Blog · Review Essay · Last reviewed June 25, 2026

The Rise of the Network Society and the Infrastructure of Power

Manuel Castells's The Rise of the Network Society is a large, dense theory of the information age. Its lasting value is not that it predicted every later platform, model, or agent. It is that it defined a social form: power organized through programmable networks that route money, work, attention, identity, expertise, and institutional authority.

For this review, network power means the capacity to set the terms of connection: who can enter, which flows are visible, what gets ranked or metered, which identities authenticate, which standards interoperate, what logs survive, and who can appeal when the network misroutes a life.

The Book

The Rise of the Network Society was first published by Blackwell in 1996 as volume one of Castells's trilogy The Information Age: Economy, Society and Culture. Wiley's record for the 2010 edition describes it as the first book in the trilogy, with a substantial new preface; Open Library records the 2010 Wiley-Blackwell second edition as ISBN 9781405196864. The trilogy became one of the canonical sociological attempts to describe what industrial capitalism was becoming after computers, telecommunications, global finance, flexible production, and media systems reorganized everyday life.

The book's ambition is structural. Castells is not only writing about the internet as a communications tool. He is describing an economy and society increasingly organized by electronically processed information networks: financial flows, production chains, managerial systems, urban forms, labor markets, media channels, and institutions whose power depends on connection to the right circuits.

That makes it a natural companion to The Control Revolution, The Master Switch, Platform Capitalism, The Stack, and Protocol. Castells's later The Internet Galaxy narrows the lens to internet culture and politics; this earlier book supplies the wider institutional map. All of them ask where power goes when it leaves the obvious command center and enters infrastructure.

Current Context

As of June 25, 2026, Castells reads less like a prophecy about the internet and more like a dependency map for regulated infrastructure. The EU Digital Services Act names very large platforms and search engines as services whose scale creates systemic-risk, audit, recommender, advertising, and researcher-access duties. The Digital Markets Act turns "gatekeeper" position in core platform services into a competition-governance category. The EU AI Act makes high-risk AI systems answerable through risk management, data governance, documentation, logging, transparency, human oversight, accuracy, robustness, and cybersecurity. NIST's AI Risk Management Framework gives organizations a voluntary process vocabulary for the same work: govern, map, measure, and manage.

The AI-era network also has a harder physical layer than 1990s information-society rhetoric often suggested. The International Energy Agency's 2025 Energy and AI report projects global data-center electricity consumption roughly doubling to around 945 TWh by 2030 in its base case. NIST's 2026 AI Agent Standards Initiative treats agent identity, trust, security, interoperability, and evaluation as standards problems. Those two sources mark the same shift from interface to infrastructure: the network society now means grids, clouds, chips, data centers, model gateways, credentials, logs, and action permissions as much as websites and media flows.

The current question is therefore practical. When an agency, school, hospital, employer, newsroom, or civic organization says it has adopted an AI system, it has also adopted a network of vendors, clouds, models, data sources, gateways, identity layers, contracts, and fallback paths. A Castells-style reading asks who can program that network, who can inspect it, who can route around whom, and who is left with no usable exit.

Networks as Social Form

The book's central move is to treat the network as more than a metaphor. A network is a rule-governed topology: flexible, expandable, selective, and able to route resources through connected nodes while excluding what does not fit its logic. Castells argues that the information-technology paradigm strengthens this form by making coordination across distance faster, cheaper, and more programmable.

This matters because network power often looks less like a boss issuing orders and more like a condition of access. To be connected to financial markets, logistics chains, search systems, cloud platforms, data brokers, standards bodies, app stores, identity providers, or model ecosystems is to be inside the field where decisions can travel. To be disconnected is not simply to be offline. It is to be socially and economically peripheral.

The stronger definition is this: a network society is one in which the ability to route, rank, meter, observe, authenticate, exclude, coordinate, and translate across systems becomes a primary form of power. The important edges are not just cables or links. They are protocols, prices, permissions, rate limits, recommender rules, procurement contracts, audit logs, identity standards, and interoperability terms.

That definition matters because it separates network analysis from simple internet boosterism. Castells's "space of flows" names a social order in which capital, decisions, images, commands, and expertise move through infrastructure faster than local institutions can often respond. The power is not only in moving quickly. It is in deciding which flows count, which nodes are legible, which interfaces become mandatory, and which places are treated as endpoints rather than authors of the system.

The language of networks can sound emancipatory: openness, decentralization, peer connection, flexibility. Castells is more useful when he shows the harder side. Networks include by selecting. They exclude by routing around. They can be adaptive without being democratic. They can break old hierarchies while installing new dependencies. That is the recurring pattern across the site's reviews of The Master Switch, Cloud Empires, and Technofeudalism: the control point moves from visible command to infrastructure access.

Labor Inside the Network

The Rise of the Network Society is especially valuable as a labor book. Its account of flexible production, informational work, global restructuring, and the "network enterprise" helps explain why digital systems so often arrive as reorganizations of employment before they arrive as consumer gadgets.

