Does Technology Redistribute Power or Centralize It?
A Person Can Hold More Power While the System Becomes More Concentrated
A smartphone gives an individual capabilities that once required multiple institutions.
It can function as a camera, map, library, publishing system, television studio, bank interface, marketplace, navigation device, translation tool, recording studio, and global communication network.
A small business can reach customers on another continent.
An independent creator can publish without owning a printing press or television station.
A programmer can build software using tools that once required substantial corporate infrastructure.
An individual can ask an artificial intelligence system to perform tasks that previously required specialized knowledge or professional assistance.
From the user's perspective, technology can look profoundly decentralizing.
But underneath that device may sit semiconductor factories, cloud providers, operating systems, telecommunications networks, app stores, payment systems, data centers, and artificial intelligence models controlled by a comparatively small number of organizations.
Both observations can be true.
Technology Redistributes Capability Before It Redistributes Ownership
This distinction is essential.
Giving someone access to a powerful tool increases what that person can do.
It does not necessarily give that person ownership of the infrastructure producing the capability.
A small company can rent computing power without owning a data center.
A musician can reach millions of listeners without owning a distribution network.
A merchant can sell internationally without owning a payment network.
A developer can build on an artificial intelligence model without training one.
Technology can therefore democratize access to capability while leaving ownership highly concentrated.
Power has moved outward in one dimension and inward in another.
The Printing Press Shows That This Tension Is Not New
Technologies have repeatedly altered who can produce and distribute information.
Printing dramatically reduced the cost of reproducing written material compared with manuscript copying.
Radio allowed one transmitter to reach enormous audiences.
Television expanded mass communication further.
The internet lowered distribution costs again and allowed many recipients to become publishers themselves.
Each transition changed the relationship between speaker, distributor, and audience.
But none eliminated intermediaries permanently.
New technologies often destroy old gatekeepers and create new ones.
The Internet Was Architecturally Decentralizing
One of the internet's extraordinary characteristics was that anyone connected to the network could potentially create a server, website, email service, discussion forum, or application.
Publishing no longer required permission from a newspaper editor or broadcasting license.
Open technical standards allowed different machines and networks to communicate.
The cost of distributing information approached levels unimaginable in earlier media systems.
Millions of people and organizations gained capabilities previously available only to institutions with substantial capital.
That was a genuine redistribution of communicative power.
But Attention Did Not Remain Decentralized
As the internet expanded, finding information became increasingly difficult.
Search engines organized the web.
Social networks organized relationships.
Marketplaces organized buyers and sellers.
App stores organized software distribution.
Streaming platforms organized entertainment.
Cloud providers organized computing infrastructure.
These services solved real coordination problems.
They also created new points of concentration.
The open network remained underneath, but an increasing portion of activity passed through intermediaries capable of ranking, recommending, approving, monetizing, or restricting access.
Convenience Can Centralize What Technology Initially Decentralized
Running your own server is possible.
Renting one from a cloud provider is easier.
Building an independent online store is possible.
Joining a large marketplace may immediately provide customers, payments, reviews, logistics, and trust.
Publishing directly on a website is possible.
Using a social platform provides an existing audience and distribution system.
Technical freedom can therefore coexist with economic incentives toward centralization.
People voluntarily choose intermediaries because intermediaries reduce complexity.
Enough such choices can make the intermediary structurally important.
Network Effects Can Turn Usefulness Into Power
The previous cases encountered network effects in competition.
They become even more important in technology.
A communications network becomes more useful as more people join it.
A marketplace attracts sellers because buyers are present and buyers because sellers are present.
A software platform attracts developers because users exist and attracts users because applications exist.
The process can become self-reinforcing.
The technology distributes capability among participants while the network itself accumulates coordinating power.
This creates an unusual form of centralization.
The platform may become powerful precisely because it successfully connected everyone else.
Scale Can Be a Feature Rather Than a Failure
Not every technological concentration indicates malfunctioning competition.
Some technologies become more useful at scale.
A fragmented payment system may be less convenient than an interoperable one.
