THE STATE–AI MERGER
How governments across the world are acquiring the compute, models, data and autonomous agents that could make human judgment optional.
Artificial intelligence is no longer merely a product governments regulate.
It is becoming infrastructure governments finance, own, procure, train, host, direct and increasingly depend upon.
The United States is considering whether the public should own stakes in its most powerful AI companies.
Britain has created a sovereign investment fund that exchanges public capital and computing power for strategic influence over domestic AI firms.
China is building a unified national computing network intended to make processing power accessible like electricity, while establishing technical standards through which autonomous AI agents can identify one another, exchange credentials, invoke tools and complete tasks across systems.
Saudi Arabia has placed a full-stack AI company directly beneath its sovereign wealth fund.
The United Arab Emirates is constructing a one-gigawatt AI complex alongside OpenAI, Oracle, Nvidia, Cisco, SoftBank and G42.
Canada is funding a publicly controlled national AI supercomputer and a separate secure facility for government and national-security work.
South Korea is building a public–private national AI computing centre while funding domestic foundation models to reduce dependence on foreign systems.
Japan is rolling out a government AI platform intended to reach approximately 180,000 officials.
Singapore is preparing a registry of the autonomous agents used by its public servants.
Brazil is connecting sovereign cloud infrastructure, national data systems, government identity and domestically governed AI.
The European Union has converted its publicly backed supercomputing network into a connected system of AI Factories spanning much of the continent.
This is not a hidden plan in the conventional sense.
The budgets are public.
The speeches are public.
The contracts are often public.
The problem is that the architecture is disclosed one component at a time.
A sovereign investment fund appears in one announcement.
A national supercomputer appears in another.
A digital-identity wallet is presented separately.
An AI assistant for civil servants arrives through a departmental pilot.
A military taskforce is described as defence modernisation.
A cloud contract is filed under procurement.
A model-access agreement is framed as innovation.
A new data centre is presented as regional employment.
No single document says:
We are constructing an automated operating layer through which the state may eventually perceive, predict, recommend and act.
No document needs to.
The system emerges from the connections.
That is the subject of this investigation.
This Is the Infrastructure Behind Algorithmic Sovereignty
In DON’T SAY I DIDN’T WARN YOU, I warned that the decisive AGI question would not be whether a machine suddenly declared itself ruler.
The real danger was institutional.
Who supplies the model?
Who owns the computing power?
Who controls its updates?
Which databases can it access?
Which actions can it execute?
Who defines the objective?
Who is legally responsible when its recommendation harms someone?
Can an elected minister meaningfully overrule a system whose reasoning neither the minister nor the civil servants beneath them can independently reproduce?
The global evidence now shows that governments are building six interconnected layers.
Layer One: Capital
States are investing directly in AI companies, creating sovereign AI funds, subsidising national champions and using government procurement to determine which companies survive.
Layer Two: Compute
Governments are financing supercomputers, GPU clusters, cloud campuses, energy connections and national computing networks.
Compute determines who can train powerful systems, who can run them at scale and who can be excluded.
Layer Three: Models
States are supporting domestic language models and “sovereign AI” systems trained on national languages, law, culture, records and policy objectives.
Layer Four: Data and identity
Digital identity, payment systems, national data exchanges, health records, tax information and public-service databases provide the inputs that make government AI operationally useful.
Layer Five: Agents and administration
AI is moving from answering questions to conducting workflows: reviewing documents, identifying fraud, processing applications, coordinating services and invoking software tools.
Layer Six: Defence and security
The same states are embedding AI in surveillance, intelligence analysis, cybersecurity, autonomous platforms, military planning and target-development systems.
Each layer can be defended individually.
A country needs computing power.
Public officials need better tools.
Citizens deserve efficient services.
Domestic companies should not depend entirely on foreign technology.
Defence forces cannot ignore what their adversaries are deploying.
The danger arises when all six layers become one operating system.
The United States: From Regulator to Customer, Partner and Possible Shareholder
The United States remains the clearest example of a government–frontier-laboratory merger occurring through procurement, national-security policy and potentially equity ownership.
In June 2025, OpenAI launched OpenAI for Government, announcing a US Department of Defense contract with a ceiling of $200 million.
The work covered administrative operations, healthcare delivery, acquisition data and cyber defence.
By June 2026, the White House had issued a national-security directive describing AI as one of the most transformative technologies in the history of national security and calling for its accelerated integration into the American security apparatus.
The administration also directed the creation of a framework through which selected frontier-model developers could provide the federal government with secure early access to advanced systems for classified benchmarking and cybersecurity evaluation.
Meanwhile, the General Services Administration has developed USAi, a government AI platform giving federal agencies access to multiple commercial models through a common interface.
Its official website is now advertising an AI-agent and Model Context Protocol hackathon intended to connect government datasets and services to systems capable of invoking tools and completing tasks.
That is already a close relationship:
government money;
government contracts;
government-secured model access;
government procurement;
private frontier models;
public-sector agents;
national-security integration.
Then came the equity proposal.
On 2 July 2026, Reuters reported that OpenAI had discussed giving the US government a 5% ownership stake, potentially through a public vehicle modelled on the Alaska Permanent Fund.
OpenAI reportedly suggested that comparable arrangements could extend to other major American AI companies.
The proposal has not been finalised.
It may never be.
But its existence marks an extraordinary shift.
The state would no longer merely regulate a frontier AI laboratory.
It could become:
a customer;
a security partner;
an early-access recipient;
a policy-maker;
and a shareholder with a direct financial interest in the company’s valuation.
That creates a conflict which cannot be solved by promises of responsible partnership.
What happens when a regulator must decide whether to delay a product release that could increase the value of its own holding?
What happens when the state must investigate a laboratory whose profits may feed a public sovereign fund?
What happens when a private company provides the analytical infrastructure used by the military while the military’s political leadership has influence over that company’s financial future?
Public ownership could distribute part of the economic return.