In networked capitalism, the firm becomes less like a sealed factory and more like a coordination system: contractors, suppliers, teams, databases, metrics, logistics, finance, and temporary projects linked by information flows. That structure can create new autonomy for some workers and severe insecurity for others. The same network that lets specialized work travel can also make labor easier to monitor, compare, outsource, and discard.

That point has sharpened in the AI era. Models and agents enter workplaces that were already modularized, measured, and routed through software. The important question is not only whether a model can perform a task. It is whether the organization has already decomposed work into units a system can observe, score, allocate, and replace. That is why this review belongs beside The Platform Society, Ghost Work, and algorithmic management: the visible interface depends on institutions, metrics, data labor, and contractual arrangements that remain easy to miss.

The governance implication is direct. Worker safety in a network enterprise cannot be reduced to a model-performance question. It needs notice when automated systems are used, limits on surveillance, access to records used in evaluation, contestable scoring, procurement review, collective voice, and a human path for consequential decisions. Otherwise the network's flexibility becomes one-way flexibility: organizations can reconfigure work instantly while workers inherit opacity and risk.

Media, Politics, and Legitimacy

Castells also helps explain why media politics became central to institutional legitimacy. A network society does not only move capital and work through information systems. It moves attention, reputation, public conflict, crisis narratives, and political identity through media circuits.

This is one reason the book pairs well with The Culture of Connectivity, The Filter Bubble, The Chaos Machine, and The Revolt of the Public. Castells is writing before the mature platform era, but he gives the social architecture that later platforms exploited: politics becomes inseparable from communication networks, and authority becomes vulnerable to flows it cannot fully command.

The practical lesson is that institutions cannot treat communications systems as mere publicity channels. In a network society, the channel becomes part of the institution's operating environment. It shapes what publics can see, what claims travel, what evidence feels current, and which actors can coordinate faster than formal authority can respond.

That is also a governance problem. A platform does not govern only by deleting posts. It governs through ranking, recommendation, monetization, account identity, ad delivery, enforcement workflow, appeal design, and researcher access. The site's platform governance page names these as public-scale controls even when they are operated by private firms.

The AI-Age Reading

Read in 2026, The Rise of the Network Society looks like a prehistory of AI infrastructure power. Generative models did not create the network society. They arrived inside it: cloud platforms, chip supply chains, training-data pipelines, API ecosystems, data centers, identity systems, app marketplaces, payment rails, standards bodies, public agencies, outsourced labor markets, and media environments already organized by informational flows.

AI intensifies Castells's frame because cognition becomes a network service. Search, writing, coding, translation, tutoring, customer support, legal triage, medicine, hiring, logistics, surveillance, and public administration can now pass through model-mediated systems. The network no longer only connects people and organizations. It can also supply statistical interpretation, prediction, and automated action.

That change makes access and exclusion more consequential. Who gets the best models, the cheapest compute, the strongest integrations, the most useful data, the audit logs, the appeal process, the human fallback, and the right to disconnect? These are not side questions. They are the politics of a society whose basic functions increasingly run through networked machine cognition.

The concrete map now has layers Castells could only partly see: compute governance, cloud dependency, model routing, prompt and tool permissions, data-center siting, evaluation labor, incident reporting, and public procurement. The International Energy Agency's 2025 Energy and AI report projects global data-center electricity demand roughly doubling to around 945 TWh by 2030 in its base case, with AI as one driver of demand growth. That does not make every AI use unjustified. It does mean the network society has a material footprint in grids, water, land, chips, cooling, and transmission capacity, not only in screens and datasets.

Agentic AI adds another layer. Once software can call tools, retrieve records, submit forms, spend funds, change permissions, or act through institutional accounts, network power includes delegated authority. The civic controls are then identity, authentication, authorization, revocation, logs, provenance, incident response, and separation between read, write, send, publish, spend, delete, and change-permission scopes. NIST's 2026 AI Agent Standards Initiative treats secure and interoperable agent systems as standards work around identity, trust, security, and evaluation, which is exactly where Castells's abstract network analysis becomes operational. The governance question is not whether an agent sounds competent; it is which network permissions let its interpretation become action.

A useful AI audit follows the path of a decision through those layers rather than stopping at the chatbot box or model card. The related pages on model routing and AI gateways, public compute, cloud infrastructure, and The Stack make the same point in more operational terms.

Governance and Safety

Current regulation makes the network-society reading less abstract. The EU Digital Services Act requires very large online platforms and search engines to assess systemic risks arising from the design, functioning, and use of their services, including algorithmic systems; it also creates obligations around mitigation, independent audit, recommender transparency, ad repositories, and data access for vetted researchers. The Digital Markets Act targets gatekeepers that operate core platform services, turning platform position into a legal category for competition and fairness obligations. The EU AI Act treats high-risk AI systems as systems that need risk management, data governance, technical documentation, record keeping, transparency, human oversight, accuracy, robustness, and cybersecurity. NIST's AI Risk Management Framework gives a process vocabulary for this work: govern, map, measure, and manage.

