If AI becomes infrastructure, will sovereignty still belong to the state?

I. From "Number of Factories" to "Invisible Smart Grid"
In the industrial age, the surface indicators of a nation's strength were the number of factories and steel production. However, what truly differentiated nations was the underlying infrastructure system: how railways were laid, who controlled the ports, how the power grid was managed, and who controlled energy transmission and distribution. These "intermediary systems," seemingly purely technical issues, actually reflected the structure of sovereignty and power.
In the internet age, the surface is filled with a dazzling array of apps and websites, but at a deeper level remains a highly centralized intermediary layer—search engines, cloud computing platforms, operating systems, and communication protocols. These determine how data flows, how users are indexed, and who controls computing and distribution capabilities.
Today, artificial intelligence is being pushed to the same position. This time, infrastructure is no longer just delivering energy or data, but "intelligence." Large-scale models and inference systems have permeated finance, scientific research, manufacturing, media, and defense. They are not only service tools but also prerequisites for all upper-level systems. AI is no longer just software but is forming a new type of "smart grid."
The core question then arises: as this smart grid is increasingly built, operated, and controlled by a few multinational corporations and private capital, is "sovereignty over infrastructure" still equivalent to "national sovereignty"?

(Image caption) As energy, computing power, data centers and model systems gradually become coupled, AI is no longer just a single tool, but is forming a new "smart grid," reshaping the underlying structure of information flow, resource flow and power flow.

II. Why is AI becoming "electrified"?
For AI to be viewed as infrastructure rather than a commodity, three key conditions must be met: universality, upstream concentration, and externalities.
1. Versatility: Becoming a cutting-edge AI capability layer. AI models have been embedded in writing, program development, customer service, automated decision-making, risk analysis, and R&D processes, gradually becoming a universal capability across various industries, rather than a single-point application. For enterprises, it is more like "callable intelligent capacity"—similar to electricity or cloud resources.
2. Upstream Concentration: Controlling the model and energy, i.e., controlling the entire chain of training and deploying cutting-edge models, requires massive computing power, data centers, and long-term capital. This leads to a high degree of upstream concentration. This concentration is not only reflected in the model itself, but also extends to the overall structure of energy-computing power-network: whoever controls cheap and stable energy, advanced process chips, and globally distributed data centers can lock other participants into a secondary intelligence level.
3. Externalities: Risks have spilled over corporate boundaries. Large-scale AI deployment will profoundly change the division of labor, education models, public opinion production, military capabilities, and financial stability. These impacts span the entire society and have been elevated to a level similar to systemic financial risks or disruptions to critical energy infrastructure. Therefore, AI can no longer be regarded as "ordinary software" but will inevitably enter the public governance sphere.
When a capability simultaneously possesses versatility, high upstream concentration, and significant externalities, its governance issues naturally escalate to "sovereignty issues." Just as electricity and payment systems require strict regulation because their collapse would destabilize social order, AI is approaching the same level.

(Image caption) High-density AI racks are driving the transformation of data centers from traditional air cooling to liquid cooling, closed-loop cooling, and higher power density architectures. As computing power density rises rapidly, power and thermal management become the true "hard boundary" of AI infrastructure.

III. The real bottleneck of AI : no longer algorithms, but electricity. In the cost structure of AI, GPUs and cloud resources are often the focus, but the real ultimate bottleneck is rapidly emerging - electricity itself.
The International Energy Agency (IEA) predicts that by 2030, global data center electricity consumption could reach approximately 945 TWh, exceeding the current total electricity consumption of Japan, with AI being the primary driver of this growth.
OpenAI's Stargate project highlights this trend. The project plans to invest up to $500 billion to build approximately 10 gigawatts (GW) of AI infrastructure, equivalent to the power load of a megacity.
This set of numbers has three meanings:

• The competition in cutting-edge AI is now directly tied to "who can secure massive amounts of clean electricity in the long term".
• Any country that cannot provide an energy-computing power combination within its own or allied systems will experience a structural gap in "intelligent sovereignty".
• When such electricity demand is planned and managed by private platforms, it is tantamount to allowing companies to partially control the load structure and prioritization of the national grid.

The energy issue of AI has become a core variable in national energy strategies and industrial restructuring.
IV. How can "smart grids" rewrite sovereignty?
Traditional national sovereignty centers on territory, military, and monetary systems, assuming that critical infrastructure can ultimately be directly controlled by the state. However, in the AI era, this premise is undergoing a qualitative change.
This change can be divided into three levels:
1. From controlling land to controlling loads: While the power grid remains nominally under national jurisdiction, the load structure is being reshaped by a few AI platforms. As data center clusters consume gigawatts of power, they are effectively defining "what constitutes rigid demand." The state may have to choose between "ensuring uninterrupted AI computing" and "residential and industrial electricity consumption," a choice often pre-determined by upfront investment agreements.
2. From Control Cloud to Control Models and Standards: Even with localized cloud computing and electricity, "intelligent sovereignty" will still be lacking if core models and algorithm standards are monopolized by foreign platforms. True sovereignty depends on the ability to autonomously deploy, operate, and adjust AI infrastructure, including localized data center design and upper-level control frameworks, under energy constraints.
3. From controlling capital flows to controlling computing power allocation: AI infrastructure is highly capital-intensive, and decisions often exceed traditional public budgets. When the largest projects rely on private equity capital and platform market capitalization, the state effectively shares the decision-making power over the "smart grid" path with the capital market.
At all three levels, sovereignty is "outsourced" to enterprises to varying degrees: enterprises select sites, negotiate energy, and design loads in advance, while the state adjusts these matters afterward through approval and regulation, rather than planning from a sovereign perspective in advance.

