The New Periphery

There is a cognitive habit that makes certain kinds of understanding harder than they need to be. It is the habit of treating each new phenomenon as genuinely unprecedented, of refusing the recognition of pattern on the grounds that the current instance is different enough from previous instances to require entirely fresh analysis. The habit is not always wrong. Genuine novelty exists. Some phenomena really do break cleanly from what preceded them and demand new frameworks.

But the habit has a cost. It discards frameworks that were built at great expense, from experience that was painful and accumulative, and that describe structural relationships with precision. When a new phenomenon resembles a familiar pattern, the question is not whether the resemblance is perfect. It never is. The question is whether the structural logic is similar enough that the existing framework illuminates more than it distorts.

The suite of essays preceding this one has examined, from multiple angles, the transformation that AI and automation are producing in the global economy: the foreclosure of development pathways, the failure of educational contracts, the political consequences of blocked aspiration, the inadequacy of distributional frameworks, the absence of adequate governance. Each essay examined a piece of the architecture.

This essay names the shape of the whole.

The shape is familiar. It is dependency.

The Structure of Dependency

#

The word is used loosely in everyday language: relying on something, needing something, finding it difficult to do without. This is not the meaning I want.

The structural definition is more precise. A dependency relationship exists when one actor’s economic and social reproduction requires access to systems that another actor controls, and the dependent actor cannot build, own, maintain, or exit those systems without the controlling actor’s cooperation.

This definition has several features that are important to hold clearly.

It does not require malice. The controlling actor need not intend to dominate. The structure of the relationship produces the dependency independent of anyone’s intentions. The British textile industry did not need to conspire against Indian weavers. The economic architecture accomplished the suppression automatically, through normal market operation, because the terms of the relationship were set by actors whose interests were served by those terms.

It does not require explicit extraction. Value flows from periphery to center through normal transactions, through access pricing, through the direction of surplus investment, without anyone behaving unethically in any conventional sense. The extraction is structural, not personal. This is what makes it durable. Personal cruelty can be reformed. Structural architecture reproduces itself.

And it is self-reinforcing. Once established, dependency relationships tend to deepen rather than resolve, because the gap between the productive capacity of the center and the productive capacity of the periphery grows when the center controls the infrastructure. The dependent actor uses the system. Using the system trains the system. Training the system makes the center more capable. The capability gap widens through normal use.

These features describe the AI dependency relationship precisely. The companies building the most consequential AI systems are not engaged in a coordinated effort to dominate the global south. They are responding to competitive pressures and investment incentives that reward scale, integration, and the accumulation of training data. The dependency that emerges from this is structural, not intentional. That does not make it less of a dependency.

Three Forms, Three Centuries

#

The current form is the third instantiation of a structural relationship that has organized the global economy for several centuries. Each form is harder to see than the previous one, and harder to exit.

The first form was colonial commodity extraction. The periphery provides raw materials. The center processes them into finished goods. The finished goods return to the periphery at prices that capture the value of processing. The periphery cannot develop its own processing capacity because the terms of the relationship, maintained through explicit political and military power, prevent it. The dependency is visible. The coercion required to maintain it is visible. The structural position of the dependent party is understood by everyone in the relationship, including the dependent party.

This visibility had a consequence that the architects of colonial extraction did not anticipate: it organized resistance. Colonial subjects knew they were colonized. That knowledge was painful. It was sometimes deliberately suppressed. But it was not in structural doubt, and the political movements that eventually disrupted colonial dependency were built on clear recognition of what the relationship was.

The second form was postcolonial financial dependency. Formal political independence without economic independence. Development finance from international institutions conditioned on structural adjustments that opened peripheral economies to goods and capital from the center while the center’s own markets remained managed. The periphery produces for the center’s consumption and borrows from the center to fund its development. Debt service flows outward. The surplus generated by peripheral production is captured by the financial architecture rather than accumulated domestically.

This form is maintained through financial architecture rather than political control. It is less visible than colonial extraction. The language of development assistance, of structural adjustment as necessary medicine, of conditionality as the price of access, obscures the structural relationship. Many of the actors implementing this form genuinely believed they were helping. The structural consequences were independent of their beliefs.

