Will Future Cities Form Around Data Centres?

In the past, most cities grew around large bodies of knowledge such as libraries, learning institutions like universities and cultural institutions. Alexandria (Egypt) became a hub of intellectual activity thanks to the legendary Library of Alexandria. Athens (Greece), developed around educational institutions like Plato’s Academy and Aristotle’s Lyceum, shaping Western philosophy and culture. In Europe, cities like Oxford and Cambridge grew around their prestigious universities, while Paris flourished intellectually with the founding of the Sorbonne.

The University of Bologna in Italy, Baghdad’s House of Wisdom, and Boston’s parallel to Harvard University all highlight how knowledge centres have historically driven urban development, attracted scholars, inspired innovation, and contributed to cities’ cultural and intellectual life. I am about to draw parallels. Patience.

Fast-forward to present-day cities. We access many, if not all our artificial intelligence (AI) services, through the Internet. While this might be the most familiar piece of AI we know, we need to look under the hood to see the infrastructure making it possible. More specifically, we should examine how Africa can play a role in the infrastructure side of AI. Africa needs AI workload-ready data centres.

These data centres are specifically designed to handle the demands of modern AI applications, which include training machine learning models, running inference tasks, and processing large datasets. The infrastructure needed for such a data centre goes beyond the traditional data centre requirements, focusing on scalability, high throughput, specialised hardware, and efficient cooling systems. There do exist data centres that handle the servers and infrastructure that accompany these AI tools. However, they are not yet at the scale we have in Europe, Asia and North America. Power is the first bottleneck. AI data centres typically require power density levels of 10-30 kW per rack, compared to the 4-10 kW per rack found in traditional data centres.

With all that power consumption, comes the need to keep things cool. Cooling becomes a key factor in ensuring that the data centres can accommodate large GPU clusters. Various methods of cooling already exist in our data centres such as Hot Aisle/Cold Aisle Containment. And some even use free cooling where appropriate as it requires the use of the external environment (cold air, water bodies) to aid in cooling the data centre. However, in warmer climates, in Africa, this is often not a viable option.

Direct-to-Chip Liquid Cooling is a more promising solution. However, there are questions about the sustainability of such methods, including immersion cooling. With large volumes of water being dedicated to this mode of cooling, some questions arise around the sustainability of this practice and its probable ethical concerns, especially when deployed where water as a resource is limited.

While the challenges mentioned above can be fixed with some retrofitting of the existing data centres, the next challenge is this; how does Africa get the GPUs and HPCs to fill the data centres? The investment cost is significantly high, and the lead times are a pain with wait times from eight months onwards. How does Africa, after building data centres that are supplied with primarily renewable energy, attract compute to the continent?

One key issue facing this rather simple, yet complex ask is that most HPCs in the world sit with a select few AI mammoths. And they would not want to risk several billions in GPUs and HPCs to sit in areas they have termed as politically and economically uncertain. There has to be a huge effort to change this narrative for the safety of these GPUs and HPCs. Some of the best data centres in the continent with public cloud providers on board are caches and storage. But AI requires proximity to training data. This is a shift that means compute and storage now must be side by side. A unique opportunity for the continent if I ever saw one.

What happens when we have the data centres present to attract HPCs in the continent with better power and cooling? We will attract alongside it, the technical competencies associated with the whole industry allowing for the disbursement of knowledge much closer to home. For us, local access to these HPCs means we can do more Africa-specific research and development.

If AI is the future, the next generation of cities might grow around data centres with hundreds of thousands of GPU clusters and HPCs.  If you build it, they will come. At least that is what happened in the beginning (of time and this article). Maybe this is something worth trying to prove. The potential to organically have cities developed around such infrastructure as well as the potential for knowledge exchange, research and development, job creation and other economic benefits is a huge opportunity for data centres and Africa.