Core Data Points

  • US tech giants invested $560B in AI last year, earned back $35B — a 16:1 input-output ratio
  • US national debt: $39 trillion, annual interest: $1.23 trillion (first time exceeding defense budget)
  • 高志凯 predicts a global financial crisis within 12-18 months

Data Verification

AI 16:1 ratio: This number needs context. If you only count direct AI revenue ($35B), yes, it looks terrible. But this ignores AI’s efficiency gains on existing businesses — Copilot driving Office 365 ARPU increases, Google Search AI Overviews improving click-through rates and ad pricing. These indirect revenues aren’t in that $35B number.

OpenAI itself: annualized revenue approaching $40B, inference costs near $20B, not counting training and headcount. It’s probably the fastest revenue growth company in history, and possibly the fastest money-losing company in history.

So 高志凯’s 16:1 isn’t a replay of the dot-com bubble. The dot-com bubble was “no revenue.” This time there IS revenue, but costs are out of control. Different problem.

$39 trillion debt: Directionally correct (actual June 2026 is ~$36-37T). Interest exceeding defense spending is confirmed. The US government now spends $1 of every $3 in tax revenue on interest alone. And a massive chunk of that $39T was borrowed at near-zero rates during COVID, maturing in the next 12-18 months. Rolling over at current rates would double interest payments.

At that point, $2 of every $3 in tax revenue goes to interest. The government either raises taxes (political suicide), prints money (dollar collapse), or defaults (global earthquake). Three doors, none good.

The “three doors” framework: This is too extreme. There’s a fourth door — gradual inflation to erode real debt burden. The US has done this multiple times in history. It’s painful but not apocalyptic. You just need inflation to run above interest rates, and the debt-to-GDP ratio slowly declines. No crash needed, just sustained discomfort.

Comparison to 2008

The author argues this round is fundamentally different: in 2008, the financial system broke but the real economy held up (Apple still made phones, Amazon still built warehouses). This time, the real economy itself might get hit. TSMC’s 3nm lines are for AI chips, NVIDIA gets 80% of revenue from datacenters. If AI capital dries up, these supply chains collapse within months.

This observation has merit. But “everyone admits AI is useful while every AI-adjacent company is laying off” — this is more likely a correction period than a systemic collapse. Useful things eventually find business models; it just takes time.

Score: 6.5/10

Impactful data points, but conclusions lean doomsday. The 16:1 ratio would look much better including indirect efficiency gains. The 12-18 month crisis window is almost certainly too precise — very few people in history have accurately predicted crisis timing. The “three doors” framework ignores gradual adjustment.

The author himself says he hopes this prediction doesn’t come true. But 16:1 and $39 trillion are two numbers that can’t be ignored. Between hope and reality lies an accounting ledger that balances very clearly.