The current frenzy is not merely a "bubble" in the sense of irrational exuberance over a new gadget. Instead, it is the apotheosis of a deeper, more systemic phenomenon: extractive capitalism. While the technology of large language models and generative neural networks demonstrates genuine, albeit uneven, utility in scientific research, coding, and administrative efficiency, the economic framework surrounding it has shifted from a model of wealth creation to one of wealth extraction. The primary driver of today's massive investments is not a collective delusion about the immediate profitability of AI chatbots. Rather, it is a calculated, defensive maneuver by incumbent firms to secure their positions as the permanent landlords of the digital world. By monopolizing the scarce resources of compute, energy, and data, these entities are ensuring that any future productivity gains are captured as rents rather than shared as widespread economic growth.
The architecture of rentier capitalism
To understand why the "bubble" narrative is insufficient, one must first distinguish between profit-seeking and rent-seeking. In the classical sense, profit is the reward for taking risks and creating new value—making a better mousetrap or a more efficient engine. Rent, by contrast, is the income derived from the control of a scarce asset. As David Ricardo and later Gordon Tullock observed, rent-seeking involves manipulating the economic or political environment to grow one’s wealth without creating any corresponding new wealth for society. It is, in effect, a "tax" on the productive sector by those who control the "toll bridges" of the economy.
The modern technology sector has undergone a profound "rentierization". Since the 2008 financial crisis, the shift from neoliberalism—which ostensibly promoted free markets—to rentier capitalism has accelerated. In this system, firms do not compete solely on the quality of their products but on their ability to enclose the digital commons, secure intellectual property monopolies, and lock users into proprietary infrastructures. AI represents the ultimate frontier of this enclosure. The staggering capital outlays are not just for research and development; they are the "entry fees" for a new era of digital feudalism.
The political economy of "extractivism" was historically associated with the Global South’s dependency on raw material exports. Rents from natural resources like oil or minerals allowed local elites to maintain power while obstructing structural change. Today, this logic has been imported into the core of the global tech economy. Data is the new oil, not because it flows naturally, but because it is being extracted, commodified, and enclosed at an unprecedented scale.
The trillion-dollar moat
The sheer scale of AI infrastructure spending is a testament to the "moat-building" strategy of Big Tech. Goldman Sachs notes that consensus estimates for AI capital expenditure consistently underestimate the actual spend, which has exceeded 50% year-over-year growth for two consecutive years. By 2026, annual investments could surpass $500 billion, with a cumulative total potentially exceeding $7 trillion by 2030.
This expenditure is directed at three critical bottlenecks: high-performance semiconductors (primarily Nvidia’s GPUs), massive data center capacity, and the electricity required to run them. Control over these inputs creates a barrier to entry that is insurmountable for all but a few firms. In this context, the "revenue gap"—the $600 billion discrepancy between infrastructure spend and AI-driven income—is not a sign of failure but a feature of the system. It is a massive "lump sum cost" that ensures only the wealthy can survive as rent seekers.
The dynamics of this investment are increasingly "circular." In a pattern eerily reminiscent of the 1990s telecom boom, tech giants are investing in or financing startups that later become their primary customers. For instance, Microsoft's multi-billion-dollar investment in OpenAI is largely recouped through OpenAI’s commitment to spend billions on Microsoft’s Azure cloud services. This creates a closed loop where capital stays within a handful of firms, inflating their reported earnings while masking the lack of broader market adoption.
The finance of fragility
The current model of financial capitalism relies on the exponential growth of debt to drive up asset prices. This creates a feedback loop where rising valuations require yet more borrowing to sustain. In the AI sector, firms are turning to leasing and new debt to maintain the "compute race" as free cash flow is devoured by relentless CapEx.
The risk of this model is that it becomes unsustainable if the underlying productivity of the rest of the economy does not catch up. As Michael Hudson argues, when debt grows faster than income, it leaves less and less disposable income for goods and services, eventually leading to a crash. The AI infrastructure boom is being built on the assumption that "Artificial General Intelligence" (AGI) will eventually emerge and justify the cost. If AGI remains a distant mirage, the "silicon mountain" of debt could collapse, triggering a crisis that far exceeds the dot-com bust.
The new enclosures of the digital commons
If the first stage of capitalism was the enclosure of the land, the current stage is the enclosure of the "data commons". For decades, the public web functioned as a shared repository of human knowledge, symbols, and creativity. AI firms have treated this commons as a "free" raw material, scraping billions of data points to train their models without the consent or compensation of the creators.
