← Back to Perspectives
Regulating a capability is not the same as owning it.
Who owns the intelligence the rules are trying to govern?
Governments No Longer Own AI.
August 5, 2026
Artificial intelligence has stopped being just a technology sector.
Governments still write the laws. They approve defence budgets. They command armed forces. They regulate markets.
But many no longer own the systems that will increasingly shape all four.
The most advanced AI models, the computing infrastructure that trains them, the cloud systems that run them, the satellites that feed them and much of the software entering national command structures are controlled by private companies.
Governments are responding with regulation, procurement and privileged-access agreements.
That looks like control returning to the state.
It is not.
But many no longer own the systems that will increasingly shape all four.
The most advanced AI models, the computing infrastructure that trains them, the cloud systems that run them, the satellites that feed them and much of the software entering national command structures are controlled by private companies.
Governments are responding with regulation, procurement and privileged-access agreements.
That looks like control returning to the state.
It is not.
Regulating a capability is not the same as owning it.
The rules arrived after the power moved
More than 1,000 AI rules are now moving across 69 countries. Europe has armed its regime with fines reaching 7 per cent of a company’s global revenue. Governments, regulators and institutions are moving at speed.
The obvious reading is that the state is taking charge.
The stronger reading is the opposite.
Governments write rules for capabilities they do not directly control. They regulate banks because they do not own every bank. They regulate energy companies because they do not operate every grid. They regulate technology platforms because the platforms grew powerful outside the state.
AI has now crossed the same line.
The rush to regulate is not proof that governments are ahead. It is evidence that they are trying to catch up.
In May, California, the European Union, India and the Vatican moved on AI within days of one another. The timing looked coordinated. It was more revealing than that. Different institutions had reached the same conclusion at the same time: the capability was moving faster than their authority over it.
The rules matter. They can limit harms, impose standards and create consequences.
But they do not answer the central question.
The obvious reading is that the state is taking charge.
The stronger reading is the opposite.
Governments write rules for capabilities they do not directly control. They regulate banks because they do not own every bank. They regulate energy companies because they do not operate every grid. They regulate technology platforms because the platforms grew powerful outside the state.
AI has now crossed the same line.
The rush to regulate is not proof that governments are ahead. It is evidence that they are trying to catch up.
In May, California, the European Union, India and the Vatican moved on AI within days of one another. The timing looked coordinated. It was more revealing than that. Different institutions had reached the same conclusion at the same time: the capability was moving faster than their authority over it.
The rules matter. They can limit harms, impose standards and create consequences.
But they do not answer the central question.
Who owns the intelligence the rules are trying to govern?
The capital shows who is in front
The world spent about $1.5 trillion on AI in 2025. That figure was on course to pass $2 trillion. Four American companies alone were projected to spend $562 billion. Sixty-one per cent of global venture investment was chasing the same field.
Capital at that scale does more than build products.
It builds infrastructure, talent concentrations, technical standards and dependency.
A government may pass a law. A company may still own the model, the data centre, the specialist engineers and the computing capacity required to make the system work.
This creates a new imbalance.
The state retains formal authority. The supplier controls practical capability.
That distinction will define much of the next decade.
A country may have the legal right to act but lack the independent systems required to act at speed.
It may have a defense doctrine but depend on privately developed software to identify targets.
It may have an intelligence service but rely on commercially owned models to process information.
It may have a space programme but depend on privately controlled satellite infrastructure for access, communications or targeting.
The machinery of sovereignty is becoming a supply chain.
Capital at that scale does more than build products.
It builds infrastructure, talent concentrations, technical standards and dependency.
A government may pass a law. A company may still own the model, the data centre, the specialist engineers and the computing capacity required to make the system work.
This creates a new imbalance.
The state retains formal authority. The supplier controls practical capability.
That distinction will define much of the next decade.
A country may have the legal right to act but lack the independent systems required to act at speed.
It may have a defense doctrine but depend on privately developed software to identify targets.
It may have an intelligence service but rely on commercially owned models to process information.
It may have a space programme but depend on privately controlled satellite infrastructure for access, communications or targeting.
The machinery of sovereignty is becoming a supply chain.
Access is replacing ownership
The clearest signal did not come from a major product launch.
It came from a quiet access arrangement.
Microsoft, Google and xAI agreed to let the United States government test their newest models before public release, including versions with safety restrictions reduced or removed. The government would see what the systems could really do before competitors, citizens and many other states.
That is valuable.
It is also revealing.
It came from a quiet access arrangement.
Microsoft, Google and xAI agreed to let the United States government test their newest models before public release, including versions with safety restrictions reduced or removed. The government would see what the systems could really do before competitors, citizens and many other states.
That is valuable.
It is also revealing.
States do not inspect what they own. They inspect what they depend on.
Early access gives a government warning, influence and advantage. It may expose security risks before release. It may allow national institutions to prepare for capabilities others have not yet seen.
But access remains permission.
It is not ownership of the model. It is not control of the infrastructure. It is not a permanent right to future versions. It is not the ability to prevent the supplier from changing terms.
A government with privileged access may be closer to the centre of power than one without it.
It is still outside the company that built the system.
This is how dependence can look like sovereignty.
The flag remains over the institution. The critical capability sits elsewhere.
But access remains permission.
It is not ownership of the model. It is not control of the infrastructure. It is not a permanent right to future versions. It is not the ability to prevent the supplier from changing terms.
A government with privileged access may be closer to the centre of power than one without it.
It is still outside the company that built the system.
This is how dependence can look like sovereignty.
The flag remains over the institution. The critical capability sits elsewhere.
