shift additionally displays a broader transfer towards modular community design, the place capabilities function as versatile, cloud-based elements moderately than a single tightly coupled system. The profit is a community that may reply dynamically to altering calls for whereas supporting extra clever, automated operations and lowering reliance on handbook, ticket-based processes. Crucially, modularity have to be paired with interoperability and vendor-neutral requirements that permit organisations to mix best-of-breed elements with out changing into locked right into a single provider.
In a skills-constrained market, this issues. By lowering handbook configuration and troubleshooting, and enabling phased upgrades throughout hybrid environments, AI-native networking makes modernisation extra achievable even in resource-constrained settings, reducing operational prices whereas easing stress on scarce abilities.
Combining AI-native networking with modular design, which lays the foundations for extra goal-driven, agentic AI, permits organisations to simplify community administration and keep dependable efficiency, at the same time as AI-driven purposes place heavier and fewer predictable calls for on the community.
Constructing networks with AI and for AI
The trail ahead just isn’t disruption for its personal sake, however deliberate, phased modernisation. AI-ready networks have to be constructed each with AI and for AI.
AI-native networks can simplify deployment, automate troubleshooting, strengthen safety, and scale back operational complexity. Constructed for AI, they ship superior, high-speed and low-latency architectures, dependable information motion and compliance by design, making certain AI workloads can run effectively and securely at scale.
Importantly, compliance and safety can now not be bolted on afterwards. As AI expands assault surfaces and regulatory scrutiny intensifies, networking and safety have to be architected collectively. When designed in tandem, compliance turns into simpler to handle moderately than more durable to implement.
South Africa’s AI second is already underway. Whether or not it turns into a sturdy benefit or a fragile stack of early wins will depend upon the energy of the foundations beneath. With out resilient, AI-native networking, AI initiatives stall at pilot stage, regardless of how promising the use case. For enterprise leaders, the problem is obvious: deal with community modernisation as a strategic enabler of scalable AI, not a technical afterthought, or threat constructing AI ambition on foundations that can’t maintain.
