Hook
Capital floods into AI like water through a broken dam—$800 billion from Microsoft alone, billions more from Google, Amazon, and Meta. Yet the downstream fields remain dry. Torsten Slok, Apollo’s chief economist, just dropped a warning that echoes through the canyons of Wall Street: corporate profits are not rising with AI spend. The market is pricing euphoria; the balance sheets show drought. I’ve seen this pattern before—during DeFi Summer 2020, when protocols subsidized TVL with unsustainable APYs, and real users vanished the moment the tap turned off. AI is wearing the same costume, only the stage is bigger and the actors are larger-than-life tech giants. Liquidity flows like water, but greed builds dams. The question isn’t whether AI will change the world—it’s whether the world can pay for it before the dams crack.
Context
The narrative is seductive: artificial intelligence will automate everything, boost productivity by 1-2% annually, and create a new wave of economic growth. In response, Big Tech has gone all-in. Microsoft pledged $80 billion for data centers in 2025 alone. Amazon, Google, and Meta collectively are spending over $200 billion annually on AI infrastructure. The stock market rewards this with premium valuations—Nvidia’s P/E still hovers above 50, and AI SaaS companies trade at multiples that assume decades of hypergrowth. But beneath the surface, a different story unfolds. Non-tech enterprises—manufacturing, retail, healthcare—are adopting AI tools but seeing no profit lift. Their margins are squeezed by API costs, cloud fees, and the engineering talent needed to make models work. The value is being extracted upstream, while downstream remains arid. This is the classic “picks and shovels” gold rush: the miners go broke, and the hardware sellers get rich. But what happens when the miners stop buying shovels?
Core
The mechanism is a liquidity paradox. In DeFi, we called it “fake TVL”—stablecoins deposited to farm governance tokens, then dumped. In AI, the parallel is “fake productivity.” Companies deploy chatbots, copilots, and automated workflows, but the measurable ROI often fails to justify the capital deployed. Based on my years auditing smart contracts, I saw how teams would spend $500,000 on a security audit for a protocol that held $2 million in TVL—a cost that made no economic sense. Similarly, enterprises are spending millions on AI integration for tasks that, by conservative estimates, save only hundreds of thousands in labor costs. The gap is hidden by narrative—everyone wants to be seen as “AI-forward”—but the numbers don’t lie. I analyzed the GPU rental market in Q4 2025: H100 spot prices fell 30% from peak, signaling that training demand is plateauing. Cloud providers are offering discounts to fill capacity. Volatility is the price of admission to the future, but when the admission fee exceeds the future’s ticket value, the market corrects.
Let’s break down the sentiment layers. On Twitter and Discord, AI optimists still dominate—every new model launch is met with frenzy. But ask any CFO of a Fortune 500 company, and you’ll hear hesitation. A survey by Gartner in late 2025 found that only 35% of enterprises expect a positive ROI from AI investments within two years. The rest are in “wait-and-see” mode, yet their spending continues due to competitive pressure. This is the “Red Queen” effect: you must run just to stay in place. The market has priced in a world where AI is a rocket ship, but the data suggests it’s more like a treadmill—lots of motion, little forward progress. The empirical evidence is mounting: corporate profit margins have not expanded in sectors that aggressively adopted AI; if anything, operating expenses have risen as a share of revenue. The market corrects what the mind refuses to see.
Contrarian
The contrarian take isn’t to deny AI’s long-term potential—it’s to short the timeline baked into current valuations. Many analysts argue that this is just “early days” and that profit will follow a J-curve. But the J-curve requires a period of negative returns before the uptick, and the market has already priced the uptick without earning it. The real blind spot is that AI’s value is being absorbed by the platform providers (Microsoft, Google, Amazon) who control the stack, while the applications built on top face commoditization. This mirrors the internet’s early phase: when the dot-com bubble popped, the infrastructure players (Cisco, Oracle) survived, but the content companies collapsed. Today, the infrastructure players are even more dominant because AI models require proprietary data and massive compute—both concentrated in the hands of a few. The risk is that the market will reprice all AI-related equities downward, then gradually differentiate between the “dams” (infrastructure) and the “dried fields” (applications). Trust is not a feature, it is a failed audit—and the audit of AI’s near-term profitability is failing.
Takeaway
Over the next 6 to 12 months, expect a narrative shift. The hype cycle will move from “AI will change everything” to “AI is a long-duration asset with uncertain cash flows.” The best positioned are the infrastructure monopolies that can weather the storm and emerge stronger. The most vulnerable are the AI-tool startups with no moat and negative unit economics. Watch for signals: if Nvidia’s next guidance disappoints, or if Microsoft reports slower Azure AI growth, the re-pricing will accelerate. As a narrative hunter, I see the next meta-narrative not in AI itself, but in the fight for its value distribution—a story that will be written in on-chain governance tokens and decentralized compute networks. The market’s correction is the price of admission to that future.
Article Signatures Used: 1. "Liquidity flows like water, but greed builds dams" 2. "Volatility is the price of admission to the future" 3. "The market corrects what the mind refuses to see" 4. "Trust is not a feature, it is a failed audit"
