The Illusion of AI Sovereignty: Washington and Beijing Still Pull the Strings

Introduction

In an era where digital data has become the new oil and AI the refinery, governments across the globe are waking up to a critical question: Who controls the intelligence that shapes their citizens’ future? As of 2024, 137 countries have enacted data protection laws, according to the Cloud Security Alliance. This signals rising concerns around digital sovereignty, national security, and citizen privacy in an increasingly globalized world. But what began as a regulatory response to safeguard sensitive data is now evolving into a broader and more ambitious mandate: the pursuit of sovereign AI.

Across the globe, governments are no longer content with being passive consumers of foreign-built AI systems. They are investing heavily to build their own AI infrastructure, foundational models, and regulatory frameworks that align with their cultural norms, linguistic diversity, national priorities, and geopolitical values. Why? Because a sovereign state cannot afford to build its intelligence on AI models that are blind to local realities. Will an algorithm built in and for Boston respect the data-sharing taboos of Bedouin cultures or understand the consent expectations in South Korea?

As a result, a new wave of digital nationalism is sweeping through both public policy and enterprise boardrooms. Countries including France, Saudi Arabia, the UAE, Singapore, and India are doubling down on sovereign AI. They are investing in localized data ecosystems, region-specific LLMs, national AI clouds, and stricter compliance regimes. The rhetoric of autonomy is powerful, even necessary. But it also raises a pressing question: Is sovereign AI a real path to digital self-reliance or just an illusion of control in a world still governed by a handful of global AI superpowers?

The Strategic Appeal of Sovereign AI

Until recently, sovereign AI was a luxury—an ambition confined to nations with deep pockets, state-backed funds, or military-industrial muscle. For most countries, especially those in the Global South, relying on foreign AI models was not just a convenience but a necessity. The turning point came in January 2025, when China’s DeepSeek-R1, an LLM rivalling Western giants, was launched at a fraction of the typical cost (reportedly under $6 million). It served as a wake-up call. Suddenly, the once-distant dream of building localized AI systems did not seem so out of reach. A new AI nationalism took root, not just among digital superpowers but also mid-sized economies seeking greater technological self-determination. Today, the momentum behind sovereign AI is being driven by three strategic imperatives:

1. 1. Data sovereignty and citizen privacy: As public awareness grows around the misuse of personal data, governments are under pressure to keep national data within their borders and under their jurisdiction. Sovereign AI allows tighter control over data governance, aligning with national laws, ethics, and constitutional values.
   2. Geopolitical resilience: The US-China tech standoff has underscored how fragile global digital dependencies can be. Many nations now view AI autonomy as a form of geopolitical insurance, serving as a safeguard against sanctions, service shutdowns, or strategic lock-ins by dominant AI providers.
   3. Cultural and linguistic relevance: Imported models often fail to capture local languages, societal structures, or values. For countries such as India, which is home to dozens of official languages and even more dialects, the ability to build multilingual, culturally aligned models is not just a feature; it is a national imperative.

While these motivations are undeniably strong, they clash with the economic and infrastructural realities many countries face. Building even a modest AI stack requires advanced compute, skilled talent, regulatory frameworks, and long-term capital, all of which remain out of reach for many. The sovereign AI race is no longer about ambition; it is now about access and execution.

The Mirage of Independence: Hidden Dependencies in Sovereign AI

Despite bold declarations of AI autonomy, true sovereignty remains largely elusive. Today, two dominant models are shaping national efforts to build sovereign AI ecosystems:

1. 1. The Stargate model: Under OpenAI’s for Countries initiative, nations can rapidly deploy AI services such as ChatGPT customized with local data governance, region-specific APIs, and sovereign safety layers. This approach leverages in-country data centers and localization controls while still relying on US-origin models, chips, and cloud infrastructure. The UAE exemplifies this model, with its Microsoft-G42 and OpenAI partnerships enabling rapid deployment of citizen-facing AI tools. However, the strategic base layers—model architecture, foundational training data, and semiconductors—remain outside national control.
   2. In-house sovereignty: Several countries are attempting to develop indigenous AI stacks through open-source models, public datasets, and domestic infrastructure. India, for example, is building multilingual LLMs such as Bhashini, setting up GPU clusters under the IndiaAI Mission, and investing in data platforms such as AI Kosh. France, too, has built models such as Mistral and alternatives to mainstream search engines such as Qwant. However, these efforts are deeply intertwined with global supply chains. France is codeveloping AI data centers with the UAE. India still relies on imported chips and foreign cloud infrastructure. Even the most sovereign initiatives are rarely free from strategic and technical entanglements.