This process mirrors Karl Marx’s "primitive accumulation"—the historical process of divorcing producers from their means of production. In the digital context, the "means of production" is the collective intelligence of humanity. By transforming this shared heritage into proprietary algorithms, tech giants are engaging in a new form of "cognitive colonialism".
The mechanics of enclosure
1. Extraction: The large-scale scraping of the web, treating human experience as a harvestable resource.
2. Fencing: The use of licensing deals, paywalls, and "walled gardens" to prevent others from accessing the same high-quality data.
3. Monetization: Selling access to the resulting models back to the very public whose data was used to build them.
This enclosure creates a "walled garden" effect that stifles innovation. Small developers and academic researchers are increasingly priced out of the market as the "open web" disappears behind licensing walls. Furthermore, copyright law is being recalibrated to serve the interests of the AI firms rather than the creators. While some propose statutory licensing to ensure creators are compensated, the current trend is toward a "pay-to-play" model that benefits the most capitalized firms.
The invisible proletariat of AI
Behind the sleek, automated interfaces of ChatGPT and Gemini lies a vast, hidden workforce. The perception that AI is a purely machine-driven miracle is a carefully maintained illusion. In reality, these systems require millions of human workers to label, annotate, and clean the data—a process known as data work or "ghost work".
This workforce is primarily concentrated in the Global South—countries like Kenya, Nigeria, the Philippines, and India—where labor is cheap and regulatory oversight is minimal. This is the "hidden infrastructure" of the AI revolution, characterized by precarious contracts, sub-minimum wages, and psychological trauma.
The exploitation of this "invisible proletariat" is not an accidental byproduct of AI development but a fundamental requirement of its extractive logic. By distancing themselves through opaque subcontracting chains and third-party vendors (like Sama or Scale AI), tech giants avoid accountability for working conditions while reaping the benefits of low-cost data.
This model of "digital colonialism" breeds resentment and undermines the long-term stability of the U.S. AI supply chain. Geopolitical rivals like China are increasingly positioned as alternatives for data-rich nations that feel exploited by American firms. Without reform, the extractive nature of AI labor could lead to "data embargoes" and a loss of the very human diversity that makes AI models useful.
The ghost in the machine: Productivity myths
The central promise of AI is a massive lift in global productivity. Projections from the IMF and others suggest that AI-related investment is already fueling GDP growth in the United States, accounting for a significant share of capital outlays in 2025. However, at the aggregate level, the "productivity paradox" remains unresolved. While micro-level studies show impressive gains—for example, a 56% increase in speed for programmers using GitHub Copilot—these have not yet translated into measurable macroeconomic improvements.
The Berkeley research into the "myths" of AI productivity provides a sobering corrective. The belief that AI reliably boosts productivity across all contexts is often undermined by "code-quality regressions" and the "cognitive overhead" of fact-checking AI output. For complex, context-heavy tasks, AI "help" can actually increase project completion time by 19% among experienced developers.
The skill compression effect
AI is not a "blanket enhancer" but a "targeted accelerant". Its most consistent impact is "skill compression": narrowing the gap between high- and low-performing workers.
Novices vs. Experts: Studies in customer support and software development show that bottom-quartile performers see throughput lifts of up to 35%, while veterans see little to no gain.
Task Redistribution: AI substitutes for routine clerical activities while complementing high-skill analytical work, potentially depressing wages at the middle and bottom as capabilities become diffuse.
Automation Bias: High levels of automation can increase commission errors by 12% as users become over-trusting of imperfect systems.
This compression is a dream for extractors. By making workers more interchangeable, AI reduces the bargaining power of specialized labor. The premium once captured by experienced professionals—the "knowledge class"—is being eroded, allowing tech platforms to capture that value as rent.
Historical echoes: From iron rails to silicon mountains
The current AI boom finds its most striking parallel in the "Railway Mania" of the 1840s. During this period, the British economy was gripped by a speculative frenzy over the locomotive. Parliament authorized thousands of miles of track, much of it redundant or fraudulent, funded by middle-class investors looking to get rich quick.
The railway bubble burst spectacularly, ruining thousands. Yet, it left behind a functional transportation network that permanently lowered the cost of moving goods and people, facilitating the next stage of the Industrial Revolution. AI infrastructure may follow a similar path. The "overbuild" of data centers and the massive investment in power grids—driven by the fear of being left behind—will remain even if the current crop of AI companies fails.
