AI governance shifts from model behavior to securing systems: military priority disables safety brakes

The UN Secretary-General, Antonio Guterres, opened the first global dialogue on AI governance with representatives of 193 member states, warning that the current generation may be the last able to set the terms for coexistence between humans and machines before this window closes quickly. Multi-stakeholder discussions inside the conference hall focused on the normative principles established by the 2023 Bletchley Park summit on AI safety, foremost among them safety, reliability, accountability, transparency, human rights, and interoperability. Yet this diplomatic rhetoric is entirely detached from the reality of field-level shifts; the focus of the major tech powers has shifted fundamentally from regulating software model behavior and its societal impacts to dominating the physical infrastructure systems that make advanced AI model operation possible.

The global governance compass today is moving toward tightening control over semiconductor supply chains, electricity generation plants, hyperscale data centers, cloud infrastructure, talent attraction, and up to military integration, which has become a top priority. Superiority is no longer measured by how disciplined a model’s outputs are, but by the ability to employ technology to provide a decisive advantage in defense domains through ultra-fast targeting, autonomous systems, and real-time intelligence analysis. This trajectory reflects a comprehensive shift in the policies of Silicon Valley giants; after Google rejected renewing the “Maven” project contract with the Pentagon in 2018 following massive employee protests against using technology for wars, the U.S. Department of Defense recently concluded final agreements with eight leading companies, including OpenAI, SpaceX, Nvidia, Reflection, Google, Oracle, Microsoft and Amazon Web Services, authorising the military to deploy their technologies for any lawful operational use.

Geopolitical and commercial competition has proven its ability to displace safety brakes and ethical considerations in favor of protecting market positioning.The reality became clear when the US administration excluded Anthropic from that system because it refused to add a legal-use clause, citing concerns that its models could be used for local mass surveillance and the development of fully autonomous lethal weapons; the company then faced a government classification as a risk to supply chains, with federal agencies directed to gradually phase out its tools. In Europe, the strict hedging did not last long; although the European AI Act was the first globally binding legislative framework, the European Commission adopted amendments through the comprehensive Digital Services Act package to postpone several obligations and compliance requirements imposed on high-risk systems, responding to industry and policy-maker demands to boost competitiveness amid a sharp global acceleration.

This race manifests as a massive competition to build sovereign capabilities, with Saudi Arabia’s Public Investment Fund supporting the state-owned company Human to develop infrastructure worth $77 billion and data centers with a capacity of 1.9 GW by 2030. The UAE’s Stargate initiative aims for sovereign computing capacity of five gigawatts, backed by Emirati investments of about $148 billion in the AI sector locally and internationally, and a partnership with France for data centers with investments ranging from €30 billion to €50 billion. In the same context, India is expanding its government stock of graphics processors from 38 thousand units to 100 thousand units by the end of 2026. These moves reveal that more than 90 percent of specialized advanced computing capacity worldwide will be in the hands of the United States and China by 2025, while the 2026 African Data Centers Association report showed the continent’s share did not exceed one percent of global capacity despite rapid digital adoption, cementing a capability gap and increasing Southern countries’ reliance on foreign cloud platforms and draining their qualified human talent.

This fundamental shift toward infrastructure governance is reshaping the operational and strategic priorities of tech companies and decision makers in the region.In the Gulf, the shift toward massive data centers and electricity grids dedicated to computing reshapes spending priorities; if they lead a technical team or manage an enterprise architecture, they must align their strategies with local sovereign clouds that respect energy consumption limits and provide licensed processors rather than relying indiscriminately on external API calls. In Egypt and the Levant, where access to ultra-high-performance infrastructure is scarce and the cost burden of cloud inference is rising, the real challenge pushes developers and companies to focus on compact lightweight models and open-source architectures, while striving to keep engineering talent from migrating to generously funded regional and international centers. As strict global regulatory standards relax in favor of competitive and operational priority, digital sovereignty becomes more dependent on owning physical computing resources locally than merely signing theoretical security declarations.