Digital Warlords: Private Authority Over AI

A handful of firms now dominate the technologies that support advanced AI. I call them digital warlords because they operate outside the boundaries of domestic regulation and multilateral oversight, setting the terms for who may access advanced AI technologies and under what conditions. Their power rests on their expertise and control over critical inputs such as advanced chipmaking (Nvidia, TSMC, ASML),Footnote ⁹ compute (Azure, Google Cloud, AWS, Alibaba Cloud), and frontier model development (OpenAI, Anthropic, DeepSeek). Unlike traditional multinationals, whose influence derives from lobbying, employment, or production, digital warlords exercise authority primarily through infrastructural control.Footnote ¹⁰ This authority makes them gatekeepers of the AI economy.

Like historical warlords, they thrive where institutional authority is weak. Their competition centers on scarce resources like chips, talent, energy, capital, favorable policy, and market access. They do not govern populations by force, but their authority over strategic technologies gives them a warlord-like position in the digital economy, which states must confront. As with historical warlords, states may attempt to contain or co-opt their authority, but they remain independent centers of power that governments cannot fully control.

Their authority is embedded in the value chains that shape the digital economy. I refer to this as infrastructural AI power: the capacity of private firms to govern access to AI systems through control over their technical foundations.Footnote ¹¹ For instance, Nvidia supplies 92 percent of data center GPUs,Footnote ¹² and TSMC produces 65 percent of the world’s foundry output.Footnote ¹³ Their platforms set standards and determine access. Nvidia’s Compute Unified Device Architecture (CUDA) is a proprietary programming interface. It has become the de facto industry standard for AI development, a unilateral rule-making power that entrenches Nvidia’s dominance. OpenAI similarly restricts access to its most advanced systems through usage and licensing terms. Like CUDA, these are privately enforced rules with global reach. Through these rules, firms can determine the terms of participation in the AI economy. Digital warlords exercise this infrastructural control largely independent of sovereign oversight.

Their infrastructural power grants them geopolitical significance. States depend on advanced AI for strategic purposes like military applications and to expand state capacity (for example, digital public infrastructure such as India Stack),Footnote ¹⁴ but private firms control its development. These private actors perform consequential functions, from moderating platform content to controlling global connectivity through proprietary submarine cables and cloud infrastructure. In this new era, leading firms operate as geopolitical actors in their own right, shaping not only markets but also the strategic decisions of states.Footnote ¹⁵ Given this leverage, these digital warlords engage in transactional cooperation, granting access to advanced systems in return for essential resources such as low cost energy or data center investments. The core externality is that this authority lies outside state control, creating political, security, and economic vulnerabilities. Social media giants shape the information environment in ways that states increasingly view as politically consequential. Governments now frame such risks in national security terms, as illustrated by US legislation banning TikTok or China’s “Delete America” campaign.Footnote ¹⁶ Dual-use AI technologies raise further security concerns, as commercial models may be repurposed for drone applications or cyber warfare.Footnote ¹⁷ Economically, digital firms shift profits across jurisdictions by exploiting the mobility of intangible assets, eroding national tax bases.Footnote ¹⁸ These fault lines emerge from the imbalance between digital warlords’ capacity to innovate and states’ capacity to restrain.

These externalities are politically salient because they expose institutional limitations—states cannot fully internalize them.Footnote ¹⁹ Regulation is inadequate, and prior social compacts inapplicable. Embedded liberalism, as originally conceived, assumed that governments could compensate those who lost out from globalization through redistribution and domestic regulation.Footnote ²⁰ That model presumes globalization’s economic harms occur within national borders. Redistributive measures, like trade adjustment assistance, could at least in principle address labor market shocks.Footnote ²¹ But the fallout from digital warlords escapes these tools. Redistribution is no remedy when the threat is to sovereign authority itself.

The Limits of Multilateral Trade Rules in the Digital Economy

The liberal economic order is inadequate to govern the digital economy for two reasons. First, it lacks mechanisms to constrain the infrastructural power of digital warlords. Second, it struggles to discipline the sovereignty strategies that states adopt in response to the growing capabilities and externalities of these firms.

