Four companies are competing to train Saudi Arabia's AI workforce, and they are selling different things

Alongside the infrastructure announcements at LEAP 2026 sat a quieter set about training, and they are worth reading together because they describe a competition that will outlast most of the hardware. The Ministry of Communications and Information Technology and the Saudi Digital Academy launched a Gemini initiative with Google to give one million Saudi university students access to advanced AI tools. An Amazon Academy launched with MCIT aims to train more than 100,000 Saudis in areas including cloud, AI, logistics and retail. An AI academy developed with WeCloudData is to train educators, university faculty, startup founders and tens of thousands of online learners. Adobe's programme with MCIT and HUMAIN puts creative AI tools in front of more than 27 million citizens and residents aged 13 and over, with a startup track through the Center of Digital Entrepreneurship and The Garage from early 2027. Underneath the vendor programmes sits a substantial state layer that predates them. The SAMAI initiative reported in November 2025 that it had passed its target of training more than one million Saudi citizens in AI skills, providing foundational fluency for civil servants and the general workforce. Tuwaiq Academy declared 2026 its Year of Artificial Intelligence and launched camps and programmes at its Riyadh headquarters from April, covering programming, AI, cloud, cybersecurity, user experience, data science and other areas, aimed at students, graduates and professionals. Read as skills policy, this is a country moving quickly and from more than one direction at once. Read as commercial strategy, it is four or five companies competing for something more durable than a contract. Training is the longest-lasting form of vendor lock-in that exists, and it is the cheapest to deploy. A cloud contract is renegotiated every few years and can be moved. An engineer who learned to build on a particular platform carries that preference through every job they hold for a decade, and when they become the person specifying systems, they specify what they know. Every major infrastructure vendor understands this, which is why certification programmes are subsidised and why the ones aimed at students are free. The targeting confirms the logic. One million university students for Gemini. Eligibility from age 13 for Adobe. Educators and university faculty in the WeCloudData programme, which reaches students indirectly and at scale. These are not workforce interventions aimed at immediate productivity. They are aimed at people who will make technology decisions between 2030 and 2045. The programmes are also not substitutes for each other, and it is worth separating what each actually teaches. Foundational AI literacy, which is what SAMAI provides at population scale, means understanding what these systems do, where they fail and how to use them responsibly. That is a public good and it is vendor-neutral. Tool proficiency, which is what most vendor academies deliver, means fluency in a specific product. Engineering capability, the ability to build, fine-tune, evaluate and operate AI systems, is the scarcest and the least addressed by anything announced. Almost every programme in the list sits in the first two categories. That gap matters against the infrastructure being built. A country standing up hundreds of megawatts of AI compute needs people who can operate it, not only people who can prompt on top of it. The engineers who run large training clusters, tune inference serving, manage data pipelines at scale and evaluate model behaviour are a small population globally and are not produced by short courses. The Lenovo model, where the first cohort of Saudi engineers completed specialised training inside the company's global manufacturing facilities, is closer to what deep capability transfer looks like, and it is slow and small by design. The counterpoint is fair. Literacy at population scale has value independent of depth, and a workforce that broadly understands what AI can and cannot do makes better decisions across every sector, not just in technology. A million people with foundational fluency is a genuine national asset even if none of them ever trains a model. And vendor programmes deliver real skills at a speed and cost no state academy could match alone, whatever the commercial motive behind them. Both are true. The honest summary is that Saudi Arabia has moved decisively on breadth and has not yet demonstrated the same movement on depth, and that the vendors filling the breadth are acquiring a preference position they intend to keep. The figures that would settle it are not published. Completion rates rather than enrolment numbers for any of these programmes. The number of Saudis working in roles that require building rather than using AI systems. And retention, since capability that emigrates is not national capability. Those are the numbers to ask for when the next training announcement arrives.