Why Governments Are Seeking for Sovereign AI

Sovereign AI: How Governments Are Building Their Own AI Infrastructure

July 2, 2026

For most of the past decade, the infrastructure underpinning artificial intelligence (AI) has been built almost entirely by private companies. A small group of hyperscalers — Amazon, Microsoft, Google, and Meta — constructed the data centers, acquired the GPUs, and developed the platforms that now power the majority of the world’s AI systems. Governments and research institutions largely operated within this ecosystem, consuming compute rather than owning it. That assumption is now being tested.

Across the world, governments are beginning to reassess their relationship with AI infrastructure — no longer content to rent compute from foreign platforms, they are investing in the capacity to develop, deploy, and control AI systems on their own terms. This shift, broadly described as sovereign AI, is moving from policy white papers to capital-intensive reality.

Understanding why sovereign AI is emerging, and what it means for the future of AI infrastructure, supply chains, and global competition — requires looking beyond headlines at the economic, geopolitical, and technical forces now converging.

A Signal Worth Noting

The scale of this shift is already visible in supplier earnings. In its first-quarter fiscal 2027 earnings call, held in May 2026, NVIDIA explicitly highlighted sovereign AI deployments as a key driver of demand. The company reported that its ACIE segment — which captures AI clouds, industrial, and enterprise customers, including sovereign builds — generated $37 billion in revenue, growing 31% quarter-over-quarter.

AI cloud revenue within that segment more than tripled year-over-year. Executives emphasized that sovereign AI projects are ensuring sustainable, long-term demand, shielding the company from potential spending shifts by any single customer group.

For a category that barely registered a few years ago, this is a notable shift in the composition of demand. Sovereign AI revenue, in NVIDIA’s accounting, refers to purchases made by national governments and state-affiliated entities investing in domestic AI infrastructure. The figure does not capture the full scope of global activity, but it provides a concrete signal: governments are becoming direct participants in the buildout of AI infrastructure.

What is Sovereign AI

Sovereign AI is not a precisely defined technical term. In practical usage, it refers to a country’s capacity to develop, deploy, and control AI systems using infrastructure located within its own borders — or at least within its own legal and regulatory jurisdiction. The term is used differently across contexts: some governments mean full-stack independence from foreign technology, while some vendors use it to describe cloud services with data residency guarantees. We will use the stronger definition in this article.

The concept encompasses several related goals. The first is compute sovereignty: the ability to run large-scale AI workloads without depending on foreign-owned cloud platforms. The second is data localization: keeping sensitive national data — government records, healthcare information, financial data and systems — on infrastructure that cannot be accessed or subpoenaed by foreign jurisdictions. The third is strategic autonomy: the ability to develop and maintain AI capabilities independently, rather than relying on systems designed, owned, and potentially restricted by companies based elsewhere.

These goals vary in emphasis by country and context, but they share a common thread: treating AI infrastructure less like a commercial service and more like a strategic asset.

However, it is a mistake to conflate Sovereign AI with the more familiar concept of the Sovereign Cloud. While Sovereign AI is the primary driver of this new infrastructure race, it must be understood in relation to the Sovereign Cloud. If Sovereign AI represents the ‘intelligence engine’—the physical compute capacity of GPUs and high-bandwidth memory (HBM) required to process data independently—then the Sovereign Cloud is the ‘digital fortress’ that houses it. The latter focuses on data residency and legal jurisdiction, ensuring that information remains under national control.

Ultimately, true sovereignty is only achieved when these two layers are unified: a nation must not only have the right to protect its data within a Sovereign Cloud but also the localized compute power to train models, run inference, and extract strategic value from that data on its own terms.

Why Governments Are Accelerating Sovereign AI Investment

Several converging pressures are driving this trend.

National security and strategic control are the most direct motivators. Governments have grown attentive to the degree to which critical systems — logistics, communications, defense planning, public health response — are becoming AI-dependent. Running those systems on infrastructure controlled by foreign entities introduces dependencies that some governments consider unacceptable.

At the 2024 World Government Summit in Dubai, Jensen Huang, CEO of Nvidia, affirmed that every country needs to have its own artificial intelligence infrastructure in order to take advantage of the economic potential while protecting its culture. Huang stated that countries “cannot allow that to be done by other people”.

Export controls and geopolitical tension have sharpened that concern. NVIDIA’s Q1 FY2027 guidance explicitly assumes no data center revenue from China, reflecting the US restrictions in place since April 2025 on advanced AI chip exports. According to a Reuters report, these restrictions were rolled back in August under a new—and controversial—revenue-sharing framework, in which NVIDIA and AMD agreed to pay a 15% fee on all Chinese sales directly to the U.S. Treasury. By December, the same report noted, this rate had increased to 25% for the highest-performance systems.

