NVIDIA’s market cap hit $4 trillion. Its trailing P/E ratio is 70. The last time a single stock had that much weight in the S&P 500 was Cisco in 2000. Cisco lost 90% of its value. The ledger does not lie.
Ray Dalio is screaming into the void again. He’s comparing the AI market to 1929 and 2000. He’s right about the pattern. He’s wrong about the cause. The AI bubble is not a mania of irrational exuberance. It is a rational response to a liquidity glut. The real question is not whether the bubble will burst, but when the liquidity cycle turns.
I’ve spent 17 years reading order flow. I’ve audited smart contracts that lost $31 million. I’ve reverse-engineered the TerraUSD death spiral before it collapsed. The same structural fragility is present in the AI infrastructure buildout. The capital expenditure is real. The revenue is not growing fast enough.
Let’s look at the numbers. In 2025, Microsoft, Google, Meta, and Amazon combined spent over $300 billion on capital expenditures. Most of it went to AI data centers and GPUs. That’s 50% of their combined operating cash flow. History shows that when CapEx exceeds 40% of operating cash flow for more than two consecutive years, a correction follows. The last time this happened was in the telecom bubble of 2000.
Survival is the first profit metric.
Now, the bulls will tell you that AI is different. They’ll say that NVIDIA’s revenue grew 200% year-over-year in 2024. That’s true. But growth rates are decelerating. The marginal return on each additional GPU is declining. I saw this exact pattern in the crypto mining boom of 2021. When ASIC orders exceeded network hash rate growth, the collapse was inevitable. The same logic applies here.
I didn’t need a Bloomberg terminal to see this. I wrote a Python script in 2020 that monitored Uniswap V2 deployment events. The principle is the same: early movers capture the spread. Latecomers get wrecked. The AI infrastructure players are the latecomers. The cloud providers are spending billions because they are afraid of missing out. Fear is a liquidity drain.
Code does not lie, but liquidity does.
Here is the contrarian angle. The common belief is that an AI bubble burst will destroy the industry. That’s wrong. The bubble burst is the best thing that can happen to AI. Just like the 2000 Internet bubble laid the fiber optic cable that enabled Web 2.0, the AI bubble will crash GPU prices and make AI inference affordable. The survivors will be the application layer, not the infrastructure layer. The smart money is already positioning for that.
I see this from my own community. In 2024, I built a copy-trading bot for Bitcoin ETFs using Rust. The latency arbitrage was 0.5% per trade. That’s a real edge. But the infrastructure cost was high. After the AI bubble bursts, the cost of compute will drop by 50-70%. That will unlock a new wave of AI applications. The companies that are actually using AI to generate revenue today—like code assistants, customer service automation, and generative search—will see their margins explode.
Trust the math, ignore the memes.
Now, let’s talk about the trigger. Dalio is focused on interest rates. He’s obsessed with the Fed and the yield curve. I think the trigger is more specific. The real catalyst will be a capital expenditure guidance cut from one of the hyperscalers. If Microsoft or Google announces a 10% reduction in CapEx, the entire supply chain will reprice. NVIDIA will drop 30% in a week. That’s the moment of maximum pain.
But here’s the hidden insight. The market is not pricing in the time lag. The current CapEx orders are for chips that will be delivered in 2026-2027. Even if the demand signal weakens in 2025, the physical supply chain will keep churning for another 18 months. That means the infrastructure glut will be worse than the market expects. The oversupply will last longer.
I’ve seen this before. In 2022, I watched the Terra Luna reserve mechanism implode. The death spiral took 72 hours from trigger to zero. The AI infrastructure death spiral will take 18 months. But the market will react instantly. The first 40% drop will happen in a week. The bottom will take two years to form.
So what do you do? The moon is a myth. The ledger is the only truth. Here is the survival playbook.
- Short the GPU-related stocks. The narrative is priced in, but the earnings are peaking. Long NVIDIA calls at this level is a loser’s bet.
- Go long on AI application software with positive unit economics. Companies like those in code assistance or customer service automation have low CapEx and high variable margins. They will benefit from the compute cost drop.
- Keep 10% cash for the crash. When the CapEx cuts come, the market will panic. That’s your entry point.
- Use gold and inflation-linked bonds as a hedge. Dalio’s “all-weather” portfolio is boring but it works.
I’m not saying AI is a scam. The technology is real. But the price is not. The gap between the technology curve and the valuation curve is a chasm. The market is pricing in a future that is 10 years away. That’s a 10-year gap. The market will close that gap violently.
Survival is the first profit metric. The survivors will be the ones who can verify the code, ignore the hype, and wait for the ledger to settle.
Trust the math, ignore the memes.