The rental problem
Dependence becomes dangerous when AI enters decisions a state cannot afford to delay, expose or lose.
Military command is the clearest case.
AI is moving beyond information support and into targeting, operational planning and recommendations about how to attack. The systems at the center of this shift are often built by private companies and integrated through contracts rather than sovereign ownership.
The risk is not only whether the system works.
The deeper risk is whether the state controls its continued use.
A rented system can be repriced.
It can be restricted.
Its terms can change.
Its supplier can be acquired.
Its service can be interrupted.
Its underlying model can be altered without the user controlling the next version.
The same logic applies beyond defense.
A government that depends on external AI for public services, critical infrastructure, intelligence assessment or emergency response has accepted a form of strategic exposure.
Some dependence will be unavoidable. No state can build every layer alone.
But leadership must distinguish between ordinary procurement and the outsourcing of sovereign judgement.
The relevant question is not whether a supplier is competent.
It is whether the capability is so essential that losing it would weaken the state’s freedom to decide.
Military command is the clearest case.
AI is moving beyond information support and into targeting, operational planning and recommendations about how to attack. The systems at the center of this shift are often built by private companies and integrated through contracts rather than sovereign ownership.
The risk is not only whether the system works.
The deeper risk is whether the state controls its continued use.
A rented system can be repriced.
It can be restricted.
Its terms can change.
Its supplier can be acquired.
Its service can be interrupted.
Its underlying model can be altered without the user controlling the next version.
The same logic applies beyond defense.
A government that depends on external AI for public services, critical infrastructure, intelligence assessment or emergency response has accepted a form of strategic exposure.
Some dependence will be unavoidable. No state can build every layer alone.
But leadership must distinguish between ordinary procurement and the outsourcing of sovereign judgement.
The relevant question is not whether a supplier is competent.
It is whether the capability is so essential that losing it would weaken the state’s freedom to decide.
Sovereignty is not isolation
The answer is not for every government to recreate every model, data centre and satellite network.
That would be slow, expensive and unrealistic.
Sovereign capability does not require complete self-sufficiency.
It requires control at the points where dependence becomes dangerous.
That may mean national computing capacity for the most sensitive systems.
It may mean domestic or allied models for defence and intelligence.
It may mean access rights that cannot be withdrawn during a crisis.
It may mean clear ownership of data, audit rights, continuity provisions and the ability to move critical functions between suppliers.
It may mean trusted alliances that treat AI infrastructure as strategically as energy, telecommunications or weapons systems.
The objective is not to own everything.
It is to ensure that no outside actor can quietly acquire a veto over an essential national decision.
That would be slow, expensive and unrealistic.
Sovereign capability does not require complete self-sufficiency.
It requires control at the points where dependence becomes dangerous.
That may mean national computing capacity for the most sensitive systems.
It may mean domestic or allied models for defence and intelligence.
It may mean access rights that cannot be withdrawn during a crisis.
It may mean clear ownership of data, audit rights, continuity provisions and the ability to move critical functions between suppliers.
It may mean trusted alliances that treat AI infrastructure as strategically as energy, telecommunications or weapons systems.
The objective is not to own everything.
It is to ensure that no outside actor can quietly acquire a veto over an essential national decision.
The questions leadership must now ask
Most organizations still ask whether they are adopting AI quickly enough.
That question is already too small.
Leadership should ask:
Which decisions will depend on AI?
Who owns the systems behind those decisions?
Who controls the data, infrastructure and updates?
What happens if access is restricted?
Which capabilities can be rented safely?
Which must remain under direct institutional control?
Where is the exit route if the supplier changes terms?
These are not technical procurement questions.
They are questions of authority.
A government can regulate a company and still depend on it.
It can negotiate access and still lack ownership.
It can sign a sovereign AI contract and still place its most sensitive functions inside a private system.
The label does not settle the issue.
Control does.
That question is already too small.
Leadership should ask:
Which decisions will depend on AI?
Who owns the systems behind those decisions?
Who controls the data, infrastructure and updates?
What happens if access is restricted?
Which capabilities can be rented safely?
Which must remain under direct institutional control?
Where is the exit route if the supplier changes terms?
These are not technical procurement questions.
They are questions of authority.
A government can regulate a company and still depend on it.
It can negotiate access and still lack ownership.
It can sign a sovereign AI contract and still place its most sensitive functions inside a private system.
The label does not settle the issue.
Control does.
The real contest
The defining divide of the AI age will not be between countries that use artificial intelligence and countries that do not.
Almost every serious institution will use it.
The divide will be between those that understand their dependencies and those that discover them during a crisis.
Between those that retain authority over critical decisions and those that rely on permission.
Between those that build control into the system and those that assume a contract is control.
Governments are now trying to recover influence through laws, access deals and defence partnerships.
That effort is necessary.
It is not enough.
Almost every serious institution will use it.
The divide will be between those that understand their dependencies and those that discover them during a crisis.
Between those that retain authority over critical decisions and those that rely on permission.
Between those that build control into the system and those that assume a contract is control.
Governments are now trying to recover influence through laws, access deals and defence partnerships.
That effort is necessary.
It is not enough.
Continue Reading
Frontier AI Models: Why AI Releases Are Becoming Clearance Events
Governments are moving AI regulation upstream, reviewing frontier AI models before release as national security and model deployment become increasingly connected.
The Sovereign Imperative: Why Nations Must Own Their Intelligence Stack.
Outsourcing intelligence infrastructure once reduced costs; in the AI era, it creates strategic dependence.
No Universal Winner. Kimi K3 Reached the Frontier Fastest.
A new generation of AI companies is reaching the frontier in record time, redefining how competitive advantage is built.