During the Q&A session, it was noted that NVIDIA’s Data Center business grew approximately 120% in the quarter when excluding China — underscoring both the strength of non-China demand and the volatility of a market now subject to abrupt state policy shifts.

China, in turn, has accelerated domestic semiconductor development efforts. These dynamics have made the geographic distribution of compute capacity a live policy issue, not just an economic one.

For governments watching this shift, the message was clear: access to frontier compute is now a strategic resource that can be taxed, throttled, or redirected by state policy at a moment’s notice.

Economic competitiveness is a third driver. Countries that have observed the economic leverage generated by early AI adopters — in cloud services, productivity gains, and technology exports — are treating AI infrastructure as part of their long-term economic strategy, in the same way earlier generations of governments approached highways, power grids, or telecommunications networks.

This shift was catalyzed in January 2025, when U.S. President Donald Trump announced the Stargate Project—a $500 billion (€485 billion) joint venture between OpenAI, SoftBank, Oracle and MGX to build massive AI data center complexes across the United States.

trump-announces-$500-billion-stargate-ai-infrastructure-project-intellaix — Donald Trump delivers remarks on AI infrastructure next to Masayoshi Son, Larry Ellison and Sam Altman at the Roosevelt Room in the White House.

The European response has been swift. At the AI Action Summit in February 2025, French President Emmanuel Macron unveiled a plant to invest $112 billion (€109 billion) in domestic AI infrastructure. Macron explicitly framed this as a counter-move to the American initiative: “It’s the equivalent for France of what the United States announced with Stargate,” he noted, highlighting that France is investing a similar ratio relative to its population size.

By committing these staggering sums, the largest economies are no longer just subsidizing research; they are entering a capital-intensive industrial race where the “entry fee” is now measured in hundreds of billions of dollars.

Concentration of advanced compute supply adds urgency to all of the above. Advanced AI processors are manufactured at scale by a very small number of companies. TSMC fabricates the chips. HBM memory is produced primarily by Micron, Samsung, and SK Hynix. NVIDIA designs the systems. The supply chain for frontier AI hardware runs through a narrow set of chokepoints. Governments aware of this concentration have reason to secure allocation before it becomes constrained — which, based on current supply chain reporting, it already is.

In this context, access to sovereign AI is increasingly synonymous with access to technological and economic power.

Sovereign AI Infrastructure as a Strategic System

The move toward sovereign AI builds on a deeper structural reality: modern AI systems depend on an interconnected infrastructure stack. Compute, memory, networking, and power must scale together. A country seeking to develop sovereign AI capability is not simply acquiring GPUs; it is investing in entire data center ecosystems capable of supporting large-scale AI models.

This significantly raises the complexity of national-level AI investment. Building sovereign AI infrastructure requires coordination across semiconductor supply chains, energy systems, cooling capacity, and networking architecture. These are multi-year investments — data centers of this scale typically require 3–5 years from planning to operation, and they cannot be scaled independently. As explored in previous articles in this series, each layer of this stack carries its own constraints.

Saudi Arabia’s Project Transcendence, a $100 billion initiative, illustrates both the ambition and the physical constraints of sovereign AI infrastructure at scale, as noted by Forbes in last February. The project requires not only GPU procurement but also dedicated power generation and cooling systems capable of operating in desert conditions where ambient temperatures exceed 45°C — a challenge that has raised questions about water consumption and energy trade-offs. The same analysis concluded that “capital is abundant, but Watts and water are not.”

In this sense, sovereign AI is not just about technological independence — it is about developing the capacity to operate integrated infrastructure systems at scale.

Ripple Effects Across the Ecosystem

The growth in sovereign AI demand does not affect only chip manufacturers. Its effects move through several layers of the technology supply chain. Sovereign buyers alter the demand profile of the semiconductor industry by introducing longer procurement cycles and state-backed capital commitments.

NVIDIA’s networking revenue, driven by deployments of next-generation XDR technology and the Spectrum-X Ethernet platform, has nearly tripled year-over-year to reach $15 billion in Q1 FY2027. This growth reflects a broader pattern — sovereign buyers are purchasing full rack-scale systems rather than individual GPUs, changing the nature of the sale and the scale of deployment, as explored in our earlier analysis of AI networking architecture.

