When Intelligence Becomes Abundant, Ownership and Access May Determine Power
1. When Knowledge Stops Being the Advantage
For much of modern history, knowledge has been a source of economic power because it has been difficult to acquire. A person who understood mathematics, engineering, finance, medicine or law possessed an advantage over someone who did not. Education converted knowledge into expertise, expertise into productivity, and productivity into income. Institutions that could reliably produce expertise became gateways to opportunity.
Artificial intelligence is beginning to alter this arrangement. Knowledge that once required years of study to access can now be explored through a conversation with an AI system. A small business owner can obtain assistance with market research, a student can receive personalised explanations, and an individual without formal technical training can begin developing software. The results are not always correct, and using a tool effectively still requires judgment, but the cost of accessing sophisticated assistance is falling.
The change is deeper than a new way to search for information. AI can increasingly help people apply what they know. It can write, calculate, analyse, design, translate and generate code. In many situations, the distance between having an idea and producing a workable first version is shrinking.
What happens when the ability to access intelligence is no longer the principal advantage? If millions of people can obtain comparable assistance, the decisive difference may shift towards what they can do with it. Knowledge will still matter, but its economic value will depend increasingly on the resources, opportunities and institutions through which it can be applied.
2. The Commoditisation of Intelligence
Consider two aspiring software entrepreneurs. Both have access to an AI assistant that can generate code, explain technical problems and help build a prototype. In an earlier era, the person with the stronger programming skills might have enjoyed a decisive advantage. Today, that advantage may be smaller, depending on the complexity of the product and the quality of the AI assistance.
Yet the two entrepreneurs may still have radically different prospects. One has savings, industry contacts, access to customers and enough time to experiment. The other has a full-time job, little capital and no reliable route to market. Their access to intelligence may be similar, but their capacity to convert it into a successful business is not.
The same pattern can emerge in education, research, consulting and creative work. AI can help more people produce a respectable first draft, analyse a dataset or develop a prototype. It does not automatically give them the judgment to identify the right problem, the resources to test a solution, the credibility to persuade customers or the distribution needed to reach a market.
This is how competitive advantage can migrate. When a capability becomes easier to reproduce, possessing that capability may no longer be sufficient to distinguish its owner. Greater importance can accrue to choosing the right problem, evaluating the output, combining it with other resources and capturing the value created.
The implication is not that expertise becomes worthless. Difficult problems still demand deep understanding, and AI-generated work can fail in ways that only an expert will recognise. Rather, expertise may become less valuable as a standalone possession in some fields, while judgment, accountability and the ability to mobilise resources become more valuable.
3. The Falling Cost of Intelligence
The decline in AI costs is already measurable, although no single statistic captures every kind of intelligence or task. Stanford University’s 2025 AI Index reported that the inference cost of a model performing at approximately GPT-3.5 level on a particular language benchmark fell from $20 per million tokens in November 2022 to $0.07 by October 2024—a reduction of more than 280 times. This comparison concerns a defined level of benchmark performance, not the cost of every AI task, but it illustrates the direction of travel. See the Stanford HAI AI Index Report 2025.
As AI becomes cheaper and more capable, activities that once required specialised teams may become feasible for smaller organisations and individuals. A founder may be able to perform preliminary legal or market research with AI assistance before hiring a professional. A teacher may prepare tailored learning materials without a large support staff. A researcher may explore more hypotheses before committing scarce laboratory time.
But the cost of intelligence is only one part of the cost of production. The entrepreneur still needs customers, the researcher still needs equipment, and the manufacturer still needs a factory. The teacher still needs students who can access the lessons, while the business still needs a way to deliver its product and receive payment.
In other words, lowering the cost of thinking does not eliminate the cost of acting. It can make action cheaper, but the remaining constraints may become more visible precisely because intelligence is no longer the main bottleneck.
4. The Resource Bottleneck
Imagine two people who use the same AI system to design an innovative energy product. Both receive good technical suggestions and produce promising plans. One can finance prototypes, hire engineers, obtain testing facilities and negotiate with manufacturers. The other has the same plans but cannot afford to test them.
