HUMAIN AI Cloud for buyers: 1.1 megawatts live, above 90 percent full, and no published price

HUMAIN launched an AI cloud in Saudi Arabia at LEAP 2026, running NVIDIA HGX B300 Blackwell Ultra systems, with 1.1 megawatts of operational capacity and reported utilisation above 90 percent. This is an assessment for buyers, from published information. TechScoop has not used the platform, has not been given access, and makes no claim about performance, availability or support quality. Start with the two disclosed numbers, because they are unusually candid for this market and they say different things. The live capacity figure means this is a real service rather than an announcement. Most regional AI infrastructure news gives a target and a date with nothing switched on behind it. Something is running here and customers are using it. The utilisation figure means it is nearly full. Above 90 percent on a platform that has just launched tells a buyer that demand exceeded supply before the service opened, and it raises availability as the first commercial question rather than price. A buyer approaching a cloud running at that level should assume capacity is allocated rather than purchased on demand, and should ask about lead time before asking about anything else. The scale needs to be understood plainly. At 1.1 megawatts this is a single-room deployment, in the order of a few racks of Blackwell Ultra systems, not a campus. HGX B300 is NVIDIA's eight-GPU Blackwell Ultra platform aimed at large-scale training and high-throughput inference, and each system draws substantial power, which is why a megawatt figure at this scale corresponds to a modest node count. HUMAIN says it plans to reach 35.1 megawatts by 2027, more than thirty times what is live. The second thing a buyer must establish is which silicon they are buying. HUMAIN is running two estates in parallel. Alongside the NVIDIA platform, AMD, Cisco and HUMAIN announced on the opening day of LEAP that AMD Instinct MI355X systems on Cisco networking had gone live and were serving HUMAIN customers, with a reported plan to deploy 13,000 AMD GPUs this year and around 35 megawatts of operational capacity by mid-2027, on a joint path to as much as one gigawatt by 2030. Reporting has named the Ministry of Commerce among early customers. Running two accelerator families is a deliberate and defensible strategy for a national AI provider, because it gives supply resilience and negotiating leverage at a time when frontier GPU allocation is the binding constraint everywhere. For a buyer it introduces a real question. A workload built against CUDA does not move to Instinct hardware for free, and while ROCm has matured considerably the porting effort is not zero. Anyone signing should establish which estate their capacity sits on, whether they can be moved, what the notice period would be, and how pricing differs between the two. On data controls, the sovereignty argument is the reason many Saudi buyers will consider this platform at all, and it deserves precision rather than acceptance. Locally hosted infrastructure operated by a domestic company under Saudi law is a genuinely different proposition from a foreign hyperscaler region, and for some regulated workloads it is the only acceptable one. What has not been published is the contractual detail that makes it assessable: where data is processed and stored, who holds administrative access, whether foreign nationals or foreign entities have any operational access path, what happens to data on termination, and whether disconnected or restricted-network deployment is available. Those are answerable questions and none of them is answered in the launch material. The model question is similarly open. HUMAIN has released M3, an open-weights Arabic model, and has partnerships spanning Microsoft, AWS, Mistral and Reflection AI, but no published list establishes which models are served on this cloud, at what versions, with what licensing, or whether customers can bring their own weights. What is not published, in summary and in the order a buyer needs it: price, instance types and configurations, minimum commitment, service level commitment, current lead time for new customers, the supported model list, and the contractual data-control terms. On a platform above 90 percent utilisation, the absence of a published service level commitment is the most significant of those. Where this platform makes sense today is for Saudi government entities and regulated enterprises whose requirement is in-country processing under domestic law, for whom the alternative is not a cheaper cloud but no compliant option at all. It makes less sense for a price-sensitive workload with no residency requirement, where global providers with spare capacity will be both cheaper and immediately available. The questions to put to HUMAIN before committing. What is the lead time for capacity today. Which accelerator family, and can I be moved. What is the SLA and what is the remedy. Who has administrative access to my environment, and can that be contractually