A cloud provider operating enormous data centers can achieve efficiencies unavailable to thousands of small server operators.
A search system becomes more useful when it can index a large portion of available information.
A global communications platform becomes valuable because people can find one another there.
Centralization can therefore produce genuine efficiencies.
The balance problem is not whether scale exists.
It is what power accompanies scale and whether that power remains contestable.
Infrastructure Is Different From an Ordinary Product
A restaurant can become extremely popular without becoming infrastructure for other restaurants.
A digital platform can be different.
Other businesses may depend on it to reach customers.
Developers may depend on an operating system.
Merchants may depend on payment infrastructure.
Publishers may depend on search or social distribution.
Startups may depend on cloud computing.
Artificial intelligence applications may depend on externally supplied models or computing resources.
When a technology becomes infrastructure, decisions made by its owner can affect entire layers of economic activity.
The Cloud Decentralized Computing by Centralizing Computers
Cloud computing contains the paradox almost perfectly.
A small company no longer needs to purchase servers, construct a data center, employ a large infrastructure team, and predict years of future demand before launching a service.
It can rent computing capacity when needed.
This dramatically lowers barriers to entry.
Computing capability becomes available to far more organizations.
But the physical infrastructure providing that flexibility can become concentrated among enormous providers capable of financing global data-center networks.
Thousands of companies become more independent from owning hardware while becoming dependent on organizations that own hardware at unprecedented scale.
Artificial Intelligence Intensifies the Paradox
Artificial intelligence can redistribute cognitive capability unusually quickly.
A small company can use AI for translation, coding, design, analysis, customer support, research assistance, document processing, and automation.
An individual without a large staff can perform tasks that once required several specialists.
Language barriers can fall.
Technical knowledge can become easier to access.
Software creation can become more accessible.
The productive capability available at the edge of the network expands.
Yet producing the most capable general-purpose AI systems can require enormous quantities of computing infrastructure, capital, specialized chips, engineering talent, and data.
The Current AI Frontier Is Highly Concentrated
Stanford University's 2026 AI Index reports that industry produced more than 90 percent of the notable AI models released in 2025.
The report also finds that the computing resources supporting frontier AI continued expanding rapidly.
This does not mean universities, independent developers, governments, or open communities have disappeared from AI research.
They remain important contributors to publications, methods, evaluation, open models, datasets, and applications.
But the resource requirements of frontier model development give organizations with access to very large amounts of capital and compute a significant structural advantage.
Compute Has Become a Strategic Layer of Power
Artificial intelligence may feel intangible when experienced through a text box.
Its physical foundation is not.
Models run on chips.
Chips operate inside servers.
Servers occupy data centers.
Data centers require electricity, cooling, networks, land, and capital.
Advanced chips require highly specialized manufacturing supply chains.
The World Bank's 2025 Digital Progress and Trends Report describes compute as a foundational component of participation in the AI economy and finds it highly unevenly distributed internationally.
As of June 2025, high-income countries hosted approximately 77 percent of global co-location data-center capacity.
Low-income countries accounted for less than 0.1 percent.
The ability to use AI may spread globally much faster than the ability to own its physical infrastructure.
A Country Can Use a Technology Without Controlling Its Foundations
This creates a national version of the same paradox experienced by individuals.
A country does not necessarily need domestic frontier-model laboratories or semiconductor fabrication to benefit from AI.
Its businesses can access models through the internet.
Its developers can build applications using international cloud infrastructure.
Its citizens can use globally available tools.
This allows countries to adopt technologies without reproducing the enormous investment required to create every underlying layer.
But access and control remain different.
Dependence on external infrastructure can expose countries to prices, contractual conditions, technical standards, service availability, export controls, and strategic decisions made elsewhere.
The AI Divide Is More Than Internet Access
The older digital divide was often described primarily as a question of connectivity.
Who has internet access?
The AI economy adds additional layers.
The World Bank describes four foundations: connectivity, compute, context, and competency.
People need networks.
Organizations need computing capacity.
Models need relevant data and linguistic or cultural context.
People need skills to use, adapt, and build the technology.
