Core Thesis
The AI revolution is the clearest example of Hoarder Capitalism—a system where productive assets (data, chips, and models) are stockpiled and fenced off, rather than circulated for the public good. The problem isn’t that tech companies are successful; it’s that they are converting society’s collective knowledge into private tollbooths. Unless we change the rules, AI won’t create widespread abundance. It will create a new digital feudalism where everyone rents their intelligence from a handful of trillion-dollar lords.
Key Arguments
1. The Data Heist: Enclosing the Public Commons
AI models learn by digesting the sum of human writing, art, and conversation—much of which was shared freely on the open internet.
- The Problem: A few companies vacuumed up this public commons, trained private models on it, and now charge us for access to the intelligence we collectively generated.
- Concrete Example: OpenAI and Google scraped vast portions of the internet (including Wikipedia, forums, and published books) to build ChatGPT and Gemini. Now, they charge $20–$30/month for advanced access, effectively selling humanity’s aggregated voice back to itself.
- Simplified Claim: It’s like a landlord taking all the books from a public library, burning the originals, and then charging you a rental fee every time you want to read a photocopy.
“The crowd built the library. The corporation took the books. Now the crowd pays to read them.”
2. The Compute Monopoly: GPUs as the New Feudal Land
To build a cutting-edge AI, you need thousands of specialized microchips (GPUs), costing billions of dollars.
- The Problem: Only Big Tech (Microsoft, Google, Amazon, Meta) can afford this fortress of chips. Universities, startups, and poorer nations are locked out of the foundational layer of AI development.
- Concrete Example: A single Nvidia H100 GPU costs around $30,000. To train GPT-4, you need roughly 25,000 of them. That’s a $750 million hardware investment. A university lab simply cannot compete—they are forced to rent processing power from Amazon or Microsoft at high markups.
- Simplified Claim: It doesn’t matter if you have the best idea for an AI. If you can’t buy the “land” (the chips) to build it on, you are a tenant paying rent to the digital lords.
“The hoarders don’t need to win every market. They just need to own the gate everyone has to walk through.”
3. The Labor Squeeze: Automating Jobs, Hoarding the Savings
Companies use AI to replace human workers, but the profits from these efficiency gains are not shared with the workforce.
- The Problem: When AI does the work of 10 people, those 10 people lose their jobs, and the surplus value goes straight to shareholders and executive bonuses.
- Concrete Example: IBM announced it would pause hiring for roughly 7,800 back-office roles that AI could handle. Translation services, junior coding, and copywriting are already being heavily displaced. The savings aren’t used to give remaining workers shorter hours or bigger paychecks; they are used to boost stock prices.
- Simplified Claim: AI makes the pie bigger, but under Hoarder Capitalism, the owners eat the whole pie while the bakers are sent home.
“The machine eliminates the need for labor without eliminating the need for income. That is the trap.”
4. The Rentier Shift: From Buying Tools to Renting Intelligence
Software used to be a durable product you bought once. Now, AI is a service you subscribe to forever.
- The Problem: You don’t own an AI model. You rent access to it by the query (tokens) or by the month. The company can change the price, downgrade the quality, or cut you off at any time.
- Concrete Example: ChatGPT Plus costs $20/month. If you stop paying, you lose your assistant. Similarly, APIs (like OpenAI’s GPT-4) charge fractions of a cent per word processed. If your business relies on that API, you are permanently dependent on their pricing.
- Simplified Claim: It turns intelligence from a permanent tool into a perpetual utility bill—like paying a tax on every thought you generate.
| Old Software Model (Ownership) | New Rentier AI Model (Tollbooth) |
|---|---|
| Buy Microsoft Office once (CD-ROM) | Pay a monthly fee for Microsoft Copilot |
| Run software on your own computer | Access models only via cloud APIs |
| Open-source code you can modify | Closed “black boxes” you can’t inspect |
| One-time purchase | Perpetual subscription and per-token fees |
5. The Structural Trap: Capital Demands Infinite Returns
It’s not because tech CEOs are “evil.” It’s because the financial system forces continuous growth.
- The Problem: Investors put billions into AI and demand massive, ever-increasing returns. Therefore, a company cannot afford to give away a perfect, permanent AI doctor—it would kill the revenue stream. They must keep users dependent and perpetually “upgrading.”
- Concrete Example: Just as Apple slows down older iPhones to encourage upgrades, AI companies regularly deprecate older, cheaper models and push expensive frontier models to maintain high average revenue per user.
- Simplified Claim: The system rewards hoarding regardless of who is in charge. Even a well-meaning CEO is forced to play this game by shareholders.
“The problem isn’t that the wrong people own the machine. It’s that owning the machine gives them the power to extract rent forever.”
6. The Ecological Toll: Digital Hoarding is Physically Destructive
AI looks like math and code, but it runs on massive physical infrastructure that destroys the environment.
- The Problem: Data centers guzzle electricity and millions of liters of fresh water for cooling. Rapid hardware turnover generates massive e-waste and requires destructive rare-earth mining.
- Concrete Example: Training GPT-3 consumed approximately 700,000 liters of fresh water (enough to fill a nuclear reactor’s cooling tower). Microsoft’s global water consumption spiked by over 30% during the development of its large language models.
- Simplified Claim: Treating AI like an infinite gold rush means burning the planet’s finite resources to build bigger, faster furnaces.
