The Commoditization Trap: Why AI Efficiency May Devour Its Own Market

The Central Provocation

AI is doing something qualitatively different from previous efficiency revolutions. It is not merely making cognitive labor cheaper. It is progressively substituting for the very class of human workers whose incomes constitute the demand for the services AI is replacing. This is not a temporary displacement. It is a structural commoditization that, if it proceeds far enough, undermines the economic basis of the market it claims to serve.

The Jevons Paradox, which holds that efficiency improvements increase total resource consumption rather than decrease it, is the standard techno-optimist defense against this concern. When steam engines became more efficient, coal consumption rose. Cheaper energy per unit meant more units demanded. Applied to AI: as intelligence becomes cheaper, demand for it will expand exponentially, generating more economic activity, not less.

The paradox holds when efficiency improvements open genuinely new categories of demand. The key condition is that the underlying resource is an input into production, not a substitute for the consumer of the product. When efficiency improvements eliminate the economic actor who generates the demand, the paradox breaks down. That is precisely what is happening now, and the consequences run from individual balance sheets to the financial architecture of the entire AI infrastructure boom.

The Commoditization Cascade: Legal Services as a Case Study

To understand the cascade structurally, follow one industry through it completely.

In 2020, a mid-sized American law firm billed associate time at $200–$350 an hour for contract review, legal research, due diligence, and drafting. These were not artificially high prices. They reflected genuine scarcity: a J.D. required three years of postgraduate education, bar passage, and several years of supervised practice before an associate could work independently. The knowledge was real, the training costs were high, and the supply was constrained by institutional gatekeepers. The American legal services market was valued at approximately $350 billion annually.

The AI incursion followed a predictable sequence. First came augmentation: tools like Kira Systems and Luminance (2016–2019) allowed junior associates to review contracts faster. Firms adopted them and passed some savings to clients while capturing the rest as margin. Headcount in document review, previously the entry point for thousands of new lawyers, began falling. Then came substitution: by 2023, large language models could perform first-pass contract review, legal research, and routine drafting at a quality firms had to acknowledge was commercially acceptable for a wide range of matters. Clients, now aware of the capability, began demanding that AI-assisted work be billed at lower rates. Several large corporate legal departments brought work in-house using AI tools directly. By 2025, major firms were reporting that associate leverage ratios, the number of junior lawyers billing under each partner, were falling for the first time in the industry’s modern history.

The price progression is the key data. Tasks that were billed at $300 an hour in 2020 now bill at $150 where AI assistance is disclosed, and sophisticated clients are beginning to treat certain categories of legal work as near-commodity services to be put out to competitive tender. Specialized legal AI platforms offer subscription access to capabilities that would have required a team of associates for a flat monthly fee that represents a fraction of a single associate’s salary. The endpoint of this progression is not difficult to project: routine legal services become a software subscription. The $350 billion market does not disappear, the underlying need for legal work is real and persistent, but the revenue per unit of work collapses. The profession restructures around a much smaller number of high-judgment practitioners and a commoditized software layer. The middle, where most lawyers actually worked, is gone.

This is not a legal industry story. It is the template. Financial analysis, software development, medical coding, architectural drafting, accounting, consulting: each follows the same logic, at roughly the same pace, because each is a domain of structured cognitive work that AI targets with the same underlying capabilities. The cascade is not moving through one industry at a time. It is moving through all knowledge-intensive industries simultaneously, and the revenue compression in each feeds deflationary pressure across the whole system.

The Infrastructure Valuation Problem

This cascade creates a direct contradiction at the heart of the AI investment thesis.

The financial case for hyperscale data centers rests on a projection of sustained, high-margin demand for AI services. But that projection embeds the assumption that AI-delivered cognitive services will remain premium-priced. The commoditization cascade demonstrates they will not. Revenue per unit of compute collapses even as the volume of compute consumed may grow, precisely the dynamic that destroyed telecommunications infrastructure valuations in 2001. Fiber optic capacity expanded exponentially. The price of transmitting a bit of data fell by 99%. Revenue collapsed even as usage exploded. The infrastructure was not wrong; the financial model financing it was.

