Futureproofing datacentre interconnect: The backbone of an AI strategy

The AI revolution is transforming networks, demanding greater capacity and agility in datacentre interconnectivity (DCI) to support diverse workloads. This report explores the challenges AI presents and offers recommendations for providers to adapt through flexible connectivity and NaaS models.

The AI revolution is impacting nearly every aspect of technology infrastructure, but one critical element has received less attention – data centre interconnect (DCI). While many discussions focus on either the software or the data centre infrastructure itself (think GPUs, power constraints and cooling requirements), the connective tissue that makes AI workloads possible has remained largely overlooked.

This oversight risks creating a bottleneck. Central to the problem is that AI does not behave like the cloud workloads that traditional DCI was designed to handle. Traditional cloud traffic was (in most circumstances) stable, predictable, and symmetrical. AI disrupts this model: not only driving greater capacity needs, but also creating more volatile, unpredictable, multi-directional flows across core, regional and edge locations. These pressures are emerging at a scale and intensity that static, fixed DCI models were never built to manage.

But the challenge goes beyond workload intensity. AI introduces a new diversity of demands:

• Not all AI networking demands are equal: different types of AI traffic, such as those generated by training versus by inferencing, place different requirements on bandwidth, latency and resilience.
• At a global industry level, future scenarios shaped by regulation, geopolitics and technical advances add further uncertainty to where and how data must flow.
• Specific market contexts at a country level matter: countries are pursuing different AI strategies, which will shape their DCI needs in distinct ways.

If you are not a subscriber, enter your details below to download an extract of the report

Taken together, this variability demonstrates that DCI strategies cannot remain fixed and uniform. Supporting AI at scale will depend on DCI connectivity models that are more flexible and adaptive than those designed for the cloud era. Network-as-a-Service (NaaS) offers a solution. By enabling bandwidth to scale dynamically in line with workload demands, NaaS reduces the inefficiencies of static contracts and over-provisioning, and allows DCI to keep pace with AI’s heterogeneous demands.

With hundreds of billions being invested in AI infrastructure, organisations that fail to evolve their connectivity strategies risk undermining these investments. Conversely, those who adapt fastest with dynamic interconnect models will be best placed to capture value as AI reshapes industries.

This report explores how AI is putting new demands on connectivity and the implications for DCI across multiple dimensions: how AI traffic breaks the mould set by traditional cloud workloads; how the AI lifecycle creates variability across workload types; how future scenarios shaped by forces such as regulation will shape DCI; and finally, how these dynamics play out in practice depending on national strategies and market dynamics.   

Together, these perspectives underline a common conclusion: DCI must evolve into a more flexible, dynamic model if it is to support the next wave of AI growth.

Table of contents

• Executive summary
• Why AI breaks the mould: How AI traffic redefines connectivity demands
• The AI workload lifecycle: Each phase has a unique and dynamic DCI requirements
• Beyond worload dynamics, future scenarios will shape how DCI must evolve
• Comparing strategic AI markets in Asia: India and Singapore
• Conclusions

Related research

• Why Layer 1 needs NaaS: A guide for enterprises
• Enabling customer centricity in enterprise networks through NaaS
• The telecom industry in the AI era: How telcos can follow the money
• Enterprise AI: What are the best opportunities for telcos?
• Networks for AI: Segmenting the growth

---