The third form is the current one. A country’s economic life becomes organized around digital and AI infrastructure it cannot build, own, or modify. Its population’s behavior trains the systems. Its institutions are administered through software governed in other jurisdictions. Its economic transactions flow through payment infrastructure designed by entities outside its legal reach. Its workers are mediated to employers through platforms whose terms of service are set elsewhere. The surplus generated by this infrastructure flows to the owners of the infrastructure, who are, with very few exceptions, concentrated in a small number of wealthy countries.

This form is maintained through technical architecture. It requires no occupation. It requires no explicit conditionality. It requires only that the systems be indispensable and that the capacity to build alternatives not exist.

It is nearly invisible. It looks like access. It feels like participation. The dependent position and the subjective experience of empowerment are fully compatible with each other. A farmer in Ghana who uses a satellite-connected AI advisory service to improve her crop yield is genuinely better off than she was without it. She is also, structurally, more dependent on infrastructure she cannot govern than she was before. Both things are true. The second thing is harder to see precisely because the first thing is real.

The Complication: Where the Frame Breaks

#

An honest argument requires naming where it fails.

China is breaking the dependency. Not completely, and not without friction. Its dependency on Taiwan Semiconductor Manufacturing for leading-edge chips is real and structurally significant. The American export control strategy is specifically designed to maintain this chokepoint, to keep China dependent on foreign fabrication capacity for the most advanced semiconductors. The chip dependency is not trivial.

But at the layer that matters most immediately, the model layer, China has built genuine alternatives. DeepSeek demonstrated, in early 2025, that frontier-capable AI models can be built outside the American AI ecosystem at dramatically lower cost and with different architectural choices than Western models assumed were necessary. Qwen, Ernie, and the broader Chinese AI stack represent a second center, not a peripheral node consuming another center’s output. China is not solving a dependency. It is building the infrastructure of a competing dependency relationship for others.

India is pursuing a partial exit through a different path. The India Stack — Aadhaar as a universal identity infrastructure, UPI as a payment architecture, the broader digital public infrastructure that the Indian government has built and made openly available — represents the most serious attempt at application-layer digital sovereignty that any developing country has achieved. It is genuinely remarkable. India has built infrastructure that hundreds of millions of people use daily, that is governed by Indian institutions, that generates data that accumulates domestically rather than flowing to foreign servers. At the application layer, India has established a form of sovereignty that most countries have not.

But India’s foundation model capabilities remain early, and its semiconductor manufacturing is essentially nonexistent. The application layer is sovereign. The infrastructure underneath it is not. India is more accurately described as a country that has demonstrated the institutional capacity and political will to attempt the transition from peripheral to participant, and that has achieved partial success on a specific dimension of that transition. Whether it completes the journey is genuinely open.

These complications matter. The dependency frame is not a claim that all countries are equally peripheral, or that no exits are possible. China’s partial exit demonstrates that exit is structurally possible for a country with sufficient scale, state capacity, and sustained strategic investment. India’s partial exit demonstrates that sovereignty can be built incrementally, at specific layers, even before foundational-layer independence is achieved.

What the complications do not change is the structural position of most of humanity.

The Bipolar Problem

#

China’s emergence as a second AI center does not resolve the dependency problem for the global south. It restructures it.

If the global AI landscape settles into two centers, an American ecosystem and a Chinese ecosystem, then peripheral countries face something structurally analogous to Cold War bipolarity. The choice is not between dependency and independence. It is between which center to depend on. The surplus still flows outward. The infrastructure is still built elsewhere, owned elsewhere, governed by rules made elsewhere.

The bipolar structure may be more constraining than a unipolar one in a specific way. Alignment with one center forecloses access to the other. A country whose economy runs on Chinese AI infrastructure, whose payment systems use Chinese platforms, whose institutional data lives on Chinese cloud infrastructure, cannot easily switch to American alternatives if the geopolitical relationship shifts. The dependency becomes directional in a way that limits the leverage that playing centers against each other might otherwise provide.