The multilateral trade regime was designed to govern the flow of goods and, later, traditional services. It has been slow to address digital services, and has not evolved to discipline AI infrastructure. Its rules rely on outdated categories, such as the distinction between goods and services or the modes of service supply, that are increasingly incoherent in the digital context.Footnote ²² Take robotics: AI systems like large language models and image recognition tools are now embedded in humanoid robots that provide services. Yet these machines are also physical goods. Such hybrid forms expose the limitations of the regime’s path dependent rules, designed for an analog economy.

To be sure, some WTO instruments cover aspects of the digital economy. The General Agreement on Trade in Services (GATS) applies to cross-border digital services when commitments have been scheduled.Footnote ²³ TRIPS protects the intellectual property of AI firms. And the Joint Statement Initiative on E-commerce (JSI), launched in 2017, seeks to liberalize digital trade through provisions on data localization and source code disclosure.Footnote ²⁴ Yet most of these rules serve to empower digital warlords rather than constrain them. The limited attempts to protect state regulatory sovereignty, such as exceptions for privacy and security, remain legally and politically contested.

The global tax regime reveals a similar institutional failure. Rooted in the early twentieth-century economy, the system was designed to tax profits based on the location of physical production. But digital firms generate income from intangible assets, often without any physical presence. Firms further exploit these gaps by shifting profits to low-tax jurisdictions, eroding national tax bases.Footnote ²⁵ With OECD-led tax reforms stalled, states have pursued unilateral digital services taxes,Footnote ²⁶ another sovereignty measure that reflects the inability of multilateral institutions to address digital externalities.

The rise of digital services taxes underscores the second deficiency of the liberal international order: it lacks effective tools to discipline how states respond to firms’ infrastructural power. Additional policies such as data localization mandates, plat- form bans, and semiconductor export controls fall largely outside WTO disciplines or are shielded by national security exceptions. Even when trade agreements do apply, enforcement is weak and tariff-based retaliation mechanisms are poorly suited to digital restrictions.

In sum, international institutions lack the capacity to constrain either the activities of digital firms or the sovereignty strategies their capabilities provoke. The liberal trade regime was built to discipline states. But as firm power increasingly threatens state authority, governments are reasserting control through ad hoc and at times illiberal measures. In the absence of effective global rules to manage either firm power or state responses, strategic digital sovereignty has emerged as the dominant mode of digital governance.

Strategic Digital Sovereignty and Techno-Blocs

Without effective multilateral rules, states are devising their own strategies to control the AI value chain. They pursue what I call strategic digital sovereignty: the exercise of state authority over core digital infrastructure through unilateral actions or selective alliances with firms and other governments. It is strategic in two respects. First, it is pragmatic: rather than formal institutions rooted in shared liberal values, states rely on flexible partnerships to secure or influence access to essential infrastructure. Second, it is geopolitical: sovereignty is projected outward to manage dependencies and externalities that cross borders.

States pursue strategic digital sovereignty through two nonmutually exclusive modes: unilateral measures and techno-bloc formation. Unilateral measures—such as export controls, data localization mandates, digital services taxes, and platform bans—aim to reduce strategic vulnerabilities and constrain firm power. Yet unilateralism cannot overcome material constraints. Because compute, talent, energy, and capital are globally dispersed, even powerful states must rely on partnerships to build AI capacity and address strategic vulnerabilities.

To that end, states form techno-blocs, aligning with firms and other governments to overcome these constraints and extend authority over private AI infrastructure. Within blocs, states contribute critical inputs and relax barriers in exchange for access to markets and advanced technologies. Techno-blocs differ from traditional state alliances in that they embed dominant AI firms directly into governance arrangements, turning critical AI inputs into leverage. They manage the externalities of infrastructural power by constraining it or co-opting it. Blocs themselves are not mutually exclusive: states may participate in several at once, perhaps joining one centered on compute partnerships while also joining another organized around energy cooperation or regulatory harmonization.

These arrangements are broadly beneficial to their participants. For firms, blocs deliver market access, resources, and regulatory stability, reinforcing their dominance at the frontier of AI. For policymakers, they are instruments to reassert sovereignty where unilateral measures fall short, demonstrating competence and securing access to advanced technologies in ways that may enhance political survival.

Although sovereignty is often framed as unilateral control, techno-blocs reflect a cooperative form. States selectively align with others to establish authority over AI where unilateral measures are inadequate.Footnote ²⁷ Techno-blocs align incentives and coordinate governance across jurisdictions, addressing vulnerabilities that unilateral tools alone cannot overcome.