The difference between them is not necessarily intelligence. It is the ability to turn intelligence into something that works in the physical and commercial world.
The relevant resources vary by industry. They may include capital, energy, advanced chips, computing capacity, land, machinery, laboratories, intellectual property, skilled collaborators, customer relationships, distribution channels or regulatory access. Not every business requires all these resources, and AI can reduce the amount needed in some cases. Nevertheless, many forms of value creation still depend on assets that cannot be conjured into existence by a prompt.
The infrastructure behind AI makes this distinction especially clear. The OECD’s analysis of AI infrastructure and competition describes the importance of advanced semiconductors, computing resources and complex supply chains, alongside significant barriers to entry in parts of the industry. The intelligence available through a digital interface may feel weightless, but the systems supporting it depend on physical equipment, energy, networks and investment. See the OECD’s work on competition in the digital age.
This creates a possible new asymmetry. More people may be able to access intelligence, while a smaller group of organisations retains substantial control over the infrastructure through which that intelligence is developed, delivered or monetised.
Access to a powerful tool is not the same as ownership of the productive system behind it. And the ability to create value is not the same as the ability to capture it.
5. From the Knowledge Economy to the Resource Economy
The knowledge economy is built around the creation and application of information, expertise and ideas. The resource economy, as used here, describes a possible shift in emphasis: as access to intelligence expands, control over the resources needed to apply it may become relatively more important.
This is a conceptual distinction, not a claim that one economic system will simply replace another. Knowledge will remain essential. Frontier scientific discoveries, rare technical skills, sound judgment and trustworthy expertise can still command substantial value. Nor will every resource remain scarce: AI may help discover new materials, improve energy efficiency, automate production and make existing assets more productive.
The hypothesis is about changing bottlenecks. When one constraint becomes less restrictive, another can become decisive. If intelligence is expensive, the person who possesses it has an advantage. If intelligence becomes widely available, the advantage may shift towards the person who can deploy it at scale, afford repeated experimentation, secure distribution or own the resulting asset.
The pattern has precedents. Mechanisation did not make human effort irrelevant; it changed the relationship between labour, machinery and capital. The internet made publishing and distributing information easier, but it did not distribute attention equally. Digital platforms created opportunities for small businesses while also enabling some companies to accumulate enormous audiences, data advantages and market power.
AI could produce a similar tension. It may lower barriers to participation while increasing the rewards available to those who control scarce complementary resources. The outcome will depend on technology, competition, public policy and the ways people organise ownership—not on technological progress alone.
6. The Education Paradox: When Learning Becomes Abundant
Education provides a revealing test of this argument. If AI can explain a mathematical concept, generate practice questions, diagnose a student’s errors and adapt an explanation to the learner, what exactly is a prestigious educational institution selling?
It is tempting to answer that it is selling knowledge. Historically, that was a large part of its value: access to expert teachers, organised curricula, libraries, laboratories and a recognised qualification. But if high-quality explanations become easier to obtain outside a classroom, the institution’s value proposition may shift.
A university can still provide rigorous assessment, mentorship, research facilities, disciplined study, a community of ambitious peers and a recognised credential. It can connect students to employers, investors and collaborators. It can place a young person in an environment where useful relationships form naturally, through shared work and repeated contact. These benefits are harder to reproduce simply by asking an AI to teach a subject.
The result is a paradox. As instruction becomes more accessible, the scarcity value of the institution may increasingly reside in the things surrounding instruction. The classroom may be only one component of what students are paying for. The rest may be selection, reputation, social capital, access and the opportunities that follow from belonging to a particular network.
This does not mean every prestigious university provides good value, or that a conventional degree will always outperform alternative routes. AI can help independent learners build skills, demonstrate competence and connect with communities beyond traditional institutions. Employers may also develop more effective ways to assess what a person can actually do.
But the question for universities becomes harder to avoid: if the teaching itself can be obtained elsewhere, what justifies the institution’s price, selectivity and social prestige? Its answer may increasingly depend on what it controls beyond knowledge.