For cloud providers, sovereign AI introduces competitive complexity. Hyperscalers have historically offered governments cloud services under data residency agreements, but sovereign AI infrastructure often implies full domestic ownership — not just residency.

AWS’s European Sovereign Cloud illustrates the boundary: it provides fully independent EU-located infrastructure with EU-based staff and no external operational control, offering strict data residency guarantees. Yet it remains a hyperscaler service model. True sovereign AI, by contrast, requires governments to operate their own compute infrastructure — the “AI factory” — rather than consuming it as a service.

For energy infrastructure, the implications are significant. Training and running large AI models at scale requires substantial and reliable power. Countries building domestic AI capacity are simultaneously confronting questions about power grid investment, cooling infrastructure, and energy sourcing. These are multi-year capital decisions that cannot be decoupled from the compute buildout itself.

The Trade-Offs: Cost and Fragmentation

Sovereign AI investment comes with real costs and risks that deserve honest attention.

Building domestic AI infrastructure is expensive. Replicating at the national level what hyperscalers have spent decades and hundreds of billions of dollars constructing is not straightforward. Smaller economies in particular face significant cost-per-unit disadvantages: they cannot amortize fixed costs across global customer bases, negotiate volume discounts at hyperscaler scale, or dynamically shift workloads to optimize utilization. The result is infrastructure that costs more per unit of compute and may deliver capabilities that lag those available through commercial platforms.

The more structural concern is fragmentation. As different countries build separate AI infrastructure stacks — each with its own regulatory frameworks, data governance rules, hardware configurations, and model development priorities — the global AI ecosystem may diverge into a set of parallel, partially incompatible systems. This risk is compounded when nations invest in data residency (the Sovereign Cloud layer) without corresponding investment in domestic compute capacity (the Sovereign AI layer), creating localized data that still depends on foreign intelligence.

This divergence is already visible. The European Union’s AI Act imposes a risk-based classification system that differs substantially from the sectoral approach in the United States, while China’s algorithmic regulations require content controls that Western platforms do not apply. Data localization laws in various jurisdictions shape what can be processed where. If hardware supply chains further bifurcate along geopolitical lines, the result could be not one global AI infrastructure but several regional ones, each with different capabilities, standards, and interoperability limitations.

Fragmentation is not catastrophic by definition. Regulatory diversity in other sectors — financial services, pharmaceuticals, telecommunications — has not prevented global markets from functioning. But it does add friction, increase compliance costs, and complicate the development of applications intended to operate across borders.

What Sovereign AI Means for the Future of AI Infrastructure

NVIDIA’s first-quarter fiscal 2027 earnings call, with its $37 billion ACIE revenue disclosure, is a data point, not a conclusion. But it is consistent with a broader shift in how AI infrastructure is being conceptualized by governments around the world.

For most of commercial AI’s history, compute power has been allocated primarily by market mechanisms — capital flows to where returns are highest, and infrastructure follows demand. Sovereign AI introduces a different logic: compute power as a strategic resource to be secured, controlled, and in some cases insulated from market access conditions set by other states.

This does not mean private investment is retreating. NVIDIA’s forward guidance — approximately $78 billion in Q1 FY2027 revenue — reflects continued strong commercial demand. The two trends are running in parallel, not in opposition.

What is changing is the composition of who is building AI infrastructure, and why. A technology that emerged primarily from commercial research labs, scaled by private capital, and deployed through consumer platforms is becoming, in parallel, a component of national infrastructure strategy.

That shift will shape the geography of artificial intelligence capability, the structure of supply chains, and who ultimately controls access to advanced compute. AI infrastructure is no longer allocated solely by market forces; it is increasingly shaped by state strategy. For enterprises, investors, and policymakers, the implication is that AI infrastructure decisions will increasingly be made with state interests in mind — not only commercial returns. That shift will influence how compute power is distributed — and who has durable access to it — in the decade ahead.

The sovereign AI deployments driving ACIE’s $37 billion quarterly revenue are not just a revenue category — they are evidence that AI infrastructure is entering the domain of state policy.

This article is being provided for educational purposes only. The information contained in this article does not constitute a recommendation from Intellaix to the recipient, and Intellaix is not providing any financial, economic, legal, investment, accounting, or tax advice through this article or to its recipient.

Sources and references:

Intellaix Focuses on explaining the infrastructure and economics behind artificial intelligence (AI) through clear, structured, and data-driven analysis.

Why Frontier AI Startups Now Require Billion-Dollar Funding

AI Model to Machine: Evolving into Integrated Systems