Transcription
You think your AI subscription is the bargain of the century. In reality, it's a trap. A power user on tools like Code Interpreter actually costs $10,000 per year to run. But you're only paying a fraction of that because venture capitalists are picking up the tab. You're in the AI "over-moment," a short-term illusion designed to get you hooked before prices skyrocket. But the money is running out. When this trillion-dollar house of cards collapses, the tools you rely on daily will either disappear or cost you ten times as much. The economics of AI are broken.
Chapter 1: The $20 Illusion. It all starts with your wallet. A serious Code Interpreter user consumes roughly 10 billion tokens per year. Tokens are essentially the thinking units of AI. Every word it reads, writes, or every decision it makes is based on a token. If you were to pay for this usage through a standard API, those 10 billion tokens would cost you about $15,000 per year. That's the real, unsubsidized price. No discounts, no incentives, just the pure compute cost. Now, the same user, on a flat-rate plan, pays about $20 per month for the same workload, down from $15,000 to $1,200. That's a 92% hidden subsidy. Imagine walking into a car dealership, picking out a $50,000 car, and being told you only have to pay $1,000 because someone else, somewhere, covered the rest. It makes no sense, and that's exactly what makes this model so strange. But the answer lies in OpenAI's own financial projections, leaked to The Information. The company expects to lose $14 billion in 2026. Not revenue, but losses. A $22 monthly subscription covers only about 1.7% of the actual costs an active power user incurs. You're not a customer, you're the bait. Every prompt entered, every line of code generated, every late-night chat session is paid for by investors who are betting that no one can live without this product once the real bill finally arrives. Entire industries are being bought at a loss. Law firms are reviewing documents for 5 cents on the dollar. Marketing agencies are creating campaigns at prices unimaginable 18 months ago. Hospitals are testing diagnostic tools at prices no vendor could sustain at scale long-term. Every single deal is propped up by investor capital expecting a tenfold return. If companies are losing money with every new user they gain, why are they trying to gain more users? Because we've seen this play before, and we know how it ends.
Chapter 2: The Ghost of Uber. In 2014, a black SUV arrived outside your apartment within 3 minutes. The driver was polite, the car spotless. The ride to the airport cost you $11. You wondered how it all added up. It didn't, and that was the point. It was never meant to. For nearly a decade, an entire generation lived in what economists would later call the "millennial lifestyle subsidy." Venture capitalists deliberately poured money into ride-sharing, food delivery, co-working spaces, and meal kits. They priced below cost to crush competitors and create habits. The goal was to take over the market first, then raise prices until it was profitable. UBER's take—the percentage of each ride the company keeps—tells the whole story. In 2022, Uber kept about 22 cents of every dollar a rider paid. By 2024, that number had risen to roughly 42 cents. Drivers got a smaller cut, riders paid more. The company finally became profitable. Now, the same thing is happening in the AI sector. Same investors, same playbook, same price memo. Industry analysts expect prices for consumer subscription tiers to roughly double in 2 years. Anthropic has introduced new limits to push power users toward more expensive plans. Google is now testing exclusive Gemini features that used to be free and are now only available in premium. A 100% price increase isn't a rumor; it's already on the calendar. Enterprise customer contracts follow the same trend. Individual contracts from 2024 will renew in 2026 at significantly higher prices. The product remains the same, but it costs substantially more. Users will have to agree or opt out. Ride-sharing only had to move cars from point A to point B. As usage increases, costs don't rise sharply; they actually become more efficient. More drivers, more density, better routing. AI is different. The underlying math of thinking doesn't get cheaper in the same way. It gets complicated fast. AI executives continue to claim that compute power gets cheaper every year. Per-unit costs will eventually level out. This isn't a direct lie; it's more of a half-truth. The cost of model inference is decreasing annually. Chips and models are becoming more efficient and leaner. AI-generated words are significantly cheaper to produce today than they were 18 months ago. That's the part they want you to see; the other part, they want to hide.
Chapter 3: The Code Interpreter Tax. Modern agentic workflows, like those powering Code Interpreter and ChatGPT's deep research tools, consume 5 to 30 times more tokens than simple chat sessions from 2 years ago. When you ask a code assistant to fix this bug, it doesn't just write 50 words in response. It quietly generates sub-tasks. Then it re-reads your files. It checks its own work. It drafts multiple versions, discards many, and then quietly tests in the background. A single request can consume hundreds of thousands of tokens before an answer appears. A model might be slightly cheaper per word than before, but it also produces far more words per request. The total bill skyrockets. This is called the token tax. It's ruining small, agile AI startups, burning through their seed capital. It threatens to extinguish one of the most profitable business models in the history of the internet.