7. Oxford, MIT and the Nature of a Moat
Consider the difference between the University of Oxford and the Massachusetts Institute of Technology. Both possess exceptional reputations, but their institutional strengths are not identical. Oxford draws on centuries of academic history, a distinctive collegiate structure and extensive international networks. MIT is strongly associated with engineering, scientific research, technological innovation and close connections between research and industry.
It would be simplistic to predict that AI will necessarily favour Oxford over MIT, or the reverse. Their futures will depend on how they adapt, the fields in which they remain strongest, the quality of their communities and the opportunities they create. Both institutions can benefit from AI while also facing pressure to justify the cost and structure of traditional education.
The comparison nevertheless exposes an important distinction. If knowledge delivery becomes easier to reproduce, an institution’s durable advantage may depend less on the information it provides and more on the ecosystem it sustains: researchers, laboratories, funding, reputation, alumni, collaborators and access to opportunities that are difficult to assemble independently.
Financial resources matter here, though they do not guarantee excellence. Harvard University reported an endowment of $56.9 billion at the end of fiscal year 2025. That figure illustrates the scale of accumulated institutional capital available to support a university’s long-term activities; it does not, by itself, establish the quality of its teaching or the value of any particular degree. See Harvard University for institutional reporting.
A university’s moat, therefore, may not be its exclusive possession of knowledge. It may be the difficult-to-reproduce combination of people, resources, reputation and relationships that turns knowledge into opportunity.
The same principle extends beyond education. A company may use the same AI models as its competitors but retain advantages through customer trust, manufacturing capacity, proprietary data, logistics or distribution. As the intelligence layer becomes easier to access, the surrounding system may matter more.
8. From Elite Education to a Technologically Empowered Elite
This raises a more unsettling possibility. If intelligence becomes widely available but the means of applying it remain concentrated, could AI strengthen the position of an already advantaged minority?
Yuval Noah Harari has warned about the possibility of a technologically empowered elite, particularly if advances in AI and biotechnology allow some people or institutions to acquire capabilities that others cannot access. His warning should be treated as a scenario to examine, not as a settled prediction. Technology does not determine a single political or economic outcome, and today’s inequalities do not prove that a permanent superhuman elite will emerge.
Still, the underlying question deserves attention. Imagine a group that controls substantial capital, advanced computing infrastructure, important intellectual property and powerful distribution networks. AI could allow that group to coordinate complex operations, conduct research, automate work and make decisions at a scale that would previously have required much larger organisations.
Now imagine millions of people with access to equally impressive AI assistants but limited ownership of the businesses, infrastructure and assets that produce the resulting wealth. Their individual capabilities may rise dramatically without a corresponding increase in their bargaining power.
That is the central tension. A person may become more capable without becoming more powerful in relation to the systems on which they depend. Greater productivity does not guarantee greater income, just as access to a market does not guarantee ownership of the market.
Yet the opposite outcome is also possible. AI may enable small teams to challenge established companies, help individuals build profitable enterprises and make specialised services accessible to people who were previously excluded. Whether AI concentrates or distributes economic power will depend partly on who can access the necessary resources, how competitive markets remain, and whether new forms of ownership emerge.
The question is not simply whether AI will make people more intelligent. It is whether the gains from that intelligence will be broadly shared or disproportionately captured by those who already control the means of production.
9. The Ownership Problem
The difference between access and ownership becomes clearer when we look at ordinary examples. Knowing how to design a factory is not the same as owning one. Having an AI assistant that can analyse a market is not the same as owning a company with customers. Being able to generate a product design is not the same as having the capital, equipment and distribution needed to manufacture and sell it.
Digital work complicates the picture because it can sometimes be produced with relatively little physical infrastructure. An individual may use AI to create software, publish a book, provide a specialised service or build a small online business. These are genuine routes to independence, and they should not be underestimated.
But even digital entrepreneurs rely on systems they may not control: cloud platforms, app stores, payment processors, search engines, advertising networks and AI services. A platform can change its prices, rules or visibility mechanisms. A business that owns its customer relationships and intellectual property may have more room to manoeuvre than one that depends entirely on a third party.