It could also make institutional separation considerably harder.
The United States is not nationalising AI in the traditional sense.
It is developing a hybrid in which the most powerful private laboratories increasingly operate as strategic extensions of the state, while retaining private management, proprietary systems and enormous commercial power.
China: Computing Power as a National Utility
China’s model is more direct.
The Chinese state does not need to create the appearance of strict separation between private technology companies and national strategy.
Its objective is to build AI into the industrial, administrative and communications architecture of the country.
China has launched a national supercomputing network and is accelerating an integrated nationwide computing-power system connecting regional hubs, data centres, telecommunications companies and provincial nodes.
Official Chinese material has described the ambition as making computing power accessible in a manner comparable to water and electricity.
Its 2026–2028 telecommunications plan calls for:
high levels of autonomous intelligence within information networks;
more than 30 high-value AI use cases;
specialised intelligent agents;
and one-millisecond access to computing resources across at least 75% of metropolitan areas.
China’s wider “AI Plus” programme aims to integrate AI into science, manufacturing, consumption, public welfare, governance and international cooperation.
Official targets envisage AI agents and intelligent terminals reaching more than 70% penetration by 2027 and more than 90% by 2030.
The country is also building the standards required for agents to work together.
In July 2026, Chinese authorities published a national agent-interconnection framework covering:
agent identity codes;
identity accounts;
credentials;
authentication;
capability registration;
agent discovery;
message routing;
access to external tools;
and task fulfilment.
This is not simply another chatbot standard.
It is an architectural specification for systems that possess identities, discover other systems, obtain credentials and execute actions across interconnected resources.
China has simultaneously prioritised domestic hardware.
Huawei’s AI clusters, domestic accelerator manufacturers and the national computing network are responses to Western semiconductor controls.
The result is not perfect technological independence.
China remains constrained by advanced chip manufacturing and access to specialised equipment.
But its strategy is unmistakable:
AI is being designed as a national industrial and administrative utility rather than a collection of optional commercial products.
The Chinese state is building the compute, the standards, the data environment, the telecommunications layer and the governance model together.
That may become the most integrated state-AI system in the world.
Britain: Sovereign Capital, Public-Service AI and Frontline Deployment
Britain is following a different model.
It does not own its leading frontier laboratory. Google acquired DeepMind years ago.
It remains heavily dependent on American hyperscale cloud providers, foreign chips and private platforms.
The government’s response is not to abandon that ecosystem but to create strategic footholds within it.
In April 2026, Britain launched the Sovereign AI Fund.
The fund is backed by up to £500 million and is explicitly designed to invest in British AI companies considered strategically important.
It provides:
equity investment;
access to government computing resources;
procurement opportunities;
research support;
and links to strategic national assets.
The government’s first announced equity investment was in the AI-infrastructure company Callosum.
This is not merely a grants scheme.
The programme’s own language describes it as targeted infrastructure deployment.
Britain is exchanging public capital and computing access for strategic participation in companies expected to occupy critical points in the AI value chain.
The government has also published a Compute Roadmap guaranteeing dedicated capacity for Sovereign AI and the AI Security Institute.
A further £1.1 billion hardware programme is intended to support chips, semiconductor capability, researchers and domestic hardware companies.
The administrative layer is advancing at the same time.
The AI Exemplars programme is testing AI across public services.
The Incubator for AI is converting government projects into reusable tools, prompts and implementation guidance.
Government systems are being developed for document analysis, casework, consultation processing, fraud detection and internal productivity.
The defence layer is even more explicit.
In June 2026, Britain created Taskforce RAID, the Rapid AI Delivery Taskforce, to accelerate AI-enabled capabilities into the Armed Forces.
The government says the programme will support faster decision-making, military planning and uncrewed systems.
Britain has also built an AI Model Arena intended to evaluate commercial models against defence benchmarks and speed their procurement into operational systems.
This is the exact convergence explored across:
The state is becoming investor, infrastructure provider, anchor customer, evaluator, regulator and operational user.
Britain calls this sovereignty.
But the sovereign layer still sits above foreign chips, American clouds and private models.
That raises a difficult question:
Is Britain obtaining sovereignty—or merely purchasing a stronger negotiating position inside infrastructure it still does not control?
The European Union: A Continental Compute and Regulation Machine
Europe’s development is often reduced to the AI Act.
That misses the physical architecture.
The EU is not merely regulating AI. It is building publicly supported computing capacity through the EuroHPC Joint Undertaking.
Its AI Factories combine:
AI-optimised supercomputers;
data;
technical support;
research institutions;
startups;
public authorities;
testing facilities;
and sector-specific deployment.
The first factories were placed in:
Finland;
Germany;
Greece;
Italy;
Luxembourg;
Spain;
and Sweden.
Further selections added:
Austria;
Bulgaria;
France;
another German site;
Poland;
and Slovenia.
A later round added:
the Czech Republic;
Lithuania;
the Netherlands;
Romania;
further Spanish capacity;
and further Polish capacity.
AI Factory “antennas” extend the system into:
Belgium;
Cyprus;
Hungary;
Ireland;
Latvia;
Malta;
Iceland;
Moldova;
North Macedonia;
Serbia;
Switzerland;
and the United Kingdom.
The European Commission says combined investment in supercomputing infrastructure and AI Factories over the 2021–2027 period will reach approximately €10 billion.
The factories will be connected to testing facilities and European Digital Innovation Hubs, creating a continental path from public compute to commercial and administrative adoption.
The Commission is also planning AI Gigafactories, designed to provide substantially larger computing capacity.
The proposed Cloud and AI Development Act seeks to reduce strategic dependence and expand Europe’s sovereign cloud and AI ecosystem.
The Apply AI Strategy encourages AI adoption across strategic industries and public services.
Europe is therefore attempting something no individual member state could easily build alone:
a regulated continental AI market resting on shared public computing infrastructure.
Yet this creates another concentration of power.