“The planet doesn’t care about your profitable transaction. It only feels the heat, the extraction, and the waste.”
7. The Open-Source Counter-Strike: Proving Abundance is Possible
There is a living alternative that works within the market: open-source, lightweight AI.
- The Argument: You don’t need a $100 million data center to make useful AI. Smaller, open-weight models can run on a laptop and are becoming astonishingly capable.
- Concrete Example: Meta’s Llama 3 and China’s DeepSeek V3 have proven that you can achieve near-GPT-4 performance for a fraction of the cost (DeepSeek V3 was trained for roughly $5 million, compared to the estimated $100–$200 million for GPT-4). These models are free to download and fine-tune.
- Simplified Claim: Open-source AI proves that intelligence doesn’t have to be a tollbooth. It can be shared infrastructure that lets small businesses, doctors, and teachers build their own tools.
“An open model doesn’t just give intelligence away for free. It prevents intelligence from becoming an exclusive club.”
8. Automation ≠ Abundance (Ownership is the Missing Variable)
A robot can produce infinite food, but if one person owns the robot, they decide who eats.
- The Problem: AI can make things cheaper and faster. But “cheaper” only helps you if you have a job that pays you. If AI replaces your job, the super-cheap goods are still too expensive for you.
- Concrete Example: Imagine an entirely automated farm owned by a single billionaire. The farm produces enough wheat for a million people, but charges high prices. The former farmhands now have no income, so they starve—despite abundant food. The problem isn’t the farm; it’s who owns it.
- Simplified Claim: Abundance is a technical possibility. Access is a political decision.
“AI can create the cake. But it doesn’t decide who gets a slice.”
9. The Wage Paradox: Breaking the Work-Consumption Loop
Capitalism relies on workers getting paychecks to buy things. If AI destroys the need for workers, who buys the things AI makes?
- The Problem: If AI replaces truck drivers, cashiers, and coders, those people lose their purchasing power. The AI factories still produce goods, but nobody can afford them. The economy chokes on its own success.
- Concrete Example: If self-driving trucks replace 3.5 million US truckers, those truckers cannot afford the autonomous taxis or the AI-generated consumer goods. Unless we find a new way to distribute purchasing power (like universal basic income or public dividends), the system collapses into a depression, despite massive productive capability.
- Simplified Claim: A system built on wages breaks when the machines take all the wages.
“Sufficiently advanced automation forces society to ask: Why do we need to sell our time to survive, when the machines already do the work?”
10. The Political Question: Infrastructure or Tollbooth?
AI doesn’t choose its future—we do. It’s a tool.
- The Problem: We are sleepwalking into letting 5 private companies own the backbone of the 21st-century economy.
- Concrete Example: The internet started as public research (DARPA) and still functions best when neutral. We should treat foundational AI models like roads and water pipes—vital infrastructure that should be publicly regulated and accessible, not owned entirely by billionaires. France’s “Mistral” or open-source efforts show that governments and communities can push back.
- Simplified Claim: We have to decide collectively: Should AI be a basic right (like clean water) or a luxury good (like a yacht)?
“AI does not choose the society it creates. The rules we write around AI choose it.”
Summary of Key Concepts
| Concept | Plain Definition |
|---|---|
| Hoarder Capitalism | A system where the wealthy don’t just get rich—they buy up and close off essential resources (data, chips, AI models) to make everyone else pay rent. |
| Rentier AI | Charging users a recurring fee just to access intelligence, rather than letting them own it. |
| Manufactured Scarcity | Making AI scarce on purpose (via secret code, high prices, or legal threats) even though copying it is essentially free. |
| Circulatory Capitalism | The alternative: sharing AI models openly so that thousands of smaller players can compete, innovate, and improve them. |
| Technological Feudalism | A society where most people don’t own the tools they use; they just rent them from a tiny digital aristocracy. |
Structural Analysis
The Core Distinction
We must separate three things:
- Technical Capacity (What AI can do physically).
- Productive Capacity (What society could produce with AI).
- Ownership Structure (Who legally controls the AI).
The Critical Logic
The system follows a destructive loop:
Public Knowledge → Corporate Scraping → Private Models → Paywalls & Rent → Massive Profits → More Enclosure
The healthy alternative would be:
Public Knowledge → Open Research → Shared Models → Broad Innovation → Lower Costs → Higher Purchasing Power
Key Tension
- Capitalism’s logic: Enclose assets to maximize long-term profit.
- AI’s logic: Copy and share at zero marginal cost (it wants to circulate).
- The Result: A massive political fight over whether AI will be locked down or let loose.
Final Insight
The most important AI breakthrough isn’t the “AGI” that answers everything. It’s the realization that we already have enough technology to radically improve everyone’s life—but we are choosing to hoard it.
The remaining barriers aren’t technical. They are institutional, legal, and political. We know how to build powerful, open models. We know how to share infrastructure. We know the ecological costs. What we lack is the courage to change the rules of ownership.
We face a stark choice:
- Enclosure (a world of AI tollbooths, subscription serfdom, and billionaire fiefdoms).
- Circulation (a world of open infrastructure, public health AIs, free educational tutors, and shared scientific discovery).
This isn’t about destroying markets. It’s about ensuring that no private institution becomes the unavoidable gatekeeper to society’s cognitive future.
Break open the silicon hoard.