A country connected to the internet can therefore remain dependent at several deeper layers of the technological stack.
UNCTAD Finds Concentration in AI Research Too
The United Nations Conference on Trade and Development reported in its 2025 Technology and Innovation Report that just 100 companies accounted for approximately 40 percent of global corporate AI research and development spending in 2022.
With the exception of companies from China, none of those firms were based in developing countries.
UNCTAD also found substantial geographic concentration in AI publications, patents, computing infrastructure, and skills.
The technology may be globally accessible while the capacity to shape its frontier remains far less evenly distributed.
But Open Source Pushes Power in the Other Direction
Technological centralization is not a one-way process.
Open-source software has repeatedly allowed knowledge produced in one organization to become infrastructure available to millions of others.
Developers can inspect code.
They can modify it.
They can combine components.
They can build businesses without licensing every layer from a proprietary supplier.
Artificial intelligence has developed an important open ecosystem as well.
Stanford's 2026 AI Index reports continued growth in open-source AI development, with millions of projects distributed through platforms such as GitHub and Hugging Face.
Open models can allow researchers, businesses, and countries to adapt AI systems without building a frontier model entirely from the beginning.
Open Technology Redistributes Knowledge More Easily Than Infrastructure
Open code does not create a semiconductor factory.
An openly available model still requires computing resources to run.
A public algorithm does not provide electricity.
A downloadable model does not automatically provide local expertise.
Open source can therefore decentralize one layer while dependence persists elsewhere.
This is another reason technological power must be examined as a stack.
Code may be open.
Compute may be concentrated.
Data may be proprietary.
Distribution may depend on a platform.
Chips may depend on a narrow supply chain.
No single label captures the entire structure.
Data Creates a Different Kind of Accumulation
Industrial capital traditionally accumulated factories, machinery, land, inventory, and financial assets.
Digital organizations can accumulate another resource.
Data.
A service with millions of users can observe enormous numbers of interactions.
Those interactions can improve recommendations, advertising, fraud detection, personalization, logistics, product development, or machine-learning systems.
Better services can attract more users.
More users can generate more data.
More data can improve the service again.
Under some conditions, information itself participates in a reinforcing loop of scale.
Algorithms Can Concentrate Attention Without Owning Content
A platform does not need to produce every article, video, song, product, or political statement to possess influence over their distribution.
Ranking systems decide what appears first.
Recommendation systems decide what is suggested next.
Search systems determine which results receive visibility.
Moderation systems determine what remains accessible under platform rules.
Advertising systems determine which messages can be targeted toward which audiences.
The content can remain decentralized while attention becomes partially centralized.
This separates ownership of expression from control over discovery.
Technology Can Give Small Actors Disproportionate New Power
The opposite process remains equally real.
A small company can use global payment infrastructure instead of building a bank.
An independent researcher can access enormous collections of scientific literature.
A filmmaker can produce professional-quality work with inexpensive digital equipment.
A programmer can distribute software worldwide.
A local merchant can reach customers beyond the local market.
A citizen can document an event and make it visible internationally within minutes.
An AI-assisted individual can perform research, analysis, translation, coding, and production tasks that previously required a team.
These are not symbolic redistributions of power.
They materially alter what small actors can accomplish.
The Cost of Coordination Has Collapsed
Technology also changes collective power.
Organizing thousands of people once required offices, mailing lists, telephone trees, newspapers, meetings, and substantial administrative resources.
Digital communication can coordinate large groups almost instantly.
Workers can communicate.
Consumers can organize.
Communities can raise funds.
Political movements can mobilize.
Open-source developers can collaborate across continents.
Scientists can share data.
Small organizations can coordinate at scales once associated with large institutions.
The reduction in coordination costs can redistribute organizational power.
The Same Infrastructure Can Also Observe the Coordination
But digital coordination leaves traces.
Messages pass through networks.
Platforms record interactions.
Devices generate location and behavioral data.
Payment systems record transactions.
Online services authenticate identities.
The technology that makes organization easier can therefore make behavior easier to observe.
Surveillance capacity can be used by businesses for commercial purposes and by governments under varying legal authorities.