Chapter 4: The Search Tax. For 25 years, Google has printed money, and it was brutally simple. A user types a search query. Google returns ten blue links from the open web. The total cost for Google—servers, electricity, indexing—is a fraction of a cent per search, yet the ads next to those results bring in far more. Margin is one of the great financial miracles of modern times. Now, Google is rebuilding this entire system on generative AI. An AI-powered search response that delivers a paragraph instead of just links is significantly more expensive to produce than a traditional keyword search. Now, multiply that by billions of daily search queries. If Google completely replaces traditional search with AI overviews, the 21st century's most reliable money machine disappears. The profit margins that funded YouTube, Android, WMO, and Gmail are vanishing. Wall Street analysts have secretly run the worst-case scenarios, and the numbers are catastrophic. And it gets worse. The advertising models become obsolete too. If AI answers directly, no one clicks links, meaning advertisers stop paying. Google faces a future where it handles more searches than ever, has higher operating costs than ever, and generates less revenue per search than at any point in its modern history. Tech giants are intentionally cannibalizing their most profitable business lines. They believe it's more dangerous to let a competitor kill a gold mine than to do it themselves. Economists call this the Innovator's Dilemma. When a new technology threatens the core business, established companies have two options: Stand still and defend the existing money machine while a competitor builds the future, or cannibalize it themselves on their own terms, hoping to capture revenue on the next platform before the old one disappears. This is the path companies like Google, Microsoft, and Meta are essentially betting on with AI. They hope AI will displace current revenue streams. Yet, no one can prove it. Everyone is already too invested to back out. If the unit economics are this bad, how can these companies report record AI revenue to Wall Street every quarter?
Chapter 5: The Roundtrip Scam. This is where it gets sophisticated. Microsoft publicly commits to investing $13 billion in OpenAI. The press release is slick, the headlines dramatic, the stock prices surge. This makes investors happy. Reading the fine print reveals that much of the investment never lands directly in OpenAI's account but is provided as Azure cloud credits. This is essentially a gift card that's only valid within Microsoft's data centers. OpenAI books the amount as capital raised. Microsoft books the cloud usage as revenue. It's simultaneously an investment and a sale. OpenAI has also committed to spending up to $250 billion on Azure services, locking in this cycle for years to come. Now, add Nvidia on top. NVIDIA announces $10 billion in commitments to OpenAI. OpenAI takes this capital and uses it to buy Nvidia GPUs. Nvidia's quarterly revenue hits a record, and its stock price soars. The entire cycle takes only a few months, and in reality, hardly any real money changed hands. At each step, it simply got a new name. Add Oracle, Coreweave, and AMD to the list. Each firm invests and then sells services to the next, booking revenue as the same dollar flows through the cycle. The technical term is "round-tripping." In Silicon Valley, it's called "strategic partnership."
Chapter 6: The Hardware Debt Trap. In 2025, Big Tech is expected to spend approximately $320 to $400 billion on AI infrastructure. Updated projections for 2026 push this number toward $500 billion. Data centers, GPUs, cooling systems, power supplies, and entire networks will be scaled up to handle this. Meanwhile, global consumer spending on AI services, according to the State of Consumer AI report by Menlo Ventures, is only about $12 billion. Hundreds of billions are flowing out, while only $12 billion is coming in. The gap is equivalent to the economy of an entire medium-sized country. It's being filled not by revenue but by debt. Corporate bonds, structured and private loans. Meta took on $30 billion in bond markets at the end of 2025. Another $30 billion flowed through a joint venture organized by Morgan Stanley, which held liabilities off Meta's balance sheet. Microsoft has signed a 20-year power purchase agreement to recommission 3 Mile Island. Google is working with Next Era Energy to reopen nuclear power plants. These commitments remain even if AI revenues disappoint, but the hardware is not permanent. A high-end Nvidia GPU driving this boom is typically only current for one to three years before being superseded by the next generation. It loses most of its book value as soon as a new generation appears. This happens roughly every 18 months. A data center full of 3-year-old chips is dead weight in the industry. Compare this to the original dot-com crash. When the bubble burst in 2000, telecom companies left behind millions of miles of buried fiber optic cable. New companies bought it at a fire-sale price and built YouTube, Netflix, and Spotify on top of it. The crash was brutal, but what remained was useful. This AI bubble will leave behind warehouses full of useless silicon tied to 20-year power contracts, as well as concrete skeletons in the middle of nowhere. No one will know what to do with them. Utility companies will pass on higher electricity prices to households for decades, regardless of AI revenues. A $100 billion gap cannot be covered for long. The companies participating in this race already know this. That's why they are taking subtle measures to slow losses before the public notices. Most users have already felt it. They just haven't recognized the connections yet.
