The technology industry is entering a new era defined by an AI Supercycle that is reshaping infrastructure, investment, and innovation worldwide, writes Samer Lutfi, head of MEA hyperscaler and AI cloud at Nokia.
This is not a passing trend or a narrow software evolution. It is a multi-decade shift in which artificial intelligence becomes deeply embedded across industries, economies, and infrastructure. As AI systems evolve from static models into real-time, distributed, and autonomous capabilities, they are triggering a powerful feedback loop. More AI drives more data, more compute, and ultimately, more demand on the networks that connect it all.
For Africa, this moment carries unusual weight. The continent is not simply adopting AI-enabled services but is still actively building the infrastructure foundations that will determine whether it participates in the global intelligence economy as a creator or a consumer.
The critical distinction is this: AI is often framed as a software revolution, but in practice, it is an infrastructure transformation.
The network takes centre stage
AI systems are fundamentally dependent on data movement at unprecedented scale. Training large models requires transferring vast datasets across distributed compute clusters, while real-time inference depends on the seamless exchange of millions of transactions between users, edge environments, and cloud platforms.
In this environment, three performance characteristics become non-negotiable.
First, throughput. AI workloads are orders of magnitude more bandwidth-intensive than traditional enterprise traffic. Networks designed even five years ago are already approaching their limits under these new demands.
Second, latency consistency. It is no longer sufficient to optimise for average latency. AI-powered applications, require predictable, deterministic performance. Variability or jitter is not just inconvenient but undermines the reliability of the service itself.
Third, traffic patterns. The traditional north-south model, where users connect to centralised data centres, is giving way to dense east-west flows within and between data centres. This shift is placing new pressure on IP routing architectures and fundamentally redefining the role of optical transport networks.
The implication is that the performance ceiling of AI is increasingly determined by the network beneath it.
Reinventing the transport layer
Nowhere is this transformation more visible than in the evolution of fibre, IP routing, and optical transport.
At the physical layer, demand is accelerating for high-count fibre deployments and dense wavelength division multiplexing to support terabit-scale capacity. Networks are being designed not for incremental growth, but for exponential expansion.
Recent investments across the continent already demonstrate how rapidly this shift is accelerating. Airtel Africa Telesonic, for example, has partnered with Nokia to build a high-capacity terrestrial fibre network spanning East and Central Africa, linking inland terrestrial infrastructure directly to major subsea systems such as 2Africa. The initiative is designed to create resilient digital corridors capable of supporting the explosive growth in cloud, hyperscale, and AI-driven traffic flows across multiple African markets. With deployments leveraging coherent optical technologies capable of supporting up to 38 terabits per second, networks are no longer being designed merely for connectivity, but for participation in the AI economy itself.
At the routing layer, operators are contending with far more dynamic and complex traffic matrices. Modern IP networks must scale to extreme throughput levels while enabling intelligent traffic engineering in real time.
But it is the optical layer that is undergoing the most profound change. Once treated as a passive conduit, it is becoming an intelligent, software-defined system that can provision capacity on demand, anticipating failures, and adapting dynamically to shifting workloads. In the context of AI, optical networks are no longer just transport mechanisms; they are active participants in service delivery.
The rise of distributed intelligence
At the same time, AI is becoming more geographically distributed. Inference workloads increasingly need to be processed closer to where data is generated, whether in cities, industrial sites, or rural environments.
This shift toward edge computing is forcing a redesign of national and regional network architectures. Intelligence is no longer concentrated in a handful of centralised data centres but rather dispersed across a continuum that spans core, metro, and access layers.
For operators, this introduces new complexity. Networks must now support stringent latency guarantees across a far broader footprint, while maintaining efficiency and scalability. Capabilities such as software-defined networking, automated traffic engineering, and network slicing are becoming essential tools for managing this complexity and ensuring that AI workloads receive the performance they require.
Africa’s unique opportunity
Unlike more mature markets constrained by legacy infrastructure, many African operators still have the flexibility to design networks with the future in mind, creating a strategic opportunity.
By building cloud-native, software-defined networks from the outset, operators can embed automation and intelligence directly into their infrastructure. By dimensioning optical and IP layers for the tenfold growth that AI-driven data consumption will deliver within the decade rather than incremental demand, they can avoid the cycle of constant retrofitting that has challenged other regions.
Equally important is the role of cross-border connectivity. Stronger terrestrial and subsea backbone networks linking data centres across the continent, can dramatically reduce latency and cost for intra-African traffic. This would enable regional cloud and AI platforms to operate natively within Africa, while improving resilience through route diversity.
The result would be a more integrated digital economy, capable of supporting innovation at scale.
The AI infrastructure opportunity is, however, not limited to hyperscale corridors and capital cities. Expanding affordable fibre access into underserved communities remains equally critical to building inclusive digital economies.
In South Africa, Fibertime is working with Nokia to expand fibre broadband access into townships and underserved communities, with ambitions to connect millions of homes through low-cost, pay-as-you-go broadband models.
By combining fibre access, IP networking, and AI-driven automation, these deployments demonstrate how modern software-defined networks can lower the cost of connectivity while broadening participation in the digital economy. Africa’s future AI ecosystem will ultimately depend not only on compute infrastructure, but also on how widely digital access is distributed across society.
Reliability in the age of physical AI
As AI moves beyond digital applications into real-world systems the tolerance for failure diminishes sharply.
In this context, reliability becomes a defining characteristic of network design. Deterministic performance and ultra-high availability are no longer aspirational. They are foundational requirements and networks must not only deliver capacity, but do so with precision and consistency, even under extreme conditions.
This is driving increased focus on automation, predictive analytics, and self-optimising network behaviours, which are the capabilities that allow infrastructure to detect, adapt, and resolve issues before they impact services.
Building for the next decade
Globally, collaboration between telecom operators, cloud providers, and hyperscalers is already shaping how next-generation infrastructure is deployed, and several lessons are emerging.
Firstly, speed matters. The ability to scale infrastructure in months rather than years is becoming a competitive differentiator.
Openness also matters. Disaggregated architectures give operators greater flexibility and bargaining power in how networks are built and evolved.
And, finally, policy matters. Governments play a critical role in creating regulatory environments that reduce investment risk while aligning infrastructure development with national digital strategies.
From consumer to compute hub
If Africa makes the right infrastructure decisions over the next three to five years, the implications could be transformative.
The continent has several structural advantages such as abundant renewable energy potential to power data centres sustainably, a young, rapidly urbanising population generating unique and valuable datasets and growing pools of technical talent. These are precisely the assets that the global AI economy will increasingly demand.
With the right backbone infrastructure in place, Africa could host AI workloads at scale, develop sovereign AI capabilities tailored to local contexts, and attract greater levels of cloud and hyperscale investment. It could shift from being a downstream consumer of AI services to an active participant in their creation and distribution.
The window to shape that outcome is open, but it will not remain so indefinitely. The AI Supercycle will not be defined solely by breakthroughs in algorithms or models. It will be defined by the strength, intelligence, and reach of the networks that support them.
And, in Africa, the opportunity to build them and lead remains wide open.