This is not an abstract geopolitical concern. It is already the operational reality for countries in the Belt and Road ecosystem that have built their digital infrastructure on Chinese technology. The dependency is real, the switching costs are high, and the political relationship that comes with the infrastructure is not separable from the infrastructure itself.

For sub-Saharan Africa, for Southeast Asia outside China, for Central Asia, for much of Latin America and the Middle East, the question is not whether to be peripheral. It is which center to be peripheral to, and under what terms. That is a narrower question than it sounds.

What Is Genuinely New

#

The historical pattern is real. But honest analysis requires naming what the current form does that previous forms did not.

The scale is without precedent. Colonial extraction targeted specific resources in specific territories. Financial dependency affected large populations but through a specific mechanism, the financial system, that could in principle be exited through debt repudiation or currency controls. The current form reaches into the organization of essentially all economic and social activity in proportion to digitization. As economies digitize, which they are doing at accelerating speed, the infrastructure dependency extends into every sector.

The speed of establishment is genuinely new. Colonial dependency was established over generations. Financial dependency deepened over decades. AI infrastructure dependency is being established over years, in some cases over months, as digital payment systems, AI administrative platforms, and algorithmic mediation of labor markets become baseline infrastructure faster than any governance framework can assess their implications.

The invisibility is qualitatively different. Colonial subjects knew they were colonized. The knowledge was politically constitutive. The invisibility of the current form is not merely a matter of insufficient information. It is structural. The empowerment that comes with access is real. The dependency that comes with access is equally real. They coexist in the same transaction, in the same device, in the same experience of using a system that is genuinely useful and genuinely not yours.

The Governance Vacuum Is Not an Accident

#

The international governance architecture is inadequate to the current form of dependency, and that inadequacy is not an oversight.

The World Trade Organization was built for goods trade, extended awkwardly to services, and has no framework for AI infrastructure dependency as a development issue. The International Monetary Fund and the World Bank were built for development finance in a world where the development path was the manufacturing ladder that the preceding essays have described as foreclosed. Their frameworks address capital flows, not the structural conditions of AI infrastructure dependency. The United Nations system conducts AI governance discussions that are advisory, non-binding, and focused primarily on safety and rights within wealthy-country contexts.

This vacuum reflects the interests of the actors with sufficient power to shape the institutions. Countries that control AI infrastructure have incentives to maintain governance frameworks that do not constrain their operation or require them to share the surplus. Countries that do not control AI infrastructure lack the leverage to demand frameworks that would. The governance architecture reflects the power architecture. This has always been true of international institutions. It is as true now as it has ever been.

What is different is the rate at which economic facts are being established relative to the rate at which governance frameworks develop. Colonial dependency was established over a period long enough that resistance movements had time to organize, that the structural relationship had time to become visible, that the international legal frameworks of the twentieth century had time to develop concepts of self-determination and sovereignty that, however imperfectly, addressed the problem.

The current form is being established faster than the institutional response cycle operates. The dependency facts on the ground may be established before the governance frameworks that would address them are developed. This is not neutral. Dependency relationships that are fully established are historically significantly harder to renegotiate than dependency relationships that are still being formed.

The Conditions for Exit

#

The historical record offers some evidence about what makes dependency relationships renegotiable, and the evidence is neither encouraging nor hopeless.

Collective bargaining capacity has sometimes worked. OPEC in the 1970s represents the clearest case of a peripheral coalition successfully extracting better terms from a previously one-sided relationship. The conditions were specific: a resource with no short-term substitutes, a coalition that held enough of the resource to have real leverage, and the political will to absorb the short-term costs of confrontation. Those conditions are not currently present for AI infrastructure dependency, and building a coalition of the affected countries around data sovereignty or AI governance terms would be organizationally and politically difficult. But it is not structurally impossible.

Technology replication has sometimes worked. Taiwan and South Korea built domestic semiconductor industries through strategic industrial policy that used foreign technology access as a training ground rather than accepting permanent dependency. This took decades, required significant state capacity, and involved deliberate confrontation with the interests that preferred to maintain the dependency relationship. The India Stack represents a partial version of this path at the application layer. Whether any country outside China has the combination of scale, state capacity, and sustained political will to replicate the model at the foundational layer is genuinely uncertain.