(Image caption) When data centers are deployed on a campus scale and planned with gigawatt-level power demand, they are no longer just back-end facilities for technology companies, but new strategic assets that can influence local power grids, land use and industrial structure.

V. The Dilemma of Shared Infrastructure and "Sovereign AI " Faced with this situation, some countries and international organizations have proposed the concepts of "sovereign AI" and "shared AI infrastructure," attempting to reduce dependence on a single private platform through multilateral or regional platforms.
The core idea is to not require each country to build its own complete stack, but to form a smart base that is "accessible but not completely privatized" through public-private partnerships or multinational data centers, model libraries and power grid expansion.
However, this path also presents structural challenges:

• Time and scale: Data center planning can be completed in a few months, but approvals for power grid expansion, land, water resources and cooling facilities often take several years.
• Capital and risk sharing: Who bears the construction costs of "quasi-public" infrastructure? Who is responsible for sunk costs when technology changes or demand falls short of expectations?
• Standards and Interoperability: Even if the infrastructure is open and shared, if the upper layer runs a closed model and API, sovereignty is still constrained by the software layer standard setters.

To some extent, the sovereignty issue in the AI era has shifted from "who owns the data center" to "who sets the interoperability standards for energy, computing power, and models." Energy sustainability indicators (such as carbon intensity and water resource consumption) will also become hard constraints.
VI. When Energy Becomes AI 's "Strategic Depth"
In discussions about energy transition, AI is often seen as an optimization tool (such as smart grid dispatching and improving the utilization rate of renewable energy). But from another perspective, energy itself is becoming a strategic focus for AI.
Multiple regions are restructuring their investment strategies around AI data centers and energy projects. Regions with abundant clean energy and land are vying for global AI data center hub status; large-scale AI projects could potentially lock in local energy exports and price structures for decades to come.
From the perspective of sovereignty, the question becomes: can a state still unilaterally decide its own energy strategy, or must it continuously negotiate between the energy needs of AI platforms and the needs of domestic society and industry?

(Image caption) As AI computing density increases, liquid cooling systems are widely regarded as a crucial infrastructure solution for improving energy efficiency, reducing thermal risks, and supporting ultra-high-power server racks. This also means that the competition in AI is increasingly embedded in the energy and industrial infrastructure layers.

VII. How can the state reclaim sovereignty over AI infrastructure?
As AI gradually becomes "smart electricity," the reclaiming of national sovereignty cannot rely solely on traditional data localization or platform regulation; instead, it requires a restructuring in the following areas:
1. The integrated energy - computing plan regards AI as a new "critical energy-consuming industry" and incorporates it into the national medium- and long-term energy plan, including setting energy consumption quotas, levying energy taxes or carbon taxes, and requiring large-scale projects to assume more obligations in renewable energy procurement and grid investment.
2. Public AI Infrastructure and Sovereign Models: Through sovereign wealth funds, public clouds, national laboratories, and regional alliances, a partially public-capital-led stack of AI infrastructure and open-source models is being built to reduce reliance on single private vendors. This is not complete nationalization, but rather ensuring the nation's ability to maintain core intelligent services even in extreme circumstances.
3. Redefining sovereignty at the institutional and standards level: Sovereignty can be reconstructed as "having the final veto power and the right to determine the direction of critical infrastructure." This includes establishing dedicated access and ongoing regulatory mechanisms for cutting-edge models and large-scale AI infrastructure, introducing systemic risk assessments for cross-border data and model output, and promoting AI industry standards with energy sustainability as a hard constraint.
The key point is that the state needs to regard AI as an institutional prerequisite on par with currency and the power grid, and incorporate it into the core of sovereignty discussions.

(Image caption) As model capabilities increasingly rely on the supply chains of advanced chips, server clusters, and data centers, whoever controls chip manufacturing processes, computing hardware, and infrastructure nodes will be closer to true dominance in the AI era.

8. Private Smart Infrastructure: A Temporary Reality or the New Normal?
Returning to the initial question: When AI becomes infrastructure, does sovereignty still belong to the state?
The reality is rapidly moving towards the "era of private smart infrastructure"—the largest AI computing projects are led by private platforms, with energy and land provided through negotiations with local governments, and funding coming from global capital markets.
However, historical experience shows that true infrastructure has never remained in a state of purely private governance for long. From railways to telecommunications, from payment clearing to internet backbones, whenever a system becomes a necessary prerequisite for other activities, the state and society will ultimately demand the embedding of stronger public control and accountability mechanisms.
The real issue may not be whether sovereignty still belongs to the state, but rather:

• Does the state have the capacity to understand and exercise "new sovereignty in the intelligent era"—including energy sovereignty, computing power sovereignty, and standards sovereignty?
• Can society establish sustainable institutional arrangements that balance private innovation and public safety?
• For countries and regions that are unable to build their own complete AI infrastructure, can the international order provide a sharing solution that does not come at the cost of complete dependence?

With AI still experiencing exponential growth and energy and power grids catching up, these questions remain unresolved. However, one thing is certain: when intelligence is built, financed, billed, and distributed like electricity, sovereignty will be silently rewritten in practice if nations do not proactively rewrite their sovereign boundaries—only the pen may be in the hands of others.