The critical difference is the concentration of power. In the 19th century, the railway network was eventually brought under more thoughtful government control. Today, the "silicon mountain" is being constructed by a handful of private firms that are actively working to capture the regulatory process itself.
The utility extraction: Shifting costs to the public
The extractive nature of AI capitalism is perhaps most visible in its relationship with the physical infrastructure of electricity. Data centers are ravenous consumers of power; the sudden surge in demand is driving utility proposals that break the traditional mold of utility rates.
In states across the U.S., utilities are striking "secret contracts" with Big Tech firms for preferential energy rates. These deals often shift the cost of grid expansion onto ordinary ratepayers. In Georgia, residential rates reportedly increased by 37% to accommodate data center demand. This is a literal extraction of wealth from the public to subsidize the CapEx of the world's most valuable companies.
Furthermore, the surge in electricity demand creates structural inflationary pressures. If AI adoption remains concentrated among a few hyperscalers, the returns may plateau while the "infrastructure tax" on the rest of the economy continues to rise. The Federal Reserve now faces a landscape where headline data may mask these structural shifts, leading to policy errors that favor the rentiers at the expense of the productive sector.
The mirage of AGI and the final enclosure
The ultimate justification for the trillion-dollar build-out is "Artificial General Intelligence" (AGI)—a machine capable of matching or exceeding human cognition across all domains. Within the extractive framework, AGI is the "Final Empire": the ownership of the means of thinking.
If the mission of educating the next generation loses its economic rationale because "thinking" is a service owned by Microsoft or Google, the very legitimacy of our social institutions—universities, expert bureaucracies, and the credentialing system—comes into question. This is not a "bubble" of technology; it is a crisis of ownership.
The Substacker AZ Mackay notes that most nations will never build sovereign AI infrastructure because the cost is measured in "decades of technical talent development" and "rare earth mineral control". Smaller nations are thus becoming dependent on "cognitive infrastructure" they do not own, subjecting them to the caprice of a few business leaders in Silicon Valley. This is the endgame of extractive capitalism: the transformation of human judgment into a rent-generating asset.
Reclaiming the generative future
The "bubble" narrative is a distraction. If we focus only on when the stock prices of Nvidia or Microsoft will fall, we miss the more important battle over the structure of our economy. AI has the potential to be a "generative" technology—one that empowers individuals, accelerates scientific discovery, and solves complex social problems. But this requires a fundamental break from the extractive model.
Towards a generative AI policy
To redirect the trajectory of AI away from rentierism, a new policy framework is required:
1. Antitrust as a Tool for Productivity: Regulatory focus should move beyond simple consumer welfare to address "infrastructure capture". The goal must be to prevent cloud providers from using their dominance to stifle rivals and lock in users.
2. Data as a Commons: We must reject the "enclosure of the commons". This could involve statutory licensing models that compensate creators or the creation of public-interest data repositories that are accessible to all, not just the hyperscalers.
3. Decoupling Utilities from Extraction: Public Utility Commissions (PUCs) must be empowered to end "secret contracts" and ensure that Big Tech pays its fair share of the grid’s expansion.
4. Ethical Supply Chain Mandates: Following the model of the CHIPS Act, government funding and licenses should be tied to "Fair Labor Certifications". This includes living wages and mental health support for the Global South workers who make AI possible.
5. Taxing Rents, Not Work: If AI-induced productivity gains are concentrated in capital, a shift toward wealth taxes or "robot taxes" may be necessary to sustain the social contract as labor’s share of national income declines.
The choice before us
We are at a "Mythos" moment—a turning point where the power of technology is outstripping our ability to govern it. The current AI boom is "overbuilt" on a foundation of debt, extracted data, and exploited labor. This is the true bubble: the belief that we can sustain a modern economy on the logic of 19th-century extractivism.
The "silicon mountain" is already here. Whether it becomes a tomb for our collective economic future or the foundation for a new era of shared prosperity depends on our willingness to challenge the rentiers. AI is not the bubble. Extractive capitalism is. And like all extractive models, it will eventually hollow out its own host unless it is restrained by democratic power and the reclamation of the commons.
As the Federal Reserve and global policymakers scramble to measure the "intangible capital" of the AI age, they must look beyond GDP and CapEx. They must look at the distribution of power, the security of our utilities, and the dignity of the workers in the shadows. The real crash to fear is not a 20% correction in tech stocks, but the permanent "enshittification" of the human intellect by a system designed to extract every last drop of value from our collective mind.