Malek M. Itani

The New Geopolitical Chokeholds

Malek Itani — Mon, 31 Aug 2026 21:00:00 GMT

Sovereignty is no longer measured by borders or armies. It is measured by who holds the kill switch to your national decision-making. It is measured by whether a state can continue to think, decide, transact, and act when the systems beneath it are placed under pressure.

Modern power does not always announce itself as occupation. It appears as dependency.

It appears as a cloud contract. A payment rail. A firmware update. A model API. A chip allocation queue. A data-center siting decision. A visa pathway. An export license.

The decisive question is no longer simply whether a state possesses territory. The decisive question is whether the systems on which its decisions depend can be denied, throttled, surveilled, priced, or rewritten by someone else.

This is the new geography of power. The map has moved upward.

Map 01 · Formal sovereignty

For centuries, the political map has made sovereignty look self-contained: a border, a capital, a flag, territory under one state’s authority.

Traditional geopolitics was built around visible choke points: straits, canals, ports, pipelines, mountain passes, maritime routes, industrial basins. These still matter.

But they are no longer sufficient to describe where power lives. Beneath the political map is an operational one: cable landings, power corridors, data centers, government data, payment links, and technical institutions.

Pull back and the system no longer fits inside the border. Cloud control, models, semiconductor fabrication, lithography, payment clearing, and talent clusters sit across several jurisdictions.

Physical location is not operational control. A data center may stand on domestic soil while the systems that provision, update, finance, or repair it terminate elsewhere.

The border is still intact. But once the administrative map fades, the infrastructure network becomes the real map of power.

Geography has not disappeared. It has been recompiled. Now follow a single decision through that map.

From Territory to Infrastructure

A severe flood is approaching the country’s largest coastal city. Before the state can order an evacuation, evidence must arrive: satellite imagery, river sensors, weather models, mobile-network signals, and emergency reports.

The data does not travel directly from the river to the cabinet room. It leaves the country for cloud infrastructure, forecasting systems, geospatial services, and a model endpoint before returning as analysis.

The model can propose evacuation zones. It cannot carry sovereign responsibility. Raw evidence becomes model interpretation; human officials must still convert that interpretation into executive judgment.

A decision without resources is only language. Emergency fuel, shelter capacity, contractors, and cash support must move through domestic banks and international payment-clearing rails.

The order must then reach the public through government cloud systems, telecommunications networks, mapping services, notification infrastructure, and public platforms. The state owns the decision. It may not own its distribution.

Only after buses move, shelters open, hospitals redirect patients, roads close, and contractors receive funds does the decision become physical.

Infrastructure Is the New Veto

The central error in weak discussions of AI sovereignty is treating infrastructure as a support function. It is not.

Infrastructure is command authority in disguised form.

Whoever controls compute can decide who trains, who scales, who waits, and who pays. Whoever controls cloud infrastructure can shape where sensitive workflows reside and under which jurisdictional exposure.

Whoever controls data pipelines can determine what is visible and what remains invisible. Whoever controls payment rails can impose friction without firing a shot. Whoever controls model distribution can alter the decision environment of institutions that believe they are merely buying software.

The veto does not need to be dramatic. It needs to be administrative.

A capacity allocation changes. A compliance requirement appears. A vendor updates terms. A model endpoint is withdrawn. A bank tightens risk rules. A cloud provider delays provisioning. A chip shipment is slowed. A software dependency becomes unavailable.

No invasion. No declaration. No spectacle. Just constraint.

A state that controls its chips but relies on a foreign-controlled energy grid for its data centers has merely moved the kill switch to the transformer.

This is why “access” is an inadequate strategic concept. Access is conditional by design. It can be granted, priced, monitored, modified, or revoked.

Control is different. Control means the ability to continue operating when access is contested.

A sovereign AI doctrine that does not begin with power, chips, networks, data centers, and talent is not doctrine. It is branding. The journey ends at a domestic data center. Walk through its doors.

The Illusion of Rented Sovereignty

There is a comfortable formulation often used in policy circles: sovereignty is partial, interdependent, and negotiated. This is true in normal conditions.

It is useless in crisis.

In a crisis, dependency hardens. The supplier does not need to become hostile; it only needs to prioritize itself. The foreign platform does not need to attack; it only needs to comply with its own state’s law. The cloud provider does not need to censor; it only needs to enforce acceptable-use boundaries written elsewhere. The chip ecosystem does not need to embargo; it only needs to reallocate scarce capacity toward larger or more strategically favoured customers.

A state running its intelligence, defence, financial supervision, industrial planning, and crisis-response systems on foreign-controlled infrastructure has not achieved digital modernization. It has transferred its nervous system into rented space.

This does not mean every layer must be domestic. For most states, full autarky is impossible. But strategic autonomy requires identifying which dependencies are survivable and which are fatal.

A country can import tools and remain sovereign. It cannot outsource the decisive layer and pretend it still commands the outcome.

The key distinction is between ordinary dependence and veto dependence.

Ordinary dependence: Manageable. It involves foreign components, commercial suppliers, and international standards.
Veto dependence: Different in kind. It exists when another actor can disable, delay, degrade, or observe a function essential to national decision-making.

A serious sovereign intelligence strategy must be built around eliminating veto dependence, not eliminating all dependence.

The Human Chokehold

The missing variable in most sovereign AI strategies is talent.

States talk about chips because chips can be purchased. They talk about data centers because data centers can be photographed. They talk about national AI strategies because strategies can be published. The infrastructure arrives: accelerators, cooling, networks, model weights, and backup power.

