Why the AI Race Will Be Won by Networks, Not Just GPUs

The AI conversation has become increasingly dominated by one metric: compute.

Every major announcement today seems to revolve around the number of GPUs deployed, the size of a new AI factory, or the billions being invested into hyperscale data centres. Yet, beneath the excitement lies a less glamorous reality. AI is only as powerful as the infrastructure that connects it.

During a recent discussion with Rohit Chowdhary, Head of Advanced Consulting Services for Europe, Middle East and Africa at Nokia, one theme stood out repeatedly: the future of AI will depend as much on networks as it does on compute.

It is a perspective that deserves greater attention, particularly in the Middle East, where governments and enterprises are rapidly positioning themselves as leaders in the global AI economy.

By Srijith KN,

Contributing Editor

The Hidden Bottleneck

Much of the industry’s focus remains centred on processing power. However, AI workloads are becoming increasingly distributed. Data is constantly moving between servers, racks, data centres and edge locations. If the network connecting those resources cannot keep pace, even the most powerful GPU clusters become constrained.

Think of it as building a city with the world’s fastest cars but failing to invest in roads.

This is where Nokia sees the next phase of AI infrastructure evolving. Rather than viewing data centres, connectivity and edge environments as separate entities, the company advocates treating them as a unified AI grid, where compute and networking capacity are designed together from the outset.

The concept is particularly relevant as countries across the Gulf continue to announce large-scale sovereign AI initiatives and next-generation data centre investments.

Beyond Connectivity

What is becoming increasingly clear is that networking itself is undergoing an AI transformation.

From AI-assisted radio networks and intelligent optical systems to autonomous operational platforms, the network is no longer a passive transport layer. It is becoming an active participant in how AI workloads are managed, routed and optimised.

One innovation showcased by Nokia that caught my attention was its compact 800G coherent switching technology. Designed to reduce physical infrastructure requirements while improving energy efficiency, it reflects a broader shift underway in the industry. Future AI infrastructure will need to be faster, denser and significantly more sustainable than today’s environments.

As AI scales, efficiency may become just as important as performance.

The Gap Between Strategy and Execution

Another observation from the discussion was how many organisations still struggle to move from AI ambition to operational reality.

Across industries, there is no shortage of AI strategies, roadmaps and proof-of-concept projects. The challenge begins when businesses attempt to integrate these initiatives into existing operational environments.

This is where Nokia’s recently established Automation Excellence Practice offers an interesting model. Rather than stopping at advisory engagements, the company is investing in multidisciplinary teams capable of building real-world automation pipelines, agentic AI use cases and operational workflows.

It reflects a broader market shift. Enterprises are no longer asking whether AI can create value. They are asking how quickly that value can be realised.

Security Must Evolve Alongside AI

Perhaps the most important takeaway is that AI infrastructure cannot be separated from security.

As governments seek greater control over data sovereignty and organisations deploy increasingly intelligent systems, security can no longer be treated as an add-on. It must be embedded directly into the architecture.

This is particularly relevant in the Middle East, where the growth of sovereign AI capabilities is creating new expectations around resilience, privacy and trust.

The next generation of digital infrastructure will not simply need to be intelligent. It will need to be secure by design.

The Road Ahead

The race to dominate AI is often framed as a competition for compute capacity. But conversations with technology leaders increasingly suggest that the winners will be determined by something more fundamental.

AI will ultimately be limited by the quality of the infrastructure that connects it.

The organisations that succeed will not be those with the most GPUs. They will be those that build intelligent, automated and secure networks capable of turning AI potential into measurable outcomes.

The future of AI may be powered by compute, but it will be enabled by connectivity.