The European Commission will influence:
which projects obtain subsidised compute;
which models comply with the AI Act;
which systems qualify as European;
which companies receive public infrastructure;
which sectors receive deployment support;
and how standards are interpreted.
Europe may reduce its dependence on Silicon Valley while increasing dependence on a continental administrative layer.
Sovereignty does not automatically mean decentralisation.
Sometimes it means moving control from one large institution to another.
France and Germany: Sovereignty Through National Champions and Foreign Capital
France has positioned itself as Europe’s most aggressive national AI hub.
Its infrastructure plan includes a proposed one-gigawatt AI campus backed by a Franco-Emirati investment arrangement potentially worth up to €50 billion.
Additional commitments from American and international investors cover cloud facilities, data centres, energy and computing infrastructure.
Mistral AI has become the symbolic European alternative to American frontier laboratories.
The French strategy combines:
domestic models;
public research;
nuclear-heavy electricity;
foreign capital;
sovereign-cloud ambitions;
and direct presidential support.
But the word “sovereign” deserves examination.
An AI campus financed partly by Emirati capital, equipped with American chips and connected to international cloud providers may be located in France while remaining dependent on several external power centres.
Germany’s model is more industrial.
It is building EuroHPC AI Factory capacity, sovereign industrial cloud systems and German-language foundation models intended for regulated industries and public administration.
France and Germany have also created a bilateral digital-sovereignty taskforce focused on cloud, AI and cybersecurity.
Their objective is to define what qualifies as a European digital service and use regulation, public investment and state aid to favour it.
That is industrial policy.
It is also political selection.
The state increasingly determines which technological dependencies are acceptable and which companies qualify as national or European champions.
India: Biometric Identity Meets Public Compute
India represents one of the most consequential cases because its AI programme sits above an already mature digital-public-infrastructure stack.
Aadhaar provides biometric identity at population scale.
The Unified Payments Interface provides instant digital payments.
Government data exchanges and service platforms connect welfare, taxation, finance, telecommunications and administration.
The IndiaAI Mission is adding national compute.
Its original target was more than 10,000 GPUs.
By the end of 2025, an official government release said capacity had reached approximately 38,000 GPUs, with further expansion planned.
The programme makes subsidised compute available to:
startups;
researchers;
universities;
government departments;
and public-sector agencies.
It is also supporting domestic foundation models and AI applications built around Indian languages and national priorities.
The government is using innovation competitions to identify systems capable of deployment through line ministries and public services.
This infrastructure must be considered alongside the biometric and data architecture examined in:
and its later evidence audit.
The speculative biological claims remain unproved.
The digital architecture does not.
India is demonstrating what happens when sovereign compute is added to identity, payments and population-scale public data.
AI no longer operates over isolated datasets.
It can operate over the practical infrastructure through which a person proves who they are, receives benefits, pays, travels, communicates and interacts with the state.
That can produce enormous public value.
It also creates an unprecedented capacity for administrative visibility.
The decisive safeguard is not whether the system is described as inclusive.
It is whether people possess:
meaningful consent;
accessible correction;
human appeal;
service alternatives;
and protection from exclusion when the algorithm is wrong.
The Gulf: Sovereign Wealth Funds Become AI Ministries in Corporate Form
The Gulf states are building some of the world’s most concentrated state–AI structures.
United Arab Emirates
Stargate UAE is planned as a one-gigawatt AI cluster in Abu Dhabi, with an initial 200-megawatt phase expected to operate during 2026.
The partners include:
G42;
OpenAI;
Oracle;
Nvidia;
Cisco;
SoftBank;
and Khazna Data Centers.
The project forms part of a wider five-gigawatt UAE–US AI campus.
The UAE is simultaneously investing in American AI infrastructure.
This is not simple foreign direct investment.
It is a strategic exchange:
the UAE offers capital, energy and location;
American companies provide chips, cloud systems and models;
both governments obtain strategic alignment;
and G42 occupies the interface between commercial technology and Emirati state power.
G42’s history in surveillance and its previous relationships with Chinese firms made US access to advanced technology politically sensitive.
The resulting arrangement required the UAE to demonstrate that its infrastructure would align with American security demands.
Sovereign AI therefore becomes a form of geopolitical allegiance.
The country hosting the computers may not control the supply chain.
The country supplying the chips may gain influence over the host’s foreign policy and technology partnerships.
Saudi Arabia
Saudi Arabia’s model is even more direct.
HUMAIN was created as a company owned by the Public Investment Fund and chaired by Crown Prince Mohammed bin Salman.
Its remit includes:
data centres;
cloud infrastructure;
AI chips;
Arabic foundation models;
and sector-specific applications.
HUMAIN is not merely a startup receiving state support.
It is a state-owned full-stack AI vehicle.
It has pursued partnerships with Nvidia, AMD, AWS, Qualcomm and xAI, while seeking financing for gigawatt-scale data-centre capacity.
Reuters reported that HUMAIN was pursuing finance for approximately two gigawatts of capacity—only part of its longer-term ambition.
Qatar
Qatar has created Qai and entered a $20 billion AI-infrastructure partnership with Brookfield.
It has also signed a five-year agreement with Scale AI intended to create more than 50 government applications using predictive analytics, automation and data analysis.
This combines sovereign capital, infrastructure and direct administrative adoption.
The Gulf states are not merely preparing to consume AI.
They are building corporate structures through which ruling institutions can own and direct the national stack.
That may allow faster implementation than democratic systems.
It also means fewer independent centres capable of resisting or auditing the system.
Israel: Civilian Innovation and Military AI Under One National Strategy
Israel has approved a national AI programme intended to expand domestic computing infrastructure and ensure access to advanced processing power.
Its TELEM programme provides a state-backed foundation for national AI research and development.
The country has also established a National AI Directorate to coordinate policy, infrastructure and industrial development.
But Israel cannot be assessed through civilian policy alone.
Its military and intelligence services have become some of the world’s most consequential operational users of AI.