Castells helps explain why those rules converge. They are not only content rules, product rules, or model rules. They are attempts to govern chokepoints in networks: who can see, rank, combine, route, audit, exit, interoperate, and appeal. The practical safety problem is not only a bad model output. It is a bad output moving through identity systems, recommender feeds, medical or hiring workflows, payment rails, API chains, cloud outages, or public-agency records with no accountable human path back.

That makes a network inventory the first safety artifact. A public agency, school, hospital, newsroom, or employer adopting AI should be able to name the model provider, cloud provider, data sources, training or retrieval records, model-routing layer, user-permission model, logging policy, human-review path, appeal route, incident owner, retention rules, subcontractors, and exit plan. If those relations are not knowable, then the institution has adopted not just a tool but an ungoverned dependency.

A stronger network safety case should answer five questions: what flow is being routed, who controls the gate, what evidence is logged, who can interrupt or appeal, and how the institution exits without abandoning affected people. That turns Castells's "space of flows" into reviewable paperwork: system inventory, bill of materials, audit trails, procurement terms, incident records, and human-review authority.

For AI governance, the right controls therefore include logs, provenance, redress, independent audit access, data minimization, role-based authorization, incident reporting, procurement records, public registers, human oversight, meaningful exit rights, and evidence that affected people can contest consequential outcomes. The site's pages on AI governance, AI audits and assurance, algorithmic impact assessments, human oversight, public AI registers, vendor and platform governance, and transparency registers are all ways of making network power inspectable.

Where the Book Needs Friction

The book's breadth is also its burden. Castells sometimes writes at a scale where the network can feel like an all-purpose explanation. Later scholarship has pushed back on that tendency. Andrea Miconi's 2022 article in American Behavioral Scientist reads Castells's theory across later work and argues that network and society are not the same thing; technical, political, and social affairs follow different patterns.

That criticism matters for AI governance. It is too easy to say "the network did it" or "the platform did it" or "the model did it" and lose the specific institutional choices that make harm possible. A procurement rule, moderation policy, labor contract, API price, safety threshold, data-sharing agreement, interface default, cloud service agreement, or standards decision can matter as much as the broad social form.

The book also predates smartphones, mature social platforms, cloud hyperscalers, large language models, data-center politics at current scale, and today's AI safety and labor debates. Readers should not use it as a finished map of the present. Its value is as a structural lens: it teaches where to look when visible institutions depend on invisible connections.

There is also a normative risk. Networks can be efficient, resilient, and creative; they can also concentrate power by making dependence feel like participation. A good reading of Castells should not romanticize decentralization or treat connection itself as liberation. The question is always who can program the network, who can switch between networks, who bears the failure cost, and who has standing to change the rules.

What This Changes

The lasting lesson of The Rise of the Network Society is that modern power often appears as connection before it appears as command.

A person, worker, city, school, newsroom, clinic, agency, or movement may experience the system as tools, feeds, dashboards, clouds, accounts, rankings, APIs, and assistants. Beneath that surface is a question of position. Which networks can they enter? Which flows can they influence? Which standards classify them? Which systems can observe them? Which institutions can route around them?

The practical reading habit is simple: when an institution says it has adopted an AI system, map the network around the decision. Identify the model provider, compute provider, data sources, API gateway, identity layer, payment or procurement terms, logging policy, human review path, appeal process, incident owner, and exit plan. A model-mediated institution is not only a smarter institution. It may relocate judgment into a network of vendors, datasets, metrics, prompts, policies, and automated actions. The humane test is whether people can understand, contest, exit, repair, and democratically govern the connections that now govern them.

Source Discipline

This review separates book interpretation from current governance claims. Bibliographic facts come from Wiley and Open Library. Castells's concepts are used as interpretive tools, not as proof that one network logic explains every social outcome. Later criticism, especially Miconi's argument that technical, political, and social patterns do not collapse into one network form, is part of the reading.

Current legal claims are jurisdiction-specific. The DSA and DMA are European Union regulations with specific thresholds, designations, and obligations. The EU AI Act is a separate risk-based framework for AI systems. NIST AI RMF and the AI Agent Standards Initiative are standards and risk-management sources, not binding global law. The IEA energy material is a projection about data-center demand, not a claim that every AI deployment has the same footprint.

Network claims need the same discipline. Do not say "the network" caused a harm when the operative choice was an API policy, recommender objective, procurement clause, cloud region, identity rule, subcontract, rate limit, or missing appeal path. The point of network analysis is to locate control points, not to blur responsibility into abstraction.

This article makes no claim that any AI system is conscious, divine, or AGI. It treats models, agents, recommenders, and answer engines as institutional systems that route information, labor, permissions, and delegated action through infrastructure.

Sources

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

Buy on Amazon Browse Books

Return to Blog · Return to Books