More fundamentally, the framing of AI value as residing in computational capacity is itself a category error. The end product is not processing cycles. It is the successful completion of a cognitive task: a contract reviewed, a drug interaction identified, a portfolio rebalanced, a piece of software debugged. The data center is a supplier to that product, not the product itself. And suppliers in competitive markets are systematically squeezed toward their cost of production.

The incentive structure of the industry drives relentlessly in one direction: perform the task at maximum proficiency with minimum energy and cost. Every algorithmic innovation that reduces compute requirements, model compression, distillation, more efficient architectures, edge deployment, is a direct attack on the revenue model of the infrastructure layer. DeepSeek’s demonstration that frontier-level reasoning could be achieved at a fraction of the compute cost of its predecessors was not a technical anomaly. It was a preview of the secular trend that will define the next decade of AI economics. The hyperscalers are financing the research that will eventually make their infrastructure less necessary.

Abundance Without Freedom: The Human Agency Deficit

The optimistic response to all of this is the abundance argument: as AI makes cognitive services cheaper, everyone benefits. Healthcare becomes more accessible. Education becomes personalized. The deflationary logic is presented as unambiguously positive, a rising tide of cheap intelligence lifting all boats.

This argument contains a hidden assumption: abundance generates human welfare only if people are actually free to benefit from it. And freedom, as the historian Timothy Snyder argues, is not merely the absence of tyranny. It is the presence of the material and institutional conditions that allow people to exercise genuine agency. A person crushed by medical debt is not free, regardless of how cheap AI diagnostics become. A family in substandard housing cannot take advantage of AI-enabled educational tools. A worker without reliable transportation cannot access the abundance that AI-optimized supply chains theoretically provide.

The “No Next Layer” Problem

There is a standard rebuttal to displacement arguments that this paper’s own analytical framework is positioned to answer directly. The rebuttal holds that creativity, judgment, relational intelligence, and physical presence will remain distinctly human domains, and that the economy will reorganize around these capacities as it has reorganized around new human strengths after every previous wave of automation. In this view, AI is just another productivity tool that expands the economic pie while shifting which human skills command premiums.

The question is not whether such domains persist. They do, and will for the foreseeable future. The question is whether they generate enough economic value at scale to sustain a middle class, whether the number of people who can command a living wage from distinctly-human cognitive work is sufficient to maintain the consumer demand that market economies require to function.

Previous automation waves displaced physical and routine clerical labor while expanding demand for cognitive labor. The historical bargain was that automation made goods cheaper while creating new, higher-paying jobs for workers who could manage, design, and improve the automated systems. The net effect was rising real incomes and expanding consumer markets. AI breaks this bargain because it targets the cognitive layer that absorbed displaced workers from previous waves. Creativity and judgment are real human advantages. They are also, almost by definition, scarce, and an economy whose middle class is composed entirely of people competing in scarce creative and relational domains is not a stable middle class. It is an economy with a very wide base of routine workers, a very narrow peak of creative professionals, and a collapsing middle. The abundance generated at the top does not automatically flow to the base. Without institutional design, it does not flow there at all.

The Freedom Pool: AI as an Amplifier of Human Freedoms

What is required is not simply a redistribution mechanism but a purposive redeployment of AI’s deflationary power toward the specific domains where the absence of infrastructure most constrains human freedom. Call this the Freedom Pool: a structural commitment to direct the surplus generated by AI efficiency toward eliminating the material impediments that prevent people from becoming their better selves.

The concept draws directly from Snyder’s argument that genuine liberty requires positive conditions, not merely negative ones. Freedom is not just freedom from, from coercion, from arbitrary power. It is freedom for: for meaningful work, for health, for learning, for participation in civic and cultural life. The great unfreedoms of contemporary life are largely material: unaffordable housing, inaccessible healthcare, underfunded education, unreliable transportation, contaminated water, inadequate nutrition. These are structural constraints that determine, more than almost anything else, what a person’s life can actually become. Applied with this purpose explicitly in mind, AI is uniquely suited to attacking each of these constraints simultaneously, not because AI is a magic solution, but because the scale of cognitive work required to optimize these complex systems has historically exceeded the capacity of human institutions to manage them.