Commercial cloud systems, military surveillance, intercepted communications, image analysis and algorithmically generated targeting have become intertwined through the Gaza war and wider regional operations.
Project Nimbus placed Google and Amazon cloud infrastructure within the Israeli government ecosystem.
Subsequent investigations have raised serious questions about military access to commercial cloud and AI services.
The state–AI merger is therefore not hypothetical in Israel.
The essential elements already exist:
national data;
commercial cloud;
intelligence collection;
domestic defence technology;
military targeting;
and a government programme intended to expand sovereign compute.
The unresolved question is not whether AI is used.
It is whether the public can audit the chain connecting raw surveillance to operational decisions affecting who is detained, targeted or killed.
A human may approve the final action.
That does not establish meaningful human control if the human cannot independently verify the evidence generated by the system.
South Korea: The Sovereign Model Becomes a National Service
South Korea is building one of the most complete democratic sovereign-AI programmes.
The government is establishing a national AI computing centre valued at up to two trillion won, using public–private investment.
It is funding domestic compute and low-interest infrastructure programmes.
It has selected national teams to build sovereign Korean foundation models intended to reduce cultural, economic and security dependence on foreign systems.
The government’s stated objective is to place South Korea among the world’s leading AI powers.
Its strategy draws upon national strengths that few countries possess:
Samsung and SK Hynix memory chips;
telecommunications infrastructure;
domestic cloud providers;
advanced manufacturing;
and major technology conglomerates capable of co-investing at state scale.
The programme is also moving toward universal public access.
Official policy discussions have included a public AI service powered primarily by domestically developed models, with a longer-term ambition to provide personalised agents to citizens.
The national model is therefore not being conceived solely as an industrial tool.
It may become a public interface.
Once a state-backed AI becomes the default system through which citizens receive information, navigate services or interact with government, its training data and policy rules become constitutional questions.
Who determines what the national assistant considers authoritative?
Can political leaders change its behaviour?
Will citizens be told when an answer reflects government policy rather than objective fact?
Can rival models compete on equal terms?
A sovereign model can protect a country from foreign manipulation.
It can also become an official narrator through which the state mediates reality.
Japan: Government AI for the Entire Civil Service
Japan is supporting domestic model development through the GENIAC programme.
GENIAC provides computing resources and institutional support to companies developing foundation models, robotics models and sector-specific AI systems.
Its 2026 projects include models intended to control:
autonomous road vehicles;
drones;
unmanned aircraft;
ships;
and industrial machinery.
The Japanese Digital Agency is simultaneously developing a government AI platform called GENAI.
The platform has been released as open-source software and is being rolled out across ministries.
Japan says approximately 180,000 national-government officials are expected to have access during the 2026 fiscal year.
A further phase is intended to provide advanced generative-AI applications and government-wide common datasets.
This is the administrative threshold that deserves far more attention.
When every official gains access to the same state-approved AI environment, the system may gradually influence:
how evidence is summarised;
how policy options are framed;
how correspondence is drafted;
how legislation is interpreted;
and which alternatives appear reasonable.
The risk does not require malicious programming.
Institutional uniformity is enough.
If thousands of officials rely on the same model, the model’s omissions and assumptions can become government-wide assumptions.
Singapore: Registering the Agents Inside Government
Singapore’s National AI Strategy is directed by a National AI Council chaired by the prime minister.
The government is investing in public research, national language models and sector-specific adoption.
SEA-LION and MERaLiON are designed to reflect Southeast Asian languages, cultural contexts and communication patterns.
Singapore’s public sector already has a centralised AI-assistant environment.
It is now developing a registry to record:
which AI agents are operating;
who owns each agent;
and what each agent does.
That is a rational control.
It is also evidence that the government expects agents to become common enough that they require an administrative registry.
The movement from chatbot to agent matters.
A chatbot produces text.
An agent may:
access a database;
call an API;
initiate a workflow;
prepare an official decision;
update a record;
or communicate with another agent.
Once agents can act, governance cannot focus solely on whether their answers are accurate.
It must govern authority.
Which agent is allowed to change what?
Which credential does it use?
Which person is responsible when agents pass a task between one another?
Can an action be reversed?
Can a citizen inspect the machine’s evidence trail?
Singapore may become one of the earliest real-world laboratories for those questions.
Canada: Public Compute, Secure Government Compute and the State as Anchor Customer
Canada has committed approximately C$2 billion to sovereign AI compute.
The strategy includes:
up to C$700 million to mobilise commercial data-centre investment;
up to C$1 billion for public supercomputing infrastructure;
compute-access support for Canadian companies;
and a smaller secure facility for government, industry and national-security research.
Canada’s national AI strategy also says government will act as a strategic anchor customer for domestic AI companies.
That language is important.
When the state becomes the anchor customer, procurement becomes industrial policy.
Government decides:
which companies receive early revenue;
which systems are validated through public deployment;
which firms accumulate government data;
and which models become embedded in essential services.
Canada says its infrastructure will remain under Canadian governance and protect domestic data and intellectual property.
That is a credible sovereignty objective.
But the architecture will still rely heavily on foreign semiconductor and cloud supply chains.
National ownership of the facility does not necessarily mean national control over every layer beneath it.
True sovereignty would require:
replaceable models;
portable data;
auditable software;
domestic operational expertise;
and the ability to maintain the system if a foreign supplier withdraws access.
Without those safeguards, “sovereign compute” can become a nationally branded dependency.
Australia: A Sovereign Government AI Platform Inside Foreign Infrastructure
Australia’s public service has created an AI plan under which the GovAI platform will provide officials with secure, sovereign AI tools.
The Australian government says data will remain within government-controlled Australian infrastructure.
Australia has also adopted a national AI plan focused on smart infrastructure, domestic capability and public-sector deployment.
In July 2026, it announced mandatory standards intended to govern AI and large data centres, including their energy and water use.
Australia illustrates the distinction between two kinds of sovereignty.
