Key takeaways from DCD Connect: De-risking MENA data centres with a resilient commercial approach

Download Listen

My key takeaway from two days in Dubai at the MENA region’s premier data centre event, DCD Connect: Regional data centre operators need to take more ownership in defining the hosting market of tomorrow.

The need for collaboration across the MENA region was a constant narrative across the two days at DCD Connect MENA. Most compelling was the keynote panel featuring CEOs and peers from Khazna, Du, Pure DC and Equinix, where a narrative emerged on the benefit of collaboration both at a company and country level, in an attempt to ‘grow the overall pie’, as opposed to considering the MENA data centre market as a zero sum game. However, while there is consensus that AI is driving growth in ‘the pie’, I didn’t feel there was a clear narrative around the target workloads with which to fill ‘the pie’, especially given significant capacity currently under development. While I agree that the AI hosting market is neither a ‘race’ or a zero-sum game, claiming the pie/ finish line are completely unknown is to me inaccurate, and akin to standing still on commercial strategy under the pretence of retaining flexibility. Two distinct markets are taking shape:

• AI training workloads will be contested by a global market, primarily driven by power cost/availability, hardware accessibility and facility design.

Data centre operators should work closely with frontier model developers (hyperscalers plus the likes of OpenAI, Anthropic, Mistral) to understand how they can collaborate – this is a global market with an increasingly crowded group of suppliers and few buyers. The MENA region is well positioned from a power availability point of view, but restricted by lack of first mover advantage, facility supply chain constraints (eg transformer wait times are c.3 years) and fundamentally a lack of leading AI companies based in the region, although the likes of SaudiAI and G42 are looking to change this.

• AI fine tuning and inferencing hosting demand will be served regionally, driven by with the specific location dependent on various factors such as the domestic data centre capacity, data sovereignty regulations, latency requirements etc.

Data centre operators should work closely with regionally dominant service provider partners to try and stimulate local demand for inferencing workloads. The hyperscalers had great success in leveraging professional services such as executive education and technical pre-sales as loss-leading yet revenue generating activities, and data centre operators should work alongside regional managed service providers, systems integrators and content providers, to raise the digital maturity of their locale, test new inferencing use cases and stimulate demand for the sector. This is a key potential area for collaboration both between data centre operators and with their service provider customers, with potential to both grow market demand and build relationships with key future customers and partners.

It felt to me that data centre operators are the other side of the initial ‘build and they will come’ wave, but have not yet identified the optimal demand generation strategies which consider future enterprise IT trends, and are instead reliant on partners with vested interests, such as the hyperscalers, for this industry trend and demand forecasting. This uncertainty was present both in terms of segmenting demand by specific use case (AI training v AI inferencing v cloud compute), as well as the AI commercialisation channels, and therefore the data centre customers, and while nobody can predict the next 5 years of hosting demand, operators can still position themselves for commercial success through a resilient and agile commercial strategy, backed by scenario planning.

It felt to me that data centre operators are the other side of the initial ‘build and they will come’ wave, but have not yet identified the optimal demand generation strategies which consider future enterprise IT trends, and are instead reliant on partners with vested interests, such as the hyperscalers, for this industry trend and demand forecasting. This uncertainty was present both in terms of segmenting demand by specific use case (AI training v AI inferencing v cloud compute), as well as the AI commercialisation channels, and therefore the data centre customers, and while nobody can predict the next 5 years of hosting demand, operators can still position themselves for commercial success through a resilient and agile commercial strategy, backed by scenario planning.

It is worth considering that cloud computing and the hyperscalers have thrived leveraging a platform model running on largely homogenous infrastructure. With AI adoption mandating increasing infrastructure diversity and increased dependence on specialised hardware, it remains to be seen whether they will be in a position to dominate in a similar oligopolistic market structure when it comes to commercialising AI with enterprises, with the emergence of GPUaaS and ‘AI hyperscalers’, such as CoreWeave, early evidence backing up this hypothesis. Under the paradigm of AI scaling laws, they are likely to dominate when it comes to frontier model development, however with cloud repatriation a developed trend in digitally mature markets, especially with the SaaS and other service providers, the raft of future SaaS providers offering agentic AI solutions may turn to other hosting opportunities, such as with colocation providers, as opposed to the hyperscalers, in search of cost optimisation and configurability benefits. Data centre operators must prepare now to build relationships with these potential future customers.

In summary, don’t sit back and rely on service provider customers to bring you revenue and guide your facility design. Building a proactive, independent commercial strategy is imperative for MENA data centre operators to position themselves to both stimulate market demand, and maximise their share of it come 2027, and looking beyond that to 2030.