The Freedom Pool differs from a Universal Basic Income in its purpose and structure. Cash transfers address the income dimension of displacement. The Freedom Pool addresses the capability dimension: the actual conditions that determine what people can do with their lives. Rather than putting money in people’s hands and trusting markets to translate it into flourishing, it uses the power of AI to target specific infrastructures of freedom directly: the housing, healthcare, education, clean water, nutrition, and mobility that allow a person to exercise meaningful agency regardless of where they were born or what market forces have done to the sector they once worked in.

The Electrification Precedent

The historical analog that clarifies both the problem and the solution is electrification, and specifically, the institutional architecture required to make it universal.

In the late nineteenth century, electricity generation was a high-margin business dominated by a few innovative companies with proprietary technology and enormous capital advantages. Within decades, competition, standardization, and regulatory intervention had converted it into a commodity service priced at cost. The economic value of electricity was not captured by electricity companies. It was captured by the industries and consumers who used cheap electricity to do things that would otherwise have been impossible.

But the distribution of that value was not left to markets alone. The Rural Electrification Administration, established in 1936, extended electrical infrastructure to the rural communities that private utilities had no financial incentive to serve. Regulated utilities were given monopoly franchises in exchange for universal service obligations. Public utilities commissions set rates to balance investment recovery with broad accessibility. The result was that the transformative power of electricity reached the people whose lives most needed transforming, not because electricity companies were philanthropic, but because institutions were designed to make universal access a structural requirement rather than a market outcome.

AI requires an analogous institutional architecture. The Freedom Pool is the mechanism by which the deflationary surplus of AI is converted into the enabling conditions for human agency. The entities that capture that surplus, technology companies, AI infrastructure owners, firms that deploy AI to reduce labor costs, are the natural contributors to a pool of capital and capabilities, just as regulated utilities were required to cross-subsidize rural service from urban revenues. A tax on automated labor substitution, a sovereign seed fund financed by returns on publicly-funded AI research, or data dividends compensating individuals for the training data that made large AI systems possible are all viable funding mechanisms. Each creates a direct institutional link between the source of the deflationary gain and the freedom infrastructure it is obligated to sustain.

Intelligence Infrastructure as an Accelerant of Positive Human Freedom

AI is on a structural trajectory toward becoming a low-margin cognitive utility, the electricity of the mind. Like electricity, its transformative power lies not in the generation infrastructure but in what it enables when reliably and universally available. Like electricity, that universal availability will not emerge spontaneously from market mechanisms. It requires deliberate institutional architecture whose purpose is explicit: not efficiency for its own sake, but freedom as the destination.

The Jevons Paradox assumes an expanding universe of demand created by expanding human capacity to use cheap resources. That assumption held as long as humans remained the primary agents of economic production. What AI introduces is the possibility of abundance produced without broad human agency: intelligence so cheap it can do almost anything, in a society where the structural conditions for human flourishing have been allowed to erode.

The Freedom Pool is the answer to that possibility: a commitment to use AI’s deflationary power not merely to make things cheaper, but to make people freer. The alternative, an AI economy that generates extraordinary aggregate wealth while concentrating freedom in the hands of those already positioned to capture it, is not merely unjust. As the structural analysis above demonstrates, it is economically unstable. Markets require consumers. Democracy requires citizens. Freedom requires the material conditions within which it can actually be exercised.

AI that destroys those conditions in the process of generating its surplus is not building the future. It is undermining the foundations of the civilization it purports to serve. The resolution of that paradox is not a technical problem. It is a political commitment: to the proposition that intelligence, like electricity before it, should be the infrastructure of human freedom rather than its replacement.