Data sovereignty
The data remains within Australia under Australian legal control.
Technological sovereignty
Australia can build, maintain, modify and replace the underlying models, chips, software and cloud systems independently.
The first is achievable.
The second is much harder.
Australia has strong data-cententre capacity, research institutions and public administration.
It does not currently possess a domestic frontier-model and advanced-chip ecosystem comparable to the United States or China.
Its sovereign government AI may therefore protect sensitive data while still depending upon foreign models or hardware.
That may be sufficient for many public tasks.
But it should not be confused with full independence.
Brazil: Building an AI Layer for a Population-Scale Digital State
Brazil has allocated approximately R$23 billion to its 2024–2028 national AI plan.
The plan explicitly prioritises AI for public services.
Brazil already possesses a substantial state technology apparatus through SERPRO, the federal government’s data-processing company.
SERPRO manages critical identity, taxation, administration and digital-government infrastructure.
It is now building sovereign cloud and AI capabilities designed to keep strategic government systems and data under Brazilian control.
The federal government has established objectives to structure at least 25 high-impact AI projects for public services.
Brazilian authorities have described government identity, data governance, procurement and the GOV.BR platform as foundations for population-scale AI.
The state of Piauí has also developed SoberanIA, a Portuguese-language public model designed to support citizens and policy formation.
Meanwhile, Brazil has entered a cooperation agreement involving SERPRO, the Science Ministry and China’s iFlytek to develop domestic AI capacity for the functioning of the state.
That relationship deserves close scrutiny.
Brazil’s pursuit of sovereignty is simultaneously drawing upon:
domestic public infrastructure;
Chinese technical cooperation;
international cloud hardware;
national identity systems;
and state-directed public procurement.
The objective is independence.
The route to independence may create new dependencies.
Brazil’s case therefore provides a warning applicable everywhere:
A sovereign label does not reveal who supplied the model, who controls the updates, where the training data originated or whether the state can operate the system without its foreign partner.
Chile and Latin America: A Regional Model Before a Regional State
Chile has led the creation of Latam-GPT, an open model designed around the languages, history and cultural context of Latin America and the Caribbean.
The project involves institutions from multiple countries and is intended to function as regional foundation infrastructure rather than a conventional commercial chatbot.
It is important precisely because it offers a different model.
Instead of one government controlling a national assistant, a regional consortium is building open infrastructure that universities, public bodies and developers can adapt.
The model does not eliminate political questions.
Who selected the training material?
Which historical narratives are represented?
Which institutions determine acceptable data?
How are Indigenous languages included?
But its open architecture could make those decisions more inspectable than a proprietary national system.
Latin America also demonstrates that sovereign AI does not always require frontier-scale expenditure.
A culturally and administratively useful model can be built for a fraction of the cost required to compete with the largest US or Chinese systems.
The real contest may therefore not be over who builds the single most powerful model.
It may be over who controls the specialised models embedded in healthcare, education, justice, taxation and public administration.
Russia: State-Aligned AI Under Sanctions
Russia’s national AI programme is strongly state-directed and dependent upon major state-aligned companies.
Its stated strategy includes AI research, education, infrastructure and deployment across healthcare, transport, public administration and defence.
Russia’s challenge is hardware.
Sanctions and export restrictions limit access to advanced chips and manufacturing equipment.
Official Russian submissions have acknowledged shortages in computing power and weaknesses in the domestic electronic-component base.
That does not make Russia irrelevant.
Constraint can push deployment toward systems requiring less compute:
surveillance;
facial recognition;
drone control;
propaganda production;
cyber operations;
battlefield decision support;
and domestic information management.
Russia’s sovereign-AI strategy is therefore less likely to mirror America’s frontier-laboratory model.
It is more likely to merge state data, security services, state companies and militarily useful systems under central political direction.
The Kremlin does not require the world’s most capable general model to create a powerful algorithmic state.
It requires models that are sufficiently capable, politically controllable and integrated into the institutions that already exercise coercive power.
Türkiye: Digital Autonomy Meets Centralised E-Government
Türkiye launched an AI action plan for 2026–2030 with an investment ambition of approximately $10 billion.
The programme includes:
computing capacity;
data centres;
a national AI growth fund;
regulatory sandboxes;
a national AI council;
domestic capability;
and public-service integration.
Türkiye already operates one of the world’s largest centralised e-government systems.
Its e-Devlet gateway serves tens of millions of registered users and thousands of public services.
That creates an obvious path for AI assistants, eligibility tools and administrative agents.
Türkiye describes the objective as digital autonomy.
But the combination of:
centralised political power;
population-scale e-government;
domestic defence technology;
biometric and identity infrastructure;
and AI agents
requires scrutiny far beyond ordinary innovation policy.
An efficient central gateway can simplify citizens’ lives.
It can also create a single administrative point through which access, eligibility, identity and state communication are mediated.
The issue is not whether Türkiye will use AI.
It is whether the citizen retains a meaningful route to a human decision-maker outside the system.
South Africa and Kenya: Sovereign Ambition Meets Physical Constraint
South Africa’s draft national AI policy calls for:
stronger data governance;
expanded computing infrastructure;
supercomputing;
research institutions;
incentives;
and new regulatory bodies.
It explicitly identifies dependence on US and Chinese infrastructure as a strategic vulnerability.
Kenya has also adopted a national AI strategy focused on digital infrastructure, data centres, public services, agriculture, health and regional leadership.
But the African cases expose the physical reality of sovereign AI.
Compute requires:
electricity;
grid stability;
cooling;
water;
fibre;
imported hardware;
and capital.
A proposed Microsoft–G42 data centre in Kenya reportedly encountered fundamental disagreement over its potential electricity demand.
Kenyan President William Ruto warned that the proposed scale could require power equivalent to an enormous portion of the national grid.
That is the hidden cost beneath the language of development.
A country can host a globally significant AI data centre while many citizens still lack reliable electricity.
