The rapid growth of AI is often framed as a compute challenge.

More GPUs. More data centres. More processing power.

However, in practice, AI is placing increasing pressure across the entire network stack.

From core transport to edge infrastructure, operators are seeing demand rise at a pace that traditional planning models were never designed to handle.

As a result, AI network demand is becoming one of the defining challenges for infrastructure teams.

Why AI Is Driving AI Network Demand So Quickly

AI workloads place very different demands on infrastructure compared to traditional applications.

Model training involves large-scale data movement between systems, often across distributed GPU clusters. In contrast, inference workloads require low latency and consistent throughput to deliver real-time responses.

As a result, in GPU-based environments, this creates significantly higher east-west traffic across data centre fabrics and places sustained pressure on switching infrastructure.

Storage constraints further increase this pressure, with enterprise SSD and NVMe supply tightening.

This means that every AI deployment drives demand not just in compute, but in:

• optical transport networks
• switching and routing infrastructure
• storage and interconnect systems

Consequently, the network quickly becomes the limiting factor in many environments.

The Impact on Infrastructure

As AI adoption accelerates, several trends are becoming clear.

1. Increased Bandwidth Requirements

In practice, AI workloads generate significantly higher east-west traffic within data centres, as well as increased demand across core and edge networks.

You can already see this in spine-leaf architectures, where increased east-west traffic creates congestion and drives demand for higher-capacity links such as 100G, 400G, and beyond.

This trend is also reflected in optical networks, where demand is outpacing supply, as highlighted in our analysis of optical networking equipment shortages.

2. Higher Equipment Utilisation

Unlike traditional workloads, AI infrastructure often runs closer to capacity.

Therefore, teams have far less tolerance for inefficiency, and degraded components, thermal performance issues, and power delivery weaknesses become much more visible.

3. Accelerated Hardware Cycles

AI demand is pushing operators toward faster upgrade cycles to keep up with growth.

However, this creates additional cost pressure and increases reliance on OEM supply chains.

4. Supply Chain Constraints

Demand for key components such as optics, switching hardware, and high-performance servers continues to outpace supply.

As a result, this leads to:

• extended lead times
• price volatility
• limited availability of critical parts

This trend also affects compute infrastructure, where enterprise server lead times continue to increase as demand grows.

In some cases, critical components are now on lead times of several months, directly impacting project delivery.

Why This Creates a Problem

As a result, the combination of rising demand and constrained supply creates a difficult position for operators.

On one hand, they need to scale quickly.

On the other, they face:

• increasing costs
• limited availability
• pressure to maintain performance

This pressure is already being felt in delayed projects, stretched budgets, and difficulty sourcing critical hardware.

In addition, oversubscription in network design makes the problem worse in many environments. As traffic patterns become less predictable, link utilisation can spike rapidly, increasing the risk of congestion and performance degradation.

Ultimately, traditional approaches based on continuous replacement become harder to sustain.

The Risk of a Replacement-Only Strategy

However, relying solely on new hardware to meet AI-driven demand introduces several risks.

Cost Escalation

Frequent upgrades significantly increase capital expenditure.

Lead Time Delays

Projects can be delayed by months due to hardware availability.

Operational Disruption

Replacing infrastructure at scale introduces complexity and risk.

Sustainability Pressure

Shorter lifecycle cycles increase waste and environmental impact.

A More Resilient Approach to AI Network Demand

To manage AI-driven growth effectively, operators need a more flexible approach.

As explored in our previous blog on reducing network energy costs, many networks already contain untapped capacity that can be optimised before upgrading.

Instead of focusing purely on replacement, the focus shifts to:

Extending Existing Infrastructure

Well-maintained equipment can continue to support increased demand when properly optimised.

Repair and component-level intervention restore performance, address inefficiencies, and extend usable life.

In addition, repair provides valuable insight into failure trends, as outlined in our work on network equipment root cause analysis.

Maximising Available Capacity

Before upgrading, it is critical to understand how existing infrastructure is being used.

In many environments, there is still untapped capacity that can be utilised more effectively through optimisation and better workload distribution.

Using Refurbished Equipment Strategically

Refurbished enterprise servers provide a practical way to scale capacity without waiting for new supply.

A generation behind current models is often:

• readily available
• proven in production
• capable of supporting increased workloads in many environments, depending on architecture and design

As a result, operators can scale quickly where new hardware is unavailable or delayed.

Reducing Dependency on OEM Supply Chains

Diversifying supply through repair, reuse, and secondary markets reduces risk and improves flexibility in how infrastructure is scaled.

The Role of Circular Economy in AI Infrastructure

As AI demand grows, the industry faces a choice.

Continue accelerating replacement cycles, or adopt a more sustainable and resilient model.

A circular approach focuses on:

• extending equipment life
• repairing rather than replacing
• redeploying assets where possible

In practical terms, this means keeping equipment in service for longer through repair, reuse, and refurbishment.

As a result, this not only reduces environmental impact but also improves resilience in the face of supply constraints.

What This Means in Practice

How do we scale efficiently with what we already have?

That shift is essential as AI network demand continues to grow.

Call to Action

If AI workloads are increasing demand across your network, now is the time to review how your infrastructure is being used.

Comtek works with operators to identify available capacity, extend equipment life, and provide practical ways to scale without relying solely on new hardware.

Speak to the team to review your current infrastructure and identify immediate opportunities

Frequently Asked Questions

Why does AI increase demand on network infrastructure?

AI workloads generate large volumes of traffic between systems, particularly in GPU-based environments. This increases pressure on switching, routing, storage, and optical transport infrastructure.

Is AI demand only a data centre issue?

No. While AI demand is often most visible inside data centres, it also affects core transport, edge infrastructure, interconnects, and the wider network supporting data movement and low-latency access.

Why does the network become a bottleneck for AI workloads?

As AI traffic increases, east-west traffic inside data centres and between systems can rise sharply. In oversubscribed or constrained environments, this can create congestion, higher utilisation, and performance issues across the network.

Can refurbished equipment support AI-related growth?

In many environments, yes. It depends on the architecture and workload, but refurbished hardware can provide a practical way to add capacity quickly where new hardware is unavailable or delayed.

How can operators scale without relying only on new hardware?

Operators can review utilisation, optimise existing infrastructure, extend equipment life through repair, and use refurbished hardware strategically. This reduces dependence on long OEM lead times and supports a more flexible scaling strategy.

How does a circular economy approach help with AI infrastructure?

A circular approach helps operators keep equipment in service for longer through repair, reuse, and refurbishment. This reduces waste, lowers environmental impact, and improves resilience when supply chains are under pressure.

Conclusion

AI is changing the way infrastructure is built and used.

Ultimately, the challenge is not just about adding more capacity.

It is about using existing infrastructure more effectively, making smarter upgrade decisions, and building a more resilient and sustainable network.