Why megawatts have become the new metric of Saudi Arabia's AI race

The number that shows up in every announcement Read enough Saudi AI infrastructure announcements in 2026 and one pattern becomes impossible to miss: almost every one of them, regardless of which company is behind it, quotes a capacity figure in megawatts. That is not a coincidence or a shared marketing habit — it reflects the fact that power availability, not physical space or even chip supply on its own, is the practical constraint that determines how much AI compute a facility can run, based on how the build-out has been documented through Saudi Arabia's national digital platform ( my.gov.sa ). This is a genuine shift in how infrastructure of this kind is communicated. A decade ago, a data-centre announcement was more likely to emphasise server count, floor space or storage capacity; today, for AI-focused facilities specifically, the megawatt figure has become the headline number, and everything else tends to be described as a secondary detail. That change in emphasis is itself evidence of how central power has become to the AI infrastructure conversation. What a megawatt of AI capacity actually buys A megawatt is a unit of power — one million watts, delivered continuously. For a data centre, that power has to cover both the electricity needed to run the servers and GPUs themselves and the electricity needed to cool them, since the dense clusters of chips used for AI training and inference generate substantial heat that has to be removed continuously to keep the hardware operating safely. The more powerful and densely packed the chips inside a facility, the more electricity — and cooling capacity — that facility needs per rack, which is why AI data centres tend to require significantly more power per square foot than a traditional enterprise data centre built for less compute-intensive workloads. This is also why comparing two AI infrastructure projects purely by their announced megawatt figure, without knowing how efficiently each facility converts that power into usable AI compute, can be misleading: a facility with excellent cooling efficiency can extract more useful AI processing from the same power allocation than one with less efficient cooling, even though both might quote an identical headline capacity number. GPU density is the reason the metric changed The shift toward quoting megawatts, rather than server count or floor space, tracks the shift toward GPU-dense AI computing generally. A facility built to run large AI models needs to pack far more processing power into the same physical footprint than a conventional data centre, and that density is exactly what drives up power and cooling requirements per rack. As a result, a data centre's usable AI capacity is now more accurately described by how much power it can deliver and dissipate as heat than by how many square metres of floor space it has — which is precisely why every serious announcement about Saudi Arabia's AI build-out, whether from a global cloud provider or a local infrastructure operator, leads with a megawatt or gigawatt figure. That density trend shows no sign of reversing. Successive generations of AI-focused chips have generally increased processing power at a pace that has outstripped comparable improvements in power efficiency, which means the power and cooling requirements per rack for a state-of-the-art AI facility have tended to rise over time even as the chips themselves become more capable per watt — a dynamic that keeps megawatts, rather than any other single metric, at the centre of how this infrastructure gets sized and communicated. Cooling as the hidden half of the story Because so much of a data centre's power draw ends up being used for cooling rather than computing directly, the efficiency of a facility's cooling system has become as strategically important as its raw power capacity. A facility that can cool its GPU clusters more efficiently can support more usable AI compute within the same power allocation than one relying on less efficient cooling methods, which is why cooling technology has become its own competitive and innovation battleground within the broader AI infrastructure race, rather than a background engineering detail. In hotter climates specifically, cooling efficiency carries even more weight than it might elsewhere, since a facility has to work harder to dissipate the same amount of heat when ambient outside temperatures are already high. That makes cooling technology and design choices a particularly consequential variable for AI data-centre operators building in Saudi Arabia's climate, even though the specific cooling approaches chosen by individual projects have not been detailed in the announcements reviewed here. What this means for anyone evaluating a project For journalists, investors and enterprise buyers trying to judge the real significance of any individual Saudi AI infrastructure announcement, the practical takeaway from this shift in metric is straightforward: ask for the megawatt