The relationship between coordination and observation becomes another technological balance.
Technology can strengthen the organized and the observer simultaneously.
Automation Can Redistribute Power Inside the Firm
Technology also changes internal workplace relationships.
A skilled worker who controls scarce knowledge possesses bargaining power.
If software captures and standardizes part of that knowledge, the organization may become less dependent on the individual.
But another technology may amplify the worker's abilities and make that worker dramatically more productive.
The same category of technology can therefore substitute for one worker while complementing another.
Automation is not merely about how many jobs remain.
It can change who controls the knowledge required to perform the work.
Standardization Can Move Knowledge From Worker to System
Consider a process known mainly through employee experience.
The worker understands unusual cases, remembers past failures, and knows which informal adjustments keep production functioning.
When that process is encoded into software, organizational knowledge becomes easier to reproduce.
The company may become less dependent on particular employees.
New workers can become productive faster.
Quality can become more consistent.
But bargaining power can also shift from the person possessing tacit knowledge toward the organization owning the system.
Technology has increased productivity while simultaneously changing the location of knowledge.
AI May Move This Boundary Into Cognitive Work
Industrial machinery mechanized substantial amounts of physical production.
Software automated many routine information processes.
Artificial intelligence extends automation and augmentation further into language, analysis, coding, design, pattern recognition, and other cognitive tasks.
A worker using AI may become substantially more capable.
A company using the same AI may require fewer workers for particular tasks.
A new entrepreneur may compete with organizations previously protected by the cost of professional labor.
A large corporation may use AI to scale operations further.
The technology does not contain one predetermined power effect.
Its consequences depend partly on who controls it and how broadly access spreads.
Access Can Decentralize Even When Ownership Does Not
This produces a strange economic structure.
Millions of people may gain access to a capability owned by relatively few organizations.
The users become more powerful relative to what they could previously accomplish.
The infrastructure owners become more powerful because millions of users depend on their systems.
Power has not simply moved from one side to another.
More power has been created at both layers.
The relevant question becomes which layer can constrain the other.
Interoperability Changes the Balance
One way technological systems remain contestable is through interoperability.
Email is powerful partly because a person using one provider can communicate with someone using another.
The web became powerful because common standards allowed different browsers, servers, and publishers to participate.
A user did not need permission from one universal website owner to publish a page.
When systems interoperate, users can sometimes change providers without abandoning the entire network.
When systems are closed, switching may require leaving relationships, data, applications, purchases, or communities behind.
Technical architecture can therefore determine economic bargaining power.
Portability Can Turn Formal Choice Into Practical Choice
The same principle applies to data.
If users can export information in usable formats, switching becomes easier.
If businesses can move workloads between infrastructure providers, dependence may decrease.
If workers can carry credentials and benefits between employers, mobility can increase.
If developers can build against open standards, they may have more alternatives.
Portability does not eliminate network effects or concentration.
It changes the cost of exit.
Once again, the architecture of balance returns to the ability to say no.
Technological Sovereignty Has Become an Economic Question
Countries increasingly discuss technological sovereignty because dependence now extends beyond ordinary imported products.
Semiconductors, cloud infrastructure, telecommunications equipment, operating systems, satellite networks, cybersecurity systems, and artificial intelligence can become foundational components of economic life.
Complete technological self-sufficiency is unrealistic for most countries and could sacrifice substantial gains from specialization and international trade.
But extreme dependence on narrow suppliers creates another kind of vulnerability.
The balance problem is therefore not simply domestic production versus imports.
It is whether critical technological dependencies have credible alternatives.
Centralization Can Occur Across Countries as Well as Companies
The World Bank's AI evidence demonstrates this clearly.
High-income countries dominate many layers of innovation and compute infrastructure.
Lower-income countries may gain substantial benefits from adopting technologies developed elsewhere.
That can accelerate development.
But if they remain consumers without building skills, local data, applications, infrastructure, or innovative capacity, technological adoption may deepen dependency even while improving productivity.
The relevant development question is therefore not merely whether a country uses advanced technology.
It is which parts of the technological value chain the country can influence.
