Nearly $1 Trillion Gap in AI Investment: Stanford Economist Warns of Bubble Risk
nashnova research
A nearly $1 trillion gap has opened between what six tech giants have spent on AI and what they've earned from it. Two Stanford economists warn that once investor patience runs out, the speed of the exit will decide whether this bubble pops or deflates.
Nearly $1 trillion spent — where does the money come back from?
Since 2024, Alphabet, Amazon, Meta, Microsoft, Oracle, and SpaceX have collectively spent nearly $1 trillion more on AI than they have earned from AI-related revenue.
This means → the money is already out the door, but the income is nowhere close to catching up — a gap too large to dismiss as "early-stage investing."
Stanford economists Jared Bernstein and Ryan Cummings estimate these companies need to triple or quadruple their annual AI revenue over the next decade just for the investments to break even.
Chips depreciate in five years — how tight is the payback window?
A large share of big-tech AI spending goes to chips, and chips have an economic life of roughly five years.
In plain terms = the money bought an asset with a very short shelf life — in five years the chips need replacing, and if revenue hasn't caught up, that capital is effectively gone.
This compresses the payback window further: it's not "earn it back slowly" — the math has to work before the hardware is written off.
What do Goldman Sachs and Anthropic's numbers show?
Goldman Sachs's analysis last week reached a similar finding: big-tech AI revenue still falls below the break-even line needed just to cover capital expenditure, though the bank remains relatively optimistic on long-term returns.
Anthropic's IPO prospectus — obtained by Reuters — shows 2025 revenue of roughly $4.6 billion, with operating losses nearly double that figure.
Yet Anthropic is accelerating: its annualized revenue run rate topped $65 billion in Q2 this year, and the company has told investors it will reach quarterly profitability.
This reflects a market stress-test of the "burn cash for growth" playbook: the growth is real, but so are the losses.
How does this investment wave rank in history?
An estimate cited at the Brookings conference puts total AI infrastructure investment at $10.3 trillion by 2032 — equivalent to 3.6% of U.S. GDP per year.
This means → the scale exceeds every previous infrastructure boom — railroads, highways, the power grid, and telecom networks included.
In plain terms = humanity has never placed a bet this large on a single technology in this short a time frame.
What do the historical parallels suggest?
Bernstein and Cummings point to two precedents: electrification took decades to enter factories and deliver productivity gains; the 1990s dot-com bubble burst when investors realized profits were far further away than expected.
Cummings told Axios: "I think these companies will eventually make AI profitable, but I'm not sure which ones. There are trillions of dollars in future profits — they're just not coming on the hyper-accelerated timeline that current valuations are built on."
This reflects a crucial distinction: the question is not "can AI make money?" but "can it make enough money within the timeline baked into today's valuations?"
What risk hasn't been priced in yet?
The Stanford analysis assumes the cost of capital won't rise significantly — but interest rates are currently on an upward trajectory.
This means → if rates keep climbing, borrowing to fund AI gets more expensive, and the trillion-dollar gap comes under more pressure, not less.
The ultimate test lands on a hard constraint: can the tech giants keep AI revenue growing at three to four times current levels before the chip depreciation clock runs out?
市场有风险，内容仅供研究参考，不构成投资建议。