A foreign-owned campus can increase domestic data capacity while consuming energy, water and land whose alternative uses were never democratically compared.
The World Bank’s private-investment arm is financing data-cententre expansion across countries including:
Uganda;
Ethiopia;
Angola;
Ivory Coast;
Mozambique;
and the Democratic Republic of Congo.
Local hosting can reduce latency and improve regulatory control.
But Africa currently holds a tiny fraction of global data-cententre capacity.
The continent may therefore become the next competitive territory for:
American cloud companies;
Chinese infrastructure providers;
Gulf capital;
development-bank financing;
and national governments seeking AI status.
The critical question is who receives the value.
Does the country gain:
domestic compute access;
transferable expertise;
ownership;
tax revenue;
local models;
and public infrastructure?
Or does it supply land, electricity and political concessions to an externally controlled compute enclave?
The Wider Global Infrastructure Ledger
Not every country possesses a frontier laboratory.
That is not the correct threshold.
A country enters the state–AI infrastructure system when it controls or materially supports one or more of the following:
public AI compute;
nationally hosted cloud;
a sovereign model;
government AI agents;
population-scale digital infrastructure linked to AI;
or operational military and intelligence deployment.
As of July 2026, materially documented infrastructure or deployment exists across the following jurisdictions.
Frontier and full-stack powers
United States: frontier laboratories, hyperscale cloud, chip design, federal AI platforms, national-security model access and potential government equity.
China: national compute network, domestic models and chips, AI-agent standards, state integration and population-scale digital infrastructure.
European sovereign-compute network
European Union: continental AI Factories, Gigafactory plans, public supercomputing, regulation and public-sector adoption.
Material factory or antenna infrastructure extends across:
Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Romania, Slovenia, Spain and Sweden.
Associated infrastructure extends into:
Iceland, Moldova, North Macedonia, Serbia, Switzerland and the United Kingdom.
Portugal has separately launched AMALIA, an open-source Portuguese sovereign language model.
Major national sovereign-AI programmes
Canada: public supercomputer, secure national-security compute and commercial data-cententre support.
India: subsidised national GPU infrastructure, domestic models and population-scale digital public infrastructure.
Japan: GENIAC compute, foundation models, robotics AI and government-wide GENAI.
South Korea: national computing centre, domestic models and public AI-service ambitions.
Singapore: national models, public-sector AI, agent registry and prime-ministerial AI council.
United Arab Emirates: Stargate UAE, G42, sovereign investment and international model partnerships.
Saudi Arabia: PIF-owned HUMAIN, gigawatt data centres, Arabic models and cloud infrastructure.
Qatar: Qai, a $20 billion infrastructure partnership and large-scale government AI deployment.
Israel: national compute strategy, public AI directorate, commercial cloud and military integration.
Türkiye: $10 billion strategy, central e-government and domestic AI capability.
Russia: state-led AI, security integration and sanctioned domestic capability.
Australia: sovereign government AI environment and large-scale data-centre regulation.
Brazil: sovereign cloud, state AI company infrastructure, national public-service projects and public models.
Chile: Latam-GPT and regional public-model infrastructure.
Emerging state-backed infrastructure
South Africa: supercomputing and regulatory plans intended to reduce external dependence.
Kenya: national strategy, data-cententre ambitions and AI-enabled cloud expansion.
Kazakhstan: a national data lake, AI-powered e-government services and an increasingly integrated digital-public-infrastructure environment.
This list does not include every country with an AI strategy.
A strategy document is not infrastructure.
Nor does it include every country using commercial AI in a ministry.
The ledger identifies states and jurisdictions where computing, models, national data, public administration or security deployment have advanced beyond general aspiration.
The Public–Private Structure Is the Obscured Part
The most important finding is not that governments are building AI.
It is how they are building it.
There are three dominant ownership models.
Model One: Direct state ownership
Examples include:
Saudi Arabia’s HUMAIN;
sovereign investment vehicles;
state-owned cloud and data companies;
and public supercomputers.
The advantage is clear accountability for strategic direction.
The danger is the concentration of capital, infrastructure and political authority in one institution.
Model Two: State-coordinated national champions
Examples include:
China’s major technology companies;
South Korean consortiums;
Japan’s GENIAC participants;
France’s support for Mistral;
and Britain’s Sovereign AI investments.
These companies remain formally private but develop through public money, government compute, procurement and national strategy.
The state gains influence without assuming complete legal responsibility.
The company gains public support without becoming fully subject to public-law transparency.
Model Three: Sovereign hosting through foreign suppliers
Examples include:
national cloud facilities operated with US hyperscalers;
Gulf campuses using American chips and models;
African data-cententre projects backed by foreign capital;
and public AI services built upon commercial model APIs.
The data may remain inside the country.
The critical technology does not.
This model can produce the appearance of sovereignty while leaving the state dependent on:
foreign chips;
external model updates;
cloud licences;
proprietary orchestration;
and companies capable of altering access or conditions.
The hybrid model is politically attractive precisely because responsibility becomes difficult to locate.
When something fails, the government blames the supplier.
The supplier says it followed the government’s requirements.
The civil servant says the system produced the recommendation.
The minister says a human remained technically in the loop.
The citizen faces an automated decision whose actual author is everywhere and nowhere.
Digital Public Infrastructure Is the Bridge
The World Bank defines digital public infrastructure around reusable systems such as:
digital identity;
payments;
and trusted data exchange.
The Bank is helping governments expand digital IDs, cloud capacity, data systems and AI readiness.
It presents these systems as modular foundations through which governments and businesses can rapidly create services.
That is accurate.
It is also why the relationship between DPI and AI requires urgent scrutiny.
AI becomes administratively powerful when it can reach reliable structured data.
A model that knows nothing about a citizen can only offer general information.
A model connected to identity, taxation, benefits, employment, health, education and payments can:
assess;
rank;
predict;
recommend;
approve;
reject;
and act.
