Saudi Arabia's AI boom is becoming an electricity and cooling story

The constraint hiding behind every AI headline Every AI announcement out of LEAP 2026 — new cloud regions, expanded compute zones, model partnerships — ultimately runs on the same two physical resources: electricity and cooling. As Saudi Arabia's AI infrastructure ambitions have scaled up, so has the importance of those two unglamorous constraints, to the point where megawatt figures now show up as headline numbers in announcements that, on the surface, are about cloud services or AI models rather than power engineering. This piece connects several of those announcements to trace the power and cooling requirements behind the country's AI build-out. AWS and HUMAIN's AI zone: capacity measured in power, not dollars AWS and HUMAIN's expansion of a dedicated Saudi AI zone to as much as 50 megawatts of capacity, covered in TechScoop's piece on the AI zone's expansion , is a clear example of this dynamic: the headline figure attached to the announcement is a power number, targeted around 2028, rather than a dollar investment figure. That framing reflects a broader shift in how AI infrastructure projects get described, because power draw — not server count or floor space — is what ultimately determines how much AI compute a facility can actually run. MIS and NHC Innovation: two more data points on the same curve MIS's $1.2 billion plan to expand its Saudi data-centre capacity to 192 megawatts, covered in TechScoop's piece on the MIS expansion , and NHC Innovation's $800 million Khuzam Digital Valley project, targeting roughly 65 megawatts of potential capacity by 2033 and covered in TechScoop's report on the Khuzam project , add two more concrete data points to the same underlying pattern. Between the AWS-HUMAIN AI zone, MIS and NHC Innovation alone, Saudi Arabia has multiple operators targeting roughly 300 combined megawatts (about 307 on the announced figures) of AI-relevant data-centre capacity across different timelines — and that tally does not yet include whatever capacity sits behind the new AWS and Microsoft general-purpose cloud regions. On the announced figures, that is 192 megawatts from MIS, roughly 65 from NHC Innovation and up to 50 from the AWS-HUMAIN zone. Why heat is the harder half of the equation Supplying power to a facility is only half the challenge; removing the heat that power generates once it flows through dense racks of AI chips is the other, often harder, half. This is precisely the problem Ferveret's immersion-cooling technology is built to address, and its $150,000 win at LEAP's Rocket Fuel competition, which TechScoop covered in its profile of the company , looks considerably more significant once placed alongside the roughly 300 combined megawatts of AI capacity being planned across MIS, NHC Innovation and the AWS-HUMAIN zone. Every one of those megawatts eventually needs a cooling solution, and traditional air-cooling increasingly struggles to keep pace as rack density rises with modern AI hardware. Land and location matter as much as power and cooling Behind the power and cooling numbers sits a third, less-discussed constraint: physical land in locations with reliable access to power grids, water or cooling infrastructure, and network connectivity all at once. NHC Innovation's choice to frame its project as a dedicated "digital valley" rather than a single facility suggests an approach built around securing and developing that land and connectivity in a coordinated way from the outset, rather than treating each new facility as an isolated site-selection exercise. What this means for how AI announcements should be read Taken together, these five pieces — AWS and HUMAIN's AI zone, MIS's expansion, NHC Innovation's Khuzam project, and Ferveret's cooling technology — describe a Saudi AI build-out in which the physical infrastructure layer sets the pace for the software and model layer that tends to dominate headlines. A new AI model or cloud service can be announced in a press release; hundreds of megawatts of reliable, cooled compute capacity take years of construction to deliver, which is reflected in the 2028 and 2033 targets attached to the AI zone and Khuzam. Readers following the build-out should watch megawatt and cooling-technology announcements as closely as model and partnership news, because physical capacity determines how much of the latter can run. Why electricity supply is its own separate question Building data-centre capacity and cooling it are only two-thirds of the underlying constraint; the electricity itself has to come from somewhere, generated and delivered reliably enough to run compute-intensive AI workloads around the clock. Every megawatt figure attached to a Saudi AI infrastructure project — whether the AWS-HUMAIN zone's 50 megawatts, MIS's 192-megawatt target, or NHC Innovation's roughly 65-megawatt Khuzam plan — represents a claim on the broader electricity grid and generation capacity that has to be planned for well in advance of a facility going live. T