Chapter 7: The Stealth Downgrade. An AI model that used to solve your code with one stroke now forgets your project mid-process. A chatbot that used to write five paragraphs at once now stops after three. An image generator that used to create a flawless portrait in 30 seconds now spits out something with seven fingers and prompts you to upgrade to the next tier. No one is imagining this. The product is getting worse. When the numbers don't add up, the easiest thing for a vendor to do is to quietly water down the visible offering. The signs are easy to spot. News limits now refresh every 8 instead of 5 hours. In an app, the default model is subtly switched from a flagship model to a smaller, cheaper version, and reminder features are reduced. Advanced logical reasoning is only available in a more expensive tier. The "god model" promised in launch keynotes is quietly replaced by a cheaper, less intelligent version. Reddit threads about AI tools are full of users reporting that their assistant has become lazier. Engineers post comparison screenshots showing the same product delivering significantly worse results than just 6 months ago. Companies almost always deny this. Often, they release selected benchmarks, clean prompts, controlled conditions, and optimized scenarios to showcase performance. This buys them time but doesn't solve the larger problem. A deeper problem has already begun destroying the first wave of the AI ecosystem.
Chapter 8: The Great AI Extinction of 2026. According to data from CB Insights, about 40% of AI startups founded in 2024 have already closed or been acquired by larger companies through so-called "acqui-hires." This is the polite term for a fire sale, where a struggling company goes to a competitor very cheaply. The buyer isn't really buying a company; they're acquiring the engineers, shutting down the product, and taking as much talent as they can. These weren't hobby projects in some garage. These companies had closed Series A funding rounds with well-known investors, generated revenue, had paying customers, and positive TechCrunch profiles. Then, within 18 months, the lights went out. The reason is almost always the same: their cost of revenue—the money they pay to model providers like OpenAI, Anthropic, and Google—is so high that it negates any profit they could possibly make. A startup that built a fancy UI on top of GPT-4 might charge $50 a month. But the API usage that same customer incurs can cost the startup $80. Every active user causes losses. Successful marketing accelerates the company's bleeding. When a foundation model provider releases a new feature, ten startups often disappear overnight. ChatGPT introduces native voice mode. Say goodbye to half a dozen voice agent startups that just closed their Series A round last quarter. Code Interpreter introduces native PDF reading. Several document tools became obsolete with one product update. An entire ecosystem of independent AI companies is collapsing under the high compute costs that no one can profitably absorb. When startups fail, cloud providers lose revenue that makes their investments in foundation models attractive. Then, the final phase begins.
Chapter 9: The Great AI Scam. Venture capital firms are no longer willing to cover losses in hopes of future glory. They want a clear, written profit plan with quarterly milestones, and they want to see it immediately. For foundation model companies, there are two options. The first is a brutal, sudden price adjustment. A $20 consumer plan becomes a $100 plan or quietly disappears and is replaced by a "Pro" tier that costs ten times as much for the same features. A Code Interpreter user who paid $20 per year is suddenly faced with a bill of nearly $10,000, which is what the API actually costs. A freelance designer who relies on a $20 image generation subscription gets an email explaining that their plan is being migrated to a new structure. Small businesses that built workflows on cheap AI face a choice: pay ten times as much, or go back to doing it the old way. The second option is even worse. The services are simply shut down. We've already seen the first signs. Smaller AI companies closed within 30 days, forcing customers to quickly migrate years of work to a remaining competitor. Specialized models for legal research, medical imaging, and customer support were discontinued because they weren't economically viable. An era of cheap AI ends with 1,000 small bills, 1,000 small shutdown notices. A deeper truth is uglier than a price hike. AI will increasingly become a luxury good rather than a basic product by 2026. Cheap versions have accustomed a generation to usage. An expensive version is now the only one that balance sheets can allow. Large corporations that can afford a new price tier will solidify their advantage. Freelancers, small businesses, and the people who drove early adoption and created the hype will be the first to be priced out. An economy built on the idea of cheap intelligence will now collide with the reality of expensive intelligence. The productivity assumptions from 2024 will no longer hold in 2027. An AI revolution will come, but not for everyone, and not at the promised price. History shows that crashes usually happen faster. The dot-com crash lasted two years from peak to trough. The AI bubble has more leverage, more concentration, and more debt built into its foundations. When it tips, everything can happen in months or even weeks. Once margins shrink or a major customer publicly defects, confidence can evaporate overnight. The tools used daily by millions were never as cheap as assumed, as they were subsidized by investor money that is now drying up. An AI age might still come. However, a cheap AI age that tricked an entire generation into building their working lives on it is already over. The bill just hasn't arrived yet. When it does, that price will never feel real again. The confidence that made the AI industry seem inevitable is beginning to crumble. What once looked like unstoppable momentum now shows the first hairline cracks beneath the surface. The question suddenly shifts from "How big can this get?" to "Who bears the consequences if it doesn't get that big?" Find out what happens to the economy when the $2 trillion AI bubble bursts, or watch this instead. M.