This does not mean platform dependence is inherently bad. Platforms can make capabilities affordable and give small businesses access to infrastructure that would otherwise be impossible to build. The problem arises when access is mistaken for control, or when the ability to create value is confused with the ability to retain its benefits.
For individuals, the practical implication is to think beyond the acquisition of skills. Skills matter, but so do ownership stakes, savings, customer relationships, intellectual property, distribution, reputation and the ability to move between providers. Not everyone can own a data centre or a factory, nor should they need to. But a wider distribution of productive assets and genuine routes to market could matter enormously in an AI-intensive economy.
The larger social question is whether people will participate in the value created by increasingly capable systems—or remain users of systems whose most important assets belong to others.
10. The Counterargument: AI Could Democratise Resources Too
There is a serious objection to the resource-economy thesis. AI may not merely make intelligence cheaper; it may reduce the cost of acquiring and using other resources as well. Better systems could improve the efficiency of energy grids, accelerate scientific discovery, optimise logistics, lower the cost of software development and help small manufacturers compete with larger firms. AI-enabled robots may eventually make some forms of physical production more accessible, although their deployment will still depend on capital, safety, infrastructure and the economics of particular tasks.
If these changes are sufficiently broad, the bottlenecks described in this article may weaken rather than intensify. Small teams could gain capabilities once reserved for large organisations. New companies could emerge faster, and more people could own productive businesses. Open models, competitive markets, public infrastructure and cooperative ownership could all help spread the benefits.
Nor is the concentration of resources inevitable. Governments can shape competition and access; institutions can share infrastructure; investors can fund new entrants; and societies can experiment with broader forms of ownership. The distribution of AI’s benefits will be influenced by these choices as much as by the capabilities of the technology itself.
There is also a danger in treating resources as a single category. Capital, energy, land, computing, reputation and human relationships behave differently. Some can be expanded, some can be shared, some are geographically constrained, and some lose value as technology changes. The future will not necessarily reward every owner of today’s scarce assets.
The argument, therefore, is not that the resource economy must arrive or that knowledge will cease to matter. It is that as the cost of intelligence falls, control over the remaining bottlenecks deserves greater attention. The question is whether technology will make those bottlenecks easier for more people to overcome—or allow their owners to extract a growing share of the value created.
11. Intelligence for Everyone, Power for Whom?
AI is changing the relationship between knowledge and economic opportunity. It can help people learn faster, produce more, explore unfamiliar fields and attempt work that previously required specialist assistance. These are substantial gains, particularly for those who have historically lacked access to expert guidance or expensive services.
But access to capability is only one part of economic freedom. People also need ways to apply that capability, reach customers, finance experiments, own assets and share in the value they create. Without those pathways, a society can become more intelligent without becoming more equal.
The emerging divide may not simply be between people who understand AI and people who do not. It may increasingly separate those who can use intelligence to build, own and direct productive systems from those who can use it only on terms set by others. That divide is neither fixed nor inevitable, but it is too consequential to ignore.
Education will need to respond by teaching more than the accumulation of information. Individuals may need to combine knowledge with judgment, adaptability, relationships and an understanding of ownership. Institutions will need to justify their value through the opportunities and capabilities they create, not merely the information they transmit. Policymakers will need to consider whether the infrastructure behind AI remains competitive and whether the gains from automation reach beyond its owners.
The knowledge economy is not disappearing. It is being reshaped by a technology that makes some forms of intelligence easier to obtain while leaving other constraints stubbornly in place. The decisive question may therefore shift from who knows the answer to who has the means to act on it—and who ultimately owns the result.
AI may make intelligence more democratic without making power more democratic. Whether those two forms of democracy eventually converge will depend not only on what AI can do, but on how people choose to distribute the resources, opportunities and ownership that make its capabilities matter.
When everyone can access intelligence, who will own the future that intelligence creates?
Read next: The AI Heat Hypothesis: Every Moat Will Melt explores how AI may erode competitive advantages built on reproducible expertise and execution.