States combine unilateral and bloc strategies in different ways. One prominent variant is AI mercantilism, which uses trade policy to secure national control over the AI stack. The United States and China both pursue this approach, combining export controls with industrial policies to build domestic capacity while constraining rivals.Footnote ²⁸ It reflects a mercantilist logic: securing control over critical AI inputs while excluding others to preserve national advantage.

In contrast, smaller states like the UAE and Saudi Arabia pursue alignment strategies, offering critical inputs such as capital, energy, or regulatory flexibility in exchange for integration into techno-blocs. A third variant is advanced by the European Union, which pursues regulatory sovereignty, using EU market access to globalize its legal standards for data and AI. The remainder of this section examines in greater detail how states implement these strategies in practice.

The United States blends unilateral action with techno-blocs in a form of AI mercantilism that constrains adversaries and secures technological access. In October 2022, the Biden administration imposed sweeping controls to restrict China’s access to advanced AI technologies, including limits on high-end chip exports and US involvement in Chinese AI development. These controls expanded to other countries in 2023 and 2024, culminating in the AI Diffusion Rule, which established a tiered framework for licensing requirements on exports of advanced chips and cloud services. The rule classified countries based on their alignment with US national security interests, imposing stricter export controls on those in lower tiers. Countries such as Saudi Arabia and the UAE were placed in restrictive tiers, limiting their access to high-end AI hardware and services.Footnote ²⁹ While formally unilateral, these measures were coordinated with allies such as the Netherlands and Japan, embedding export controls within a broader techno-bloc.Footnote ³⁰

The recent Gulf partnerships with the United States exemplify bloc-based ap- proaches by smaller states. In return for access to leading Nvidia chips, Emirati authorities agreed to adopt US technology security standards, while offering to fi- nance infrastructure and supply low cost energy for AI systems.Footnote ³¹ The deal includes a large compute campus in Abu Dhabi, operated by US cloud providers and the Emirati firm G42.Footnote ³² A parallel alignment is underway in Saudi Arabia. A consortium of US firms (Google, Oracle, Salesforce, AMD, Uber) and Saudi-backed DataVolt announced USD 80 billion in joint AI projects, including $20 billion for US-based data centers.Footnote ³³ Nvidia also agreed to supply chips to Humain, a Saudi firm, under terms involving chip access and joint infrastructure investments.Footnote ³⁴

China’s model of strategic digital sovereignty combines unilateral control at home with techno-bloc formation abroad. Domestically, Beijing asserts authority over private technology firms through mandated alignment with national priorities.

Chinese authorities use laws governing data storage and cross-border flowsFootnote ³⁵ to scrutinize the activities of foreign and domestic firms when their influence threatens state control over the information environment.Footnote ³⁶ Together, these efforts assert state authority over AI infrastructure.

Like the United States, China pursues strategic digital sovereignty abroad through AI mercantilism, treating advanced technologies as strategic assets. The Digital Silk Road bundles cloud services, fiber-optic infrastructure, surveillance platforms, and e-government tools into Belt and Road Initiative projects.Footnote ³⁷ It extends beyond physical infrastructure to cloud networks. A recent study identifies ninety-seven Chinese cloud zones in thirty-seven partner countries, especially where trade ties are strong or states participate with China in regional arrangements such as the Shanghai Cooperation Organization.Footnote ³⁸ Chinese technology exports often bundle digital governance tools such as biometric ID systems with cloud services, embedding Chinese surveillance technologies in partner states. For example, Huawei’s Safe City project in Islamabad integrates facial recognition and real time suspect tracking into local security systems.Footnote ³⁹ China’s digital infrastructure exports deepen partner dependence while advancing its model of digital authoritarianism.Footnote ⁴⁰

Rather than AI mercantilism, the European Union anchors its techno-bloc in regulatory sovereignty, using its market size to globalize EU standards for digital governance. It establishes influence through legal frameworks such as the General Data Protection Regulation (GDPR), the Digital Markets Act (DMA), and the AI Act.Footnote ⁴¹ This approach reinforces norms of consumer protection but relies on extraterritorial regulation rather than multilateral institutions. The AI Act, for example, applies to any company offering AI systems to EU users, regardless of location. This strategy trades access for compliance within a regulatory techno-bloc anchored in European norms.Footnote ⁴² With few dominant firms in the AI stack, the EU converts market access into authority by embedding firms in its regulatory bloc through legal conditionality.