Changing terms of the relationship as the center’s interests shift has occasionally produced partial redistribution, though this is the least reliable mechanism and the least available to the dependent party by deliberate action.

None of these mechanisms is currently well-developed for AI infrastructure dependency at global scale. The first requires organization that does not exist. The second requires conditions that few countries possess and a timeline that may not be available given the speed of establishment. The third is not something a dependent country can choose.

I do not offer this as a reason for despair. I offer it as a reason for urgency. The conditions for exit become harder to meet the more thoroughly the dependency is established. The window is not indefinitely open.

Recognition as the Prior Step

#

Every disruption of a dependency relationship in the historical record was preceded by recognition. The relationship had to be seen clearly — named as a dependency, understood as structural rather than natural, traced to the interests it served — before any of the mechanisms for disruption could operate.

This is where the current form presents its deepest challenge.

Colonial subjects knew they were colonized. The knowledge was painful, sometimes deliberately suppressed, but not in structural doubt. The political mobilization that eventually disrupted colonial dependency was built on that recognition.

Financial dependency was harder to recognize. The language of development assistance and necessary structural adjustment obscured the structural relationship. Recognition came slowly and unevenly, and the disruptions that occurred were partial and contested.

AI infrastructure dependency is being established in the language of inclusion and empowerment. The phone that connects the Kenyan farmer to market prices is genuinely useful. The AI diagnostic tool that extends healthcare access in a country with too few doctors is genuinely valuable. The platform that allows the Vietnamese small business to reach global customers is genuinely empowering. These are real goods. They arrive in the same transaction as the dependency. And the fact that the dependency arrives dressed as empowerment is precisely what makes recognition so much harder than in previous instantiations.

The essay this suite has been building toward is, ultimately, about recognition. Not about solutions, which are complex and partially available and unevenly distributed across the geographies this series has examined. About the prior step: seeing the shape of what is being built clearly enough to make choices about it.

What This Series Has Built

#

These seven essays have moved through education, political consequence, technology threshold, development foreclosure, the inadequacy of distributional frameworks, and the absence of adequate governance. Each examined a facet of the same underlying structure.

The structure is a new form of an old relationship, built faster, at larger scale, and with greater invisibility than previous forms. It is not inevitable in its current trajectory. The China case demonstrates that alternative centers can be built. The India Stack demonstrates that application-layer sovereignty is achievable under specific conditions. The history of dependency relationships demonstrates that the terms of such relationships are not permanently fixed.

What the structure requires, before any of the available disruption mechanisms can operate, is recognition. Not the performance of recognition — not the declaration that AI is a new form of colonialism, which is the kind of statement that produces heat without producing the analytical precision that recognition actually requires. But careful, structural recognition: here is what is being built, here is how it resembles what was built before, here is what is genuinely new, here is where the exits are and what conditions they require, here is why the window may not remain open indefinitely.

I wonder whether the institutions that need to provide that recognition, the universities, the development finance institutions, the international governance bodies, the governments of the countries most exposed, have the frameworks and the incentives to do so. The frameworks are underdeveloped. The incentives are complicated by the genuine benefits of the infrastructure they would need to name as dependency-creating.

I do not know whether the window is still open. I think the honest answer is that no one does. What seems clear is that it is not indefinitely open, that the speed of establishment is accelerating, and that the frameworks for thinking about what is being built are lagging the building itself by a widening margin.

The most useful thing this series can do is reduce that lag, even slightly, for the readers who are positioned to act on the recognition.

That is where the analysis ends. Not in despair, and not in false resolution. In the precise description of a condition, offered to people who are living inside it and may not yet have the language to name what they are experiencing.

The shape is familiar. It has been built before. It can be changed.

But first it has to be seen.

The Approximate Mind is a philosophical essay series examining how artificial intelligence transforms human work, identity, development, and society. Parts 63-69, The New Periphery suite, trace the arc from broken educational contracts through the civilizational consequences of automation to the structural dependency that organizes the whole.