The OECD’s 2026 Digital Government Outlook found AI being used in at least one area of government in 35 of 36 surveyed OECD countries.
The strongest adoption was found in internal administration and public services.
This means the state–AI merger is not confined to the countries constructing giant sovereign models.
Commercial and smaller models are already entering ordinary government almost everywhere in the developed world.
The political transition is occurring before the technical debate is settled.
Governments are adopting systems whose long-term institutional effects they cannot yet know because the promised efficiency is available immediately.
The efficiency is real.
So is the dependence.
When the Assistant Becomes the Official
The first generation of public-sector AI drafts documents and summarises information.
The second generation will coordinate tools.
The third will execute workflows.
The transition can happen gradually enough that no moment appears revolutionary.
A civil servant asks an assistant to summarise a case.
The assistant is then allowed to retrieve the relevant files.
Next it recommends an outcome.
Then it prepares the official response.
Later it updates the system automatically unless the official intervenes.
Eventually the human checks exceptions while the agent processes routine cases.
At each stage, the human technically retains authority.
In practice, authority shifts toward the machine because:
the machine handles more cases;
the human has less time;
the machine’s recommendation becomes the default;
divergence requires justification;
and officials are judged against automated productivity targets.
The system does not need to command the official.
It only needs to establish the path of least resistance.
That is how human judgment becomes optional without ever being formally abolished.
The Military Layer Removes the Margin for Error
The military adoption of AI is accelerating even faster.
NATO’s strategy calls for AI to be mainstreamed across capability development and military operations.
Britain’s Taskforce RAID is intended to deliver AI to the frontline.
The United States is integrating frontier models into its national-security enterprise.
Israel has already demonstrated the speed at which surveillance and algorithmic targeting can be joined.
Russia and China are developing AI-enabled command, cyber, drone and autonomous systems.
The United Nations Secretary-General continues to call for a ban on lethal autonomous weapons that can select and engage targets without human control.
But the legal debate focuses too narrowly on the final trigger.
The kill chain begins long before a weapon fires.
AI may:
collect the data;
identify the person;
assign a confidence score;
rank targets;
calculate collateral risk;
select the weapon;
propose the time;
and generate the operational plan.
A human can approve the strike while possessing almost no independent ability to inspect how the system reached its conclusion.
That is procedural human control.
It is not necessarily substantive control.
Once military tempo depends on machine speed, the commander who pauses to investigate may be accused of allowing the adversary to gain an advantage.
Human review becomes an obstacle the system is designed to minimise.
The machine does not have to decide who lives and who dies alone.
It only has to make disagreement operationally unaffordable.
The Physical State Beneath the Digital State
AI is frequently described as weightless intelligence.
It is not.
It is land, concrete, copper, fibre, transformers, substations, cooling equipment, water and electricity.
The state–AI merger therefore becomes a resource-allocation system.
Governments decide:
which data centres receive expedited planning;
which projects obtain grid connections;
who pays for transmission upgrades;
how much water can be consumed;
whether public guarantees reduce financing costs;
and whether AI infrastructure receives priority over housing, manufacturing or domestic electricity.
Scotland alone has faced a multi-gigawatt data-cententre pipeline capable of placing extraordinary pressure on the national energy system, as documented in Scotland Data Centre Crisis.
Canada is marketing land, electricity and climate as national AI advantages.
The Gulf is using cheap energy and sovereign capital.
France is using nuclear power.
Kenya is offering geothermal potential while confronting severe grid constraints.
Australia is creating rules requiring major facilities to underwrite new energy and water infrastructure.
The computational state is therefore not merely an information system.
It competes with citizens and industries for physical resources.
The public deserves to know:
the expected power demand;
the expected water demand;
the number of permanent jobs;
the public subsidy;
the tax treatment;
the ownership;
the percentage of compute reserved for domestic use;
and the consequences if the project fails.
A data centre that consumes national resources while serving foreign models is not automatically national infrastructure.
It may be an extraction industry for computation.
What Has Not Been Proved
A serious investigation must state the limits.
The evidence does not prove that one organisation controls every national AI programme.
It does not prove that a single global AGI system has already been selected to govern humanity.
It does not prove that every sovereign model is designed for surveillance.
It does not prove that digital identity inevitably becomes social credit.
It does not prove that every public-sector agent will be allowed to make autonomous legal decisions.
It does not prove that OpenAI’s proposed 5% US government stake will occur.
It does not prove that all AI Factories will become political-control infrastructure.
It does not prove that governments have abandoned human authority.
What it proves is narrower and more important:
States across opposing political systems are acquiring the same foundational capabilities required to automate increasing parts of government.
They are building:
capital structures;
computing infrastructure;
national models;
interoperable data;
identity systems;
public-service agents;
and military deployment pathways.
Whether those systems remain subordinate to democratic authority will depend upon decisions being made now.
Ten Questions Every Government Must Answer
Before a state deploys AI into essential public functions, it should answer ten questions publicly.
1. Who owns the model?
Not merely the brand or service—the weights, intellectual property and authority to alter its behaviour.
2. Where does it run?
Which country, cloud provider, data centre and legal jurisdiction control the infrastructure?
3. Which data can it access?
Can it reach identity, health, tax, benefits, policing, education, employment or payment information?
4. What actions can it execute?
Can it merely recommend, or can it update records, trigger investigations, suspend access, move funds or reject applications?
5. Can the supplier change it remotely?
Can an update alter the system’s behaviour without parliamentary approval or public notice?
6. Is there a complete audit trail?
Can every model input, retrieved document, recommendation, tool call and human override be reconstructed?
7. Does the citizen have a human route of appeal?
Not a chatbot escalation. A legally responsible human capable of changing the outcome.
8. Can the government replace the supplier?
Are the data, workflows and interfaces portable—or has the state become operationally dependent?
9. Who receives the economic value?
Do public investment, electricity, data and procurement create public ownership—or merely private monopoly?