Within this framework, some member states pursue distinct techno-blocs. France, for instance, extends the EU’s model by promoting an “AI Third Way,” a sovereign alternative to both US and Chinese dominance in AI. Its strategy combines domestic investment in data centers (jointly financed by the UAE investment fund MGX) with the development of domestic platforms like Mistral Compute.Footnote ⁴³ Reflecting its position in the AI stack, this approach pursues unilateral AI development within the EU techno-bloc. Likewise, Germany is developing a sovereign AI system for public services, built in partnership with OpenAI but hosted in the cloud of SAP, a German firm.Footnote ⁴⁴ This approach draws on OpenAI’s capabilities while complying with national data security and regulatory requirements.

In sum, states are fragmenting the digital order, producing rival centers of technological authority. Digital sovereignty strategies reflect states’ geopolitical alignments, institutions, and positions in the AI stack. Each strategy, however, seeks to manage vulnerabilities created by private firms’ control of key inputs.

Implications for the Liberal International Order

The pursuit of strategic digital sovereignty is reshaping the liberal international order. Techno-blocs often embed digital warlords in governance, elevating them from subjects of regulation to strategic partners. The result is illiberal alignment: opaque rather than transparent, transactional rather than rules based, exclusive rather than universal. Some sovereignty initiatives still pursue liberal aims like protecting privacy or banning high-risk AI applications. The EU exemplifies this approach, using legal frameworks to discipline firms. Yet this remains an exception, and may come at the cost of limiting the capacity to develop advanced systems.

The consequences of techno-bloc competition go beyond the splintering of the digital order. Techno-blocs create a hierarchy in which a few states and firms control the AI frontier, while others are relegated to subordinate roles with more limited sovereignty.Footnote ⁴⁵ These arrangements extend beyond technical dependence. In some ways, they resemble extractive institutions of the past. In what has been called digital colonialism,Footnote ⁴⁶ dominant firms extract and monetize data, while contributing little in tax revenue or employment. This could lead to rising inequality across states as digital warlords consolidate their economic dominance.

Such outcomes raise the question of whether AI will follow the trajectory of earlier technological shocks, or whether its features make it a distinct challenge.

Previous infrastructural advances, such as electricity or telecom, were eventually brought under international regimes or subject to regulated monopolies. Like those earlier cases, AI is a general purpose technology. But its combination of features makes it harder for states to contain. AI innovation has outpaced the capacity of institutions to adapt. The mobility of intangible assets and AI-enabled digital services complicate the establishment of rules for trade and taxation. And its dual use features combine economic and security capabilities in ways rivaled perhaps only by nuclear technology. AI thus represents a technological exception, with features that make multilateral solutions especially difficult to achieve.

Crucially for international cooperation, AI also represents a systemic catalyst. It is accelerating a fragmentation of the international order in which authority over technological infrastructure becomes a central axis of division.

Despite growing fragmentation, alternative approaches to digital governance are emerging. Scholars have advocated models that disperse authority across multiple levels of governance, rather than within hierarchical blocs,Footnote ⁴⁷ building on broader insights from the institutional complexity literature.Footnote ⁴⁸ And international institutions themselves are developing models of transparency and regulatory convergence. For instance, the WTO promotes early notification of AI regulations to manage cross- border frictions, and a new emerging technologies unit builds negotiator capacity.Footnote ⁴⁹ Other initiatives, such as the US–EU Trade and Technology CouncilFootnote ⁵⁰ and OECD work on multinational tax reform, remain limited but show that international coordination has not disappeared entirely.

For international organizations to play a meaningful role, however, they would need to evolve beyond their original design. Existing trade and tax institutions were built to discipline states, not firms that control AI infrastructure. Constraining digital warlords would require new mandates: oversight of data flows and privacy, transnational competition rules, and mechanisms to tax digital revenues at their source.Footnote ⁵¹ The initiatives noted before point in this direction but lack the necessary authority and scope.

The stakes are profound. As states prioritize control over core digital technolo- gies, the liberal international order risks becoming a relic of the analog era. Unless institutions evolve to constrain infrastructural power and coordinate governance, unilateral barriers and techno-blocs will entrench illiberal fragmentation. AI governance is emerging as a central axis of international politics, accelerating the shift from cooperation under rules to competition over the world’s most consequential technologies.