The world is moving not toward full-spectrum AI sovereignty but into a bifurcated AI ecosystem with national strategies aligning around two gravitational poles:

• • The US sphere: It includes Europe, the Gulf, and much of the Global South. These regions rely heavily on US-built models (GPT and Claude), US chips (NVIDIA and AMD), and hyperscale cloud providers (AWS, Azure, and Google Cloud). OpenAI’s push for regional customization and localized data storage is designed to deepen this alignment, enticing nations with compliance flexibility while reinforcing backend dependency.
  • The China sphere: It includes countries tied into China’s digital infrastructure through the Belt and Road Initiative or Huawei’s AI/cloud offerings. China is also aggressively exporting full-stack solutions, including chips (Ascend and Cambricon), foundational models (DeepSeek and Baidu ERNIE), and sovereign cloud services.

Caught in the middle, emerging economies are seeking strategic hedges. They do not want to be trapped in the geopolitical crossfire of tariffs, export restrictions, or tech bans. Building partial AI stacks such as local models, custom datasets, and regional data centers is seen as a way to gain bargaining power, even if not full independence. But in reality, most are racing for strategic alignment, not autonomy.

OpenAI’s ambitions to expand sovereign deployments in at least 10 nations, particularly across the Asia-Pacific, reflect a broader US strategy to create sticky, end-to-end dependencies that offer localization at the surface while preserving control over the stack. Meanwhile, China’s DeepSeek-R1 and Huawei’s AI ecosystems offer a low-cost, high-performance counterweight, aimed at undercutting Western dominance in the Global South. This dual-stack realm presents an illusion of choice, not a path to genuine sovereignty.

What Real Sovereignty Would Require

No country today has achieved 100% sovereignty across all layers. The goal is not isolation but strategic autonomy. Countries must decide which stack layers are mission-critical to control. Nations can leverage the following framework to assess sovereign AI readiness across foundational, infrastructural, and regulatory dimensions.

Despite efforts to localize AI capabilities, true independence remains elusive in a world where compute, foundational models, and semiconductor supply chains are dominated by a handful of US and Chinese players. This level of independence is currently out of reach for all but two nations.

Conclusion: The Rise of Layered Sovereignty

No nation today enjoys absolute, full-stack AI sovereignty. But the global landscape is rapidly fragmenting into what can best be described as layered sovereignty, where countries selectively assert control over different layers of the AI stack based on strategic priorities, capabilities, and geopolitical constraints.

• • The US remains the most sovereign nation in AI, strategically dominant through its leadership in chips, cloud, and foundational models. Yet, it continues to rely on Taiwan (TSMC) and South Korea (Samsung) for leading-edge chip fabrication. Its AI models are often trained on global internet-scale data, raising friction around data localization, consent, and digital colonialism. While technically ahead, the US is facing growing global resistance from Europe’s regulatory assertiveness to China’s self-sufficiency push and the Global South’s demand for AI equity.
  • China is arguably the only nation pursuing end-to-end AI sovereignty at scale. Despite current dependencies on legacy electronic design automation tools and lagging in advanced node fabrication, it is building a vertically integrated stack, from domestic chips (SMIC and Huawei) and closed-loop datasets to large-scale foundation models (DeepSeek, Baidu, and Alibaba) and internal cloud infrastructure. Its AI development is tightly interwoven with state policy, national security, and digital governance, often limiting foreign model deployment and ensuring strategic control.
  • The Gulf countries, notably Saudi Arabia and the UAE, are redefining sovereignty through strategic investment rather than internal capability. By funding foundational models, acquiring compute infrastructure, and shaping AI narratives, they are securing geopolitical influence without having to build everything in-house. This capital-as-sovereignty model emphasizes access, equity, and long-term narrative control.
  • Europe is wielding regulatory sovereignty as a strategic tool. While it still depends on foreign infrastructure and chips, the EU is leading global debates on AI safety, ethics, and human rights, using its regulatory power via the AI Act to influence deployment standards worldwide. Technical independence is a longer-term goal, but sovereignty of usage norms is already underway.
  • India is carving out functional sovereignty by prioritizing control over datasets, citizen-facing AI services, and multilingual language models. While its foundational models and compute stack are still supported by global partnerships, India is focused on societal alignment, cultural context, and digital public infrastructure as levers of autonomy.

Rather than chasing total independence, the trend is toward layered sovereignty: choosing which parts of the stack to control, influence, outsource, or codevelop. The sovereign AI game is not binary; it is composable. In the coming decade, national AI strategies will be defined not by what a country builds alone but by what it chooses to own, protect, and shape on its own terms.

By Chandrika Dutt, Research Director, Avasant