But the decisive asset is harder to procure: the small population of researchers, engineers, systems architects, security specialists, and operator-analysts capable of building and maintaining these systems under real constraints.

Five rare roles may carry the operational knowledge of thousands of machines. A missing successor, unique production access, tacit recovery knowledge, or dependence on foreign support can turn an impressive facility into an organizational single point of failure.

The global AI labour market is already a strategic battlefield. The top researchers, engineers, and infrastructure specialists are not evenly distributed. They cluster around capital, compute, prestige, frontier problems, compensation, and professional community.

This is not a labour-market inconvenience. It is a national-security constraint.

The infrastructure may sit in Abu Dhabi, Paris, London, Seoul, or Riyadh. But if the people who understand the stack are recruited, retained, promoted, and culturally absorbed elsewhere, the sovereign project becomes a shell.

A sovereign compute cluster without sovereign technical talent is an empty cathedral. The racks keep blinking for a time, while versions drift, alerts accumulate, access freezes, and recovery knowledge disappears.

The gap is filled with remote support contracts. Foreign specialists can provision, update, diagnose, and recover the system. The building is domestic; command remains rented.

Durable capability cannot be solved by filling several vacancies. It requires institutions that continuously produce, retain, connect, and replace technical talent.

Human capital policy is therefore national-security policy. A serious sovereign intelligence doctrine requires:

domestic elite technical education;
security-cleared AI engineering cadres;
compensation structures that can retain rare talent;
prestige pathways outside academia and generic civil service;
technical reserve corps for crisis mobilization;
immigration channels for aligned foreign talent;
institutions where engineers, statisticians, economists, intelligence analysts, and operators work together rather than passing reports across bureaucratic walls.

The state that cannot retain its cognitive elite will not retain technological sovereignty.

The Real Question

The question is not whether a state uses foreign technology. Every state does. The question is which dependencies become kill switches.

Can the state continue to operate if foreign compute access is restricted?
Can it preserve critical data under legal and commercial pressure?
Can it audit the systems shaping public-sector decisions?
Can it maintain payment, logistics, and crisis-response functions under platform denial?
Can it retain the human capital required to adapt under stress?

If the answer is no, sovereignty is formal rather than operational.

The battlefield has moved into the stack. The kill switch is infrastructure.

And the states that fail to understand this will not lose sovereignty all at once. They will lose it administratively, contractually, technically, and silently — one dependency at a time.

The illusion of small ponds

Malek Itani — Mon, 17 Aug 2026 21:00:00 GMT

For a long time, career advice relied on a comforting metaphor: be a big fish in a small pond. Find a narrow niche. Enter an overlooked market. Become the person everyone knows. Build an advantage before the larger players arrive.

It was never entirely wrong. But it has become dangerously incomplete. Today, the pond is rarely as small as it looks.

A freelance designer in Beirut may be competing with specialists in Warsaw, Bangalore, Lagos, Buenos Aires, and San Francisco. A local consultant is no longer compared only with other consultants in the city; clients can now access global expertise, automated research, AI-generated drafts, and highly polished work at a fraction of yesterday’s cost. A student who once stood out for writing a competent report now competes with peers using AI tools to research, code, analyze data, edit prose, translate sources, and produce professional-quality presentations in hours.

The small pond has opened into an ocean. And the fish you thought were big may simply have been competing in a temporarily enclosed pool.

1. The old advantage: limited comparison

Before global digital markets, most competition was constrained by geography, institutions, and access.

A company hired from the people it could find. A university compared applicants within a relatively limited pool. A client relied on local networks and referrals. Information was scarce, coordination was costly, and switching suppliers was difficult.

Under those conditions, being competent could be enough to create distinction.

You did not need to be the best analyst, designer, writer, programmer, teacher, or entrepreneur in the world. Often, you only needed to be better than the alternatives visible to the people making a decision. That is what created many “small ponds.” Not necessarily a lack of talent, but a lack of connection between talent and opportunity. The internet weakened those barriers. Remote work weakened them further. Global platforms made skills searchable and comparable. AI is now removing another barrier: the cost of producing credible work.

The result is not merely more competition. It is more competent competition.

2. Draw the timeline

Sketch when each feature becomes available relative to when the prediction is needed. Leakage is almost always a timeline problem wearing a statistics costume.

3. Ask what the baseline was

If nobody can tell you what the naive rule scores, the model has not been evaluated — it has been described.

Note

This is a placeholder post. Replace the text; keep the front matter shape.

4. Re-run it once

Not to reproduce the numbers. To find out how long it takes, what breaks, and how much of the pipeline lives in someone’s head.

We have no clue what we are doing

Malek Itani — Mon, 17 Aug 2026 21:00:00 GMT

Every few weeks someone hands me a model and asks whether it’s any good. The honest answer is usually “I can’t tell yet,” so I’ve written down the order I work through it.

1. Find the target before the features

The first question isn’t what went into the model, it’s what came out. Ask what the label actually measures, when it was recorded, and who decided it. A surprising share of models fail here: the target is a proxy for the thing anyone cares about, and nobody wrote down the gap.

2. Draw the timeline

Sketch when each feature becomes available relative to when the prediction is needed. Leakage is almost always a timeline problem wearing a statistics costume.

3. Ask what the baseline was

If nobody can tell you what the naive rule scores, the model has not been evaluated — it has been described.

Note

This is a placeholder post. Replace the text; keep the front matter shape.

4. Re-run it once

Not to reproduce the numbers. To find out how long it takes, what breaks, and how much of the pipeline lives in someone’s head.