I didn't plan to write this piece. The market made me do it. Another earnings season wrapped, another record-high headline about US stocks and the AI miracle. Goldman Sachs cheers. Société Générale nods along. The numbers look absurd: 27% profit growth, three straight quarters above 25%. Impressive. Except when you scratch the surface, it's not a market rally. It's a reflexivity loop wearing a trench coat, selling tickets to its own echo. Algorithms smell fear, but they respect speed. And right now, the speed is all going in one direction: into the pockets of the AI supply chain.
Let's cut through the noise and talk about what that 27% actually is. Because based on my audit experience — and I've been tracking this cycle since the DeFi yield days when APY was just subsidized TVL — this number stinks of the same trick. The market is watching the revenue side of AI spending and confusing it with the health of the entire economy. You see, AI spending is not a monolith. It's a knife that cuts both sides. For hyperscalers, that $50 billion capex bill is a weight on their income statement, a depreciation anchor dragging down net income. For NVIDIA, Broadcom, and the power utilities selling the electrons to keep those data centers humming, that same spending is pure, sweet revenue. One man's cost is another man's cocaine. The article wants you to believe AI spending drives earnings. It does. But only for the suppliers. For the spenders, it's a margin squeeze disguised as strategic vision. Yield is a drug; exit liquidity is the cure. And the market is currently high on the supply-side high without asking who's paying for the next hit.
The real issue here is the quality of the growth. The source material — a flimsy four-data-point note that doesn't even name a single company — wants you to believe the whole US stock market is booming. That's a lie wrapped in an index. The S&P 500 is market-cap weighted. The top ten stocks control over a third of the index. A 27% earnings growth rate with that structure doesn't mean 500 companies are thriving. It means roughly ten giant tech firms are holding a massive capex party and inviting their suppliers to drink the overflow. Use an equal-weight index and the number drops significantly. I've seen this pattern before. In 2020, when DeFi protocols were printing their own liquidity metrics, everyone looked like a genius until the incentives dried up. Same playbook. Same missing variable: terminal demand.
Let me give you the contrarian angle nobody on CNBC wants to touch. The AI earnings cycle is partly a circular trade. Chip suppliers buy cloud credits from hyperscalers. Hyperscalers buy chips from the suppliers. Both declare record revenue. But is there a real end customer paying for the compute? Some of it, sure. But a significant chunk is companies stockpiling AI capability like it's toilet paper during a pandemic, hoping the applications will materialize. MIT's 2025 study stirred the pot by claiming most enterprise AI pilots show zero measurable ROI. The market ignored that. It's too busy calculating next quarter's capex guidance. This is the tell. When we shifted from DeFi yield farming to AI capital expenditure, we didn't change the underlying psychology. We just changed the name of the drug. The high is in the hardware, not in the actual business outcome.
What's actually happening on the ground? I hosted a roundtable in Toronto last week with exchange heads and infrastructure guys. The mood was manic. Data center power procurement is the new NFT flipping. But nobody in that room could tell me how their AI investments are directly generating a return beyond the speculative value of having an AI strategy. They're building capacity because their competitors are. That's a herd mentality, not an investment thesis. The energy constraints are real. If you look at the power grid, there isn't enough electricity to run all the planned data centers. Something's got to give. Either AI compute gets less energy-hungry, or the expansion slows down. And if the expansion slows, the whole earnings narrative shifts from "hypergrowth" to "digestion."
Chaos is just data waiting for a narrative. And the current narrative is conveniently ignoring the concentration risk. This isn't an era of widespread prosperity. It's a narrow, deep, and fragile boom. Semiconductors, HBM memory, advanced packaging, power generation — those sectors are printing money. They have pricing power because they sit on bottlenecks like TSMC's CoWoS packaging or the global supply of electricity. But the moment the hyperscalers whisper about cutting capex — just a whisper — the earnings expectations for the entire supply chain will collapse simultaneously. That's the definition of a structural vulnerability. When every stock in your portfolio is trading on the same macro signal, you don't have diversification. You have one big bet in ten different clothes.
Let's talk about the missing pieces. The spending is real. The construction cranes are real. The power lines are real. But the product that this infrastructure is supposed to serve — that's still mostly a theoretical science project. The business model is "we build first, we monetize later." That worked for Amazon in the late 90s because they were building a logistics empire with clear consumer demand. For AI, we're building a computing empire while the consumer demand is still in beta testing. The depreciation schedules tell you a lot. If you assume a 5-year depreciation on a GPU cluster, your earnings look one way. If you assume a 2-year schedule because the hardware becomes obsolete, your earnings look terrible. Everyone conveniently picks the longer schedule to make the quarter look better. I know this game. I've audited these numbers. The willingness to stretch depreciation is the market's way of kicking the can down the road.
The final piece that's missing from the mainstream analysis is the excessive focus on the US market. We don't see the parallel competition. Chinese AI infrastructure is scaling at a pace that doesn't appear in any US earnings report. The export controls are a temporary shield, not a permanent one. And if global capital expenditure shifts or accelerates differently, the entire global supply chain equation changes. The US doesn't operate in a vacuum, but the earnings reports often read that way.
What happens next? In 2025, we're at a fork in the road. The 27% growth rate is not the new normal. It's a distortion caused by a low base effect, massive stimulus-era liquidity lingo, and the most concentrated earnings contribution from a handful of players in modern history. The base case for me is a slowdown to mid-teens growth for the index overall, with a massive rotation out of unprofitable AI wannabes. The bull case is that AI applications find that magical product-market fit and the revenue follows. The bear case — and I'd put this at around 20% probability — is that the ROI questions become too loud to ignore. When the CFOs start forcing the CTOs to justify the GPU purchases with actual business metrics, we'll see a brutal re-pricing of the whole stack.
I'm not here to tell you to sell everything and build a bunker. I'm here to tell you that when you see a headline about "impressive earnings," always ask: impressive for whom? And who is paying the bill? Because someone always pays the bill. The market's been putting this rising AI spend on the tab for two years now. Credit card bills come due. It's a matter of when, not if. The smart money isn't selling the narrative. It's just getting positioned for the inevitable moment when the music stops and we all have to look at the actual business models underneath the piles of silicon. So watch the next quarters of hyperscaler capex guidance. Watch the power grid data. Watch whether AI vendors start reporting actual usage-based revenue. But most of all, watch whether the markets continue to confuse spending with earning. Because confusion is a finite resource, and the market is running low. I didn't want to be the one to say it. But someone had to clear the noise.