NVIDIA powers 52% of sovereign AI

The latest sovereign AI debate starts with a hardware count. CNAS’s Sovereign AI Index, published May 7, says Nvidia supplies GPUs for 52% of tracked sovereign AI infrastructure projects worldwide, making it the most embedded company in a category built around national control of compute, models and data. That matters because “sovereign AI” is usually sold as a way for governments and regions to reduce dependence on foreign technology providers. But the projects CNAS tracks are still concentrated in infrastructure — data centers, supercomputers, GPU clusters and compute-access programs — where access to advanced chips remains the binding constraint. Infrastructure makes up 59% of all projects in the index, CNAS said. (interactives.cnas.org) Africa offers one of the clearest examples of the tradeoff. Silicon Canals reported on May 22 that the continent’s $60 billion AI sovereignty push, announced at the April 2025 Kigali summit, still runs through roughly 12,000 Nvidia GPUs, Google Cloud credits and Microsoft data-center partnerships. Rest of World separately reported that the fund targets infrastructure, talent and startups, with the 12,000-GPU buildout focused on the “Big Four” countries and Morocco. (interactives.cnas.org) ### If the goal is sovereignty, why is Nvidia so central? Nvidia is central because sovereign AI projects are not just policy exercises; they are compute procurement programs. CNAS says its index tracks infrastructure, models and data, but infrastructure is the largest category, and that means demand for GPU clusters and AI factories. Benzinga, citing the CNAS research, said Nvidia is the supplier in 52% of tracked sovereign AI infrastructure projects. (siliconcanals.com) CNAS has also framed the broader policy issue in explicit terms. In an April 16 commentary, CNAS researchers Ruby Scanlon and Vivek Chilukuri wrote that the U.S. Commerce Department’s American AI Exports Program is designed to deploy the full American AI stack abroad, including infrastructure, tools and services. That suggests many “sovereign” deployments will still be assembled from U.S. components even when ownership or hosting is local. (interactives.cnas.org) ### What does Africa’s plan show about the model? Africa’s plan shows that local ambition and foreign dependency can coexist. Silicon Canals said the Kigali initiative’s largest single hardware line is 12,000 Nvidia GPUs allocated across major African markets and Morocco, alongside Microsoft data-center partnerships and Google Cloud support. (cnas.org) Rest of World described the same structure in similar terms, reporting that Africa’s AI plans still depend on Google, Microsoft, Nvidia and Meta even as governments talk about digital sovereignty. The publication said the $60 billion fund was announced in April 2025 and is meant to support infrastructure, talent and startups. ### Does that make “sovereign AI” a contradiction? CNAS’s own framing suggests the answer is more complicated than a yes-or-no test. (siliconcanals.com) The index tracks government-led projects across infrastructure, models and data, which implies sovereignty is being pursued across layers, not only through chip ownership. In practice, a country can host compute locally, shape access rules and back domestic model development while still buying foreign accelerators. (restofworld.org) That does not remove the dependency. Benzinga’s summary of the CNAS data said Nvidia is “by far the most embedded company” in the emerging sovereign AI ecosystem. The concentration creates obvious exposure to export controls, supply bottlenecks and vendor roadmaps, even when the political branding emphasizes independence. (interactives.cnas.org) ### What should readers watch next? CNAS said more than 130 sovereign AI projects are underway worldwide, with disclosed investment heavily concentrated in the Middle East and East Asia. As those projects move from announcements to procurement and operations, the next useful evidence will be in supplier mix, cloud partnerships and whether countries can diversify beyond a single GPU ecosystem. (interactives.cnas.org) (cdn4.benzinga.com)