10. What happens during an emergency?
Can normal safeguards be suspended, models expanded into new datasets or agents granted additional powers under emergency authority?
Without public answers, “responsible AI” remains a slogan.
The Constitutional Question Nobody Has Voted On
Every constitutional system assumes identifiable decision-makers.
A minister signs.
An official decides.
A court explains.
A commander orders.
A doctor assesses.
A police officer acts.
Responsibility can be assigned because authority has a human location.
The algorithmic state disrupts that arrangement.
A decision may emerge from:
a commercial model;
running on foreign-designed chips;
hosted in a nationally located cloud;
retrieving government data;
following rules written by civil servants;
adjusted by vendor updates;
approved by an official;
and executed by an autonomous workflow.
Who made the decision?
Everyone contributed.
Nobody fully controlled it.
This is why the debate cannot be confined to whether AI is accurate.
A system can be statistically accurate while constitutionally illegitimate.
It can produce better average outcomes while denying an individual the right to understand the decision made about them.
It can reduce administrative delay while making policy dependent upon a private supplier.
It can preserve human approval while making rejection of the machine’s recommendation practically impossible.
The deepest danger is not artificial intelligence becoming conscious.
It is institutions becoming unable to function without it.
Once government knowledge, workflows, military planning and public services are rebuilt around proprietary models and automated agents, removal becomes economically and operationally painful.
The infrastructure becomes policy.
The Merger Is Already Underway
The United States is debating government equity in frontier laboratories.
China is building computing power as a national utility.
Britain is investing directly in strategic AI companies and accelerating AI onto the battlefield.
Europe is connecting publicly funded AI Factories across the continent.
India is combining sovereign compute with biometric and payment infrastructure.
The Gulf states are turning sovereign wealth into full-stack AI companies.
Japan is preparing government-wide model access.
South Korea is building domestic models and national compute.
Singapore is registering public-sector AI agents.
Canada is constructing a public national supercomputer.
Brazil is connecting sovereign cloud, identity and AI for government.
Israel has already fused commercial AI infrastructure with national security and war.
Russia and Türkiye are embedding AI within centralised state structures.
Africa is becoming a competitive territory for foreign-funded data centres whose resource demands may rival the capacity of national grids.
These systems do not share one ideology.
They share one conclusion:
Intelligence has become strategic infrastructure, and no serious state intends to leave it entirely outside state influence.
The contest is no longer between government and technology companies.
It is between different arrangements of government and technology companies.
Some will call themselves democratic.
Some will call themselves sovereign.
Some will call themselves innovative.
Some will call themselves secure.
All will say the system remains under human control.
The public should ask what that phrase actually means.
Does a human have the legal authority to intervene?
Does the human understand the recommendation?
Does the human have enough time to inspect it?
Does the human possess an independent alternative?
Can the human act without being punished for reducing efficiency?
Can the citizen reach the human?
If the answer is no, the human is not in control.
The human is part of the interface.
The Last Decision Must Remain Human
AI can assist the state.
It can identify waste, process documents, translate languages, detect disease, improve logistics and make public services easier to navigate.
Refusing every use would not protect society.
It would leave public institutions weaker than the private and hostile systems operating around them.
But capability must not be confused with authority.
The machine may calculate.
It may advise.
It may identify uncertainty.
It may present options.
It must not become the unchallengeable source of truth through which rights, freedom and access are distributed.
The state–AI merger is being built now because governments fear dependence upon other governments and foreign corporations.
That fear is rational.
The proposed cure can still create another dependency—this time between the citizen and an automated state whose decisions cannot be meaningfully escaped.
The essential battle is not against intelligence.
It is for accountable authority.
The objective must be simple:
No essential public decision without an identifiable human decision-maker.
No automated action without a complete audit trail.
No identity-linked AI system without a non-digital alternative.
No military target without genuinely independent human verification.
No sovereign model beyond democratic scrutiny.
No public infrastructure captured by a supplier the state cannot replace.
Governments are acquiring the intelligence.
The public must retain the judgment.
Sources
Global government adoption and digital infrastructure
United States
Reuters — OpenAI proposes handing the US government a 5% stake
White House — National Security Presidential Memorandum on AI
White House — Framework for advanced AI innovation and security
US General Services Administration — Artificial intelligence and USAi
China
Chinese Government — Three-year plan for AI integration with communications networks
China Digital Summit — Unified national computing-power system
China Digital Summit — National standards for AI-agent interconnection
United Kingdom
European Union
European Commission — European approach to artificial intelligence
European Commission — Portugal launches the AMALIA sovereign model
India
Gulf states
Reuters — HUMAIN seeks financing for AI data centres and GPUs
Reuters — Qatar and Brookfield establish $20 billion AI-infrastructure venture
Israel
Associated Press — Use of commercial AI models in Israeli military operations
TIME — Google cloud services and the Israeli Defence Ministry
East Asia
South Korean Ministry of Science and ICT — National AI strategy and computing centre
South Korean Ministry of Science and ICT — Sovereign AI Foundation Model project
Japanese Ministry of Economy, Trade and Industry — GENIAC projects
Japanese Ministry of Economy, Trade and Industry — Foundation-model compute support
Canada and Australia
Canadian Government — AI Sovereign Compute Infrastructure Program
Canadian Government — National Artificial Intelligence Strategy
Brazil and Latin America
Brazilian Government — Brazilian Artificial Intelligence Plan
Brazilian Government — AI in national digital-government infrastructure
Russia, Türkiye and Africa
UNCTAD — Russian submission on national AI capacity and limitations
South African Government — Draft National Artificial Intelligence Policy
Kenyan Government — Artificial Intelligence Strategy 2025–2030
Reuters — World Bank backing for African data-cententre expansion
Is there any instance where a citizen gets the choice to opt out of this forced terrorism? I trust no human programmed machine to make decisions for me. Sure, I use a desktop computer which is a programmed machine, but I don't let it make my decisions as it is for mostly used for info gathering and other practical purposes.