Intelligent Data Centres Issue 68 | Page 42

E D I T O R ' S Q U E S T I O N

IVO IVANOV , CEO , DE-CIX

Looking firstly at data centre operators , in particular colocation , we see the need for a mix of customers within their data centres . It may be tempting to lease these facilities to single hyperscalers – something which is happening with increasing frequency – but digital ecosystems depend on diversity .

Certainly , it can be tempting to pre-lease an entire facility to a single large-scale tenant , but many data centre operators will agree that this is not always an advantage in terms of margin , and it ’ s against diversity . You can ’ t build an ecosystem in isolation . Therefore , data centre operators need to aim for a healthy mix between single-tenant operations and multi-tenant facilities that enable network density , as well as AIrelated builds .
The future will demand dense connectivity hubs for enterprise networks , similar to what we see in the carrier world today . Looking at the network presence of enterprises like Apple , Netflix , or Disney , their data centre requirements are no different from that of a carrier network presence .
Following the model of the carrier hotel , the future will demand more inclusive ‘ network operator hotels ’. In the enterprise world , this means enterprisedense data centres where they can have their network Points-of-Presence ( PoP ).
The need for balance can also be seen in enterprise use of AI . Now , there is a strong demand for deployments for the training of LLMs and other large AI models , requiring specialised power densities . In current AI-related deployments , we see a ratio of around 70 % high power density for training purposes compared to 30 % for AI inference .
I have heard from multiple data centre operators the forecast of a significant shift in the coming years , in which this ratio will be reversed , so that 70 % of AIrelated deployments will be focused on inference and only 30 % on training .
Given that AI inference does not demand equivalent power densities as training , this processing can be carried out in standard colocation facilities , taking some of the focus back away from the hyperscalers . There is a clear need for enterprise-focused data centres .
When we look at enterprises , the strength of the hyperscalers can be seen as being to their advantage , because the cloud players are heavily focused on serving enterprises .
However , enterprises very often also want to operate their own infrastructure in hybrid scenarios , rather than placing everything in the cloud . Here , enterprise-rich colocation data centres again come into play .
Now , in some of the major business hubs , like New York City , colocation space is becoming scarce , so enterprises need to be creative about how to find the space they need .
The use of a distributed , data centre and carrier-neutral Internet Exchange offers two solutions to this challenge :
• On the one hand , the distributed interconnection model allows businesses to maintain efficient and secure data flows , even when their primary data centres are located outside of the main hub , as long as they choose a data centre that houses such a platform . This ensures they remain part of a vibrant digital ecosystem regardless of their physical data centre location .
• On the other hand , there is still space in these crowded hubs , but it can be that the available spaces are small and a bit piecemeal . Through distributed interconnection platforms , it is possible to connect these less ideal spaces with their current spaces as companies grow – providing seamless connectivity across multiple colocation data centre spaces , as well as with the larger ecosystem . �
42 www . intelligentdatacentres . com