Decentralization and Centralization Can Happen at the Same Time
This is the central paradox.
The internet decentralized publishing while concentrating search and social distribution.
Cloud computing decentralized access to servers while concentrating physical infrastructure.
Smartphones distributed computing capability while concentrating operating-system ecosystems and application distribution.
Digital marketplaces empowered small sellers while creating powerful marketplace operators.
Artificial intelligence can distribute cognitive capability while concentrating frontier-model development and compute.
Technology repeatedly pushes power outward at one layer and inward at another.
The Important Question Is Which Layer Controls the Others
A concentrated infrastructure layer may not dominate the system if users can switch easily.
A large platform may remain constrained if competitors can interoperate with it.
A proprietary technology may coexist with meaningful alternatives.
A concentrated semiconductor supply chain may be manageable if substitution and geographic diversification remain possible.
Conversely, an apparently decentralized ecosystem can depend on one hidden bottleneck.
Counting companies or users therefore reveals only part of technological power.
The architecture matters.
Technology Redistributes Power When It Creates New Outside Options
A worker gains power when technology gives access to more employers.
A seller gains power when technology creates alternative routes to customers.
A developer gains power when applications can run across multiple infrastructures.
A country gains power when critical technologies can be sourced from several suppliers.
A creator gains power when audiences can be reached without one compulsory intermediary.
An individual gains power when sophisticated tools become affordable.
In each case, technology creates alternatives.
Those alternatives strengthen the ability to refuse an existing relationship.
Technology Centralizes Power When It Removes Outside Options
The reverse is also true.
A platform gains power when leaving means losing an audience.
An infrastructure provider gains power when migration becomes prohibitively expensive.
An employer gains power when technology makes workers interchangeable while workers have few alternative employers.
A technology supplier gains power when customers cannot use competing systems without abandoning accumulated data or workflows.
A country becomes vulnerable when critical infrastructure depends on one external source.
Centralization becomes structurally important when dependency loses credible alternatives.
The Real Unit of Technological Power Is Dependency
Technology itself is neither naturally decentralizing nor naturally centralizing.
Its power effect depends on the dependencies it creates and destroys.
A technology can destroy dependence on one institution while creating dependence on another.
The printing press weakened some information monopolies but required presses and distribution.
The internet weakened traditional publishing gatekeepers but created new discovery platforms.
Cloud computing removed the need to own servers but increased dependence on cloud infrastructure.
AI can reduce dependence on specialized human expertise while increasing dependence on models, chips, data, and compute.
Technological history is partly a history of dependencies being rearranged.
Capability at the Edge, Control at the Core
This may describe much of the current technological landscape.
Extraordinary capability is moving toward individuals and small organizations.
A person with a laptop can access tools that would have been unavailable to entire companies a generation ago.
At the same time, some of the infrastructure supporting those tools requires capital and scale beyond the reach of almost any individual.
The edge becomes stronger.
The core can become stronger too.
The two developments are not contradictory.
They are occurring together.
Technology Does Not Decide Who Holds Power
Technical architecture matters.
Ownership matters.
Competition matters.
Open standards matter.
Interoperability matters.
Skills matter.
Public institutions matter.
Capital requirements matter.
Geography matters.
Collective organization matters.
The same technological breakthrough can therefore support very different power structures.
The machine creates capabilities.
The surrounding institutions determine who can own them, access them, challenge them, and leave them.
Technology Can Do Both
Technology redistributes power when it lowers the cost of doing something that previously required a powerful intermediary.
It centralizes power when the new capability becomes dependent on infrastructure, networks, or resources controlled by actors that cannot easily be replaced.
Modern digital systems frequently do both simultaneously.
That is why technological progress cannot be evaluated merely by counting how many people receive access.
We must also ask who controls the layer beneath that access.
And artificial intelligence makes this question unusually urgent.
If AI gives individuals and small organizations capabilities previously reserved for large institutions, it could become one of history's strongest technological counterweights.
But if its most important infrastructure remains concentrated, AI could strengthen the very forms of corporate power it appears capable of challenging.