International AI governance is entering a new phase as geopolitical competition shifts attention from regulating AI systems to securing the infrastructure that makes frontier AI possible. From semiconductors and compute to energy, talent and defence integration, the emerging contest over AI ecosystems is reshaping what responsible AI governance means.
“We may be the last generation able to set the terms on which humanity and machines coexist. The door is still open. But it will not stay open long.” With this striking warning, UN Secretary-General António Guterres opened the first Global Dialogue on Artificial Intelligence Governance, attended by all 193 member states of the UN. The conversations at the conference largely focused on the normative principles that have been driving AI governance frameworks since the 2023 Bletchley Park AI Safety Summit. These principles include safe and trustworthy AI, interoperability, accountability, transparency, human rights, and international cooperation on AI. Therefore, the conference reflected a governance paradigm which primarily views AI as a technology whose societal impacts require regulation. Meanwhile, outside of these multilateral discussions, the focus of the world’s leading AI powers has drastically shifted.
When Governance Weighs Against Commercial Interests
Global powers are investing billions to govern and control something much larger than AI models. Their concern is no longer focused on regulating how AI behaves but rather on controlling the ecosystem that makes advanced AI possible. With increasing calls for AI sovereignty, states are more invested in securing semiconductors, electricity generation, hyperscale data centres, cloud infrastructure, talent, and, most importantly, the military integration of AI. Military integration of AI matters the most because AI is now seen as a decisive edge in military capability with faster targeting, autonomous systems, and intelligence processing. States that could successfully achieve military integration of AI would get an advantage that competitors can’t quickly match. Therefore, investments and regulation of the AI ecosystem have become a critical priority of global powers to secure capabilities.
The priorities of leading AI companies have also changed. A few years ago, AI businesses used to distance themselves from military applications and government contracts publicly. In 2018, Google declined a Pentagon contract, called Project Maven, after thousands of Google employees protested over concerns about the use of Google’s technology for warfare. Today, these companies are competing to secure government contracts and are investing heavily to gain the lead in military applications of AI.
Just a few weeks before the Global Dialogue on AI Governance, the Pentagon announced the finalisation of its agreement with eight leading AI companies, including OpenAI, SpaceX, Nvidia, Reflection, Google, Oracle, Microsoft, and Amazon Web Services. According to the Pentagon, these companies have agreed to the deployment of their technologies by the US military for any “lawful operational use.” Anthropic, however, objected to the inclusion of the lawful use clause in the contract, and was, therefore, excluded from this group by the US government. According to the company’s position, it was concerned about the use of its technology for mass domestic surveillance and fully autonomous lethal weapons. The US government also designated Anthropic as a ‘supply-chain risk’ and ordered the federal agencies to phase out its tools. This episode suggests that military utility of AI can take precedence over some concerns for ethics, safety and accountability. More importantly, the Anthropic episode illustrates that normative AI governance is no longer sufficient and ethics can be negotiated when commercial interests are involved. Therefore, those who are willing to uphold these principles risk sacrificing market competitiveness. However, the US is not the only global power that is writing this playbook.
Competitiveness over Caution
The European Union’s AI Act became the world’s first comprehensive and legally binding regulatory framework on AI. It positioned the EU as the global standard for committing to develop trustworthy and human-centric AI. However, with the adoption of the European Commission’s Digital Omnibus, several high-risk obligations and compliance requirements given in the EU AI Act have now been delayed. This move came in response to calls by European industry leaders and policymakers to reduce regulation and enhance Europe’s competitiveness, especially in the face of the accelerating AI race. Consequently, while the core architecture of the AI Act remains intact, the urgency to implement strict compliance with ethical AI development has certainly been pushed back. This indicates that even the world’s leading voice for ethical AI, is now trading some of that caution for competitiveness and if EU is willing to do so, few others are likely to hold the line. This pattern is becoming even more evident in other AI hubs across the world.
Saudi Arabia’s Public Investment Fund has backed state-owned AI company, Humain, which aims to achieve $77 billion in infrastructure and 1.9 gigawatts of data centres by 2030. Similarly, the UAE’s Stargate initiative is targeting a sovereign computational capacity of five gigawatts. Emirati entities have also invested around $148 billion in the AI industry at home and abroad. India is planning to scale up its public GPU stock from the existing 38,000 units to 100,000 units by the end of 2026. France has focused its AI strategy on energy and chip supply. France has also launched a €30-50 billion data-centre venture jointly with the UAE.
When viewed together, these initiatives indicate a common pursuit of building sovereign AI capabilities by regulating chip supply, energy, data infrastructure and talent. Ethical safeguards for ensuring safe and responsible AI development seem to have been eclipsed by this urgency to achieve technological sovereignty and competitiveness. Almost all of these states and firms still endorse global declarations and regularly participate in multilateral conferences and summits on AI. However, they invest their resources and efforts in regulating the AI ecosystems more than in regulating the AI systems themselves. This trend poses an acute governance challenge for the countries in the Global South that have limited capacity in AI development.
The Capacity Gap
The official consultation results from the UN Global Dialogue on AI governance show that the governments ranked capacity-building as their top priority. AI development capacity remains limited across much of the world. As of 2025, over 90 per cent of AI-specialised compute capacity is concentrated in only two countries: the US and China. As per the 2026 Africa Data Centres Association (ADCA) Economic Report, the data centre footprint of Africa remains below 1 per cent of the global capacity, despite the fact that the rate of digital adoption remains among the fastest in the world. Similarly, limited sovereign compute capacity means many countries in the Global South remain reliant on foreign cloud providers. On top of that, the AI talent trained in these countries ends up migrating to a handful of states and firms that offer market competitiveness. As a result, many Global South countries have limited ability to shape the development and governance of AI systems that are largely developed and controlled elsewhere.
The shift in the priority of global AI governance does not mean that the responsible AI agenda should be abandoned. If anything, that agenda has become even more critical now than ever. The harms that the global AI governance sought to address, such as biased algorithms, weak content moderation, and mass domestic surveillance, have not disappeared. In fact, these risks have further compounded as AI becomes increasingly embedded in military and governmental decision-making processes. Therefore, ethics and accountability frameworks must remain the guiding principles on how AI gets developed, deployed and distributed around the world. Yet these principles can only hold weight if they extend to the AI ecosystem itself, not to AI systems alone.
Muhammad Faizan Fakhar is a researcher, and academic specialising in strategic affairs, emerging technologies, and public policy. His academic contributions encompass peer-reviewed journal articles, book chapters, research reports and opinion pieces on South Asian security dynamics and societal impacts of emerging technologies. With Master’s in Strategic Studies and Bachelor’s in Electrical Engineering, he brings a multidisciplinary approach to analysing complex issues at the intersection of technology, security, and society.
This article is published under Creative Commons License and may be republished with attribution.
