Is AI a bubble?
Genuinely unresolved, and the honest framing is conditional. The Bank for International Settlements concluded in January 2026 that macroeconomic and financial stability risks appear moderate, but that the boom's sustainability hinges on AI firms meeting high earnings expectations — and warned that equity prices have run far ahead of debt market pricing. That gap is the whole argument.
Why — the first-principles explanation
A bubble isn't "prices went up a lot." It's prices detached from the cash the assets will ever produce. That definition is why the AI question is hard: nobody disputes the spending is happening; they dispute what it will earn. And future earnings aren't a fact you can look up. So anyone who tells you flatly that AI is or isn't a bubble is expressing a forecast in the grammar of an observation.
What you can observe is the financing structure, and that's where the BIS focused — because it's the part that's already measurable. Historically, big IT firms paid for investment out of operating cash flow: money the business generated. The BIS found the scale of AI investment is now so vast that firms are shifting from cash flows to debt, with private credit expanding rapidly. This matters because of an asymmetry. If you build a data center with your own cash and it disappoints, you had a bad year. If you built it with borrowed money and it disappoints, you still owe the money. Debt converts disappointment into distress. The shift from cash to debt doesn't prove a bubble — it raises the stakes of being wrong.
The numbers are concrete. Private credit lending to AI-related companies went from near zero to over $200 billion, climbing from under 1% of outstanding loan volumes to almost 8%; the BIS projects $300–600 billion by 2030. Roughly 20% of private credit funds now touch AI-related sectors, up from 5% in 2010. Meanwhile AI-related investment reached about 1% of US GDP and has driven almost half of US GDP growth in recent quarters — meaning this is no longer a tech-sector story. It's a GDP story. The single sharpest tell the BIS names is that equity prices have run far ahead of debt market pricing. Read that carefully. Stock investors and bond investors are looking at identical companies and disagreeing about the risk. Bond markets get paid to be pessimistic and are pricing more caution than equity markets. Both cannot be right.
And here's the part that should temper doom: bubbles and real infrastructure aren't opposites. The 1990s telecom bust vaporized fortunes and left behind the fiber that carried the internet. The BIS notes the AI investment boom isn't especially large by historical standards — around 1% of GDP — but that the end of previous investment booms was associated with a GDP growth slowdown of more than 1 percentage point on average, and that there's little evidence such booms translate into sustained higher growth. The data centers may well be useful and the people who financed them may lose money. Those can both be true.
An example that makes it click
Picture a gold rush town. Everyone can see the miners arriving, the tents going up, the shovels selling out. That part isn't speculation — it's happening, and you can count it.
What nobody knows is how much gold is actually in the hills. If there's plenty, the town becomes a city and the shovel sellers look like geniuses. If there isn't, the tents come down, and whoever borrowed to build the general store still owes the bank — while whoever paid cash just has an empty store. That's the entire AI question. The construction is real and visible. The gold is a forecast. And the reason economists watch the borrowing so closely is that debt decides whether a disappointment is a bad year or a bankruptcy.
Key facts
- The BIS (Bank for International Settlements) Bulletin No 120, published January 7, 2026 by Iñaki Aldasoro, Sebastian Doerr, and Daniel Rees, concluded macroeconomic and financial stability risks from the AI boom appear moderate, but sustainability hinges on AI firms meeting high earnings expectations.
- The BIS flagged that equity prices have run far ahead of debt market pricing — stock and bond markets disagreeing about the same companies' risk.
- AI-related investment (data centre equipment and construction) reached roughly 1% of US GDP; total IT-related investment including other equipment and software rose to about 5% of GDP.
- AI-related investment has accounted for almost half of US GDP growth in recent quarters; data centre investment contributed an average 0.4 percentage points to GDP growth.
- Private credit lending to AI-related firms grew from near zero to over $200 billion, rising from under 1% to almost 8% of outstanding loan volumes; the BIS projects $300–600 billion by 2030.
- The end of previous investment booms was associated with a GDP growth slowdown of more than 1 percentage point on average, and the BIS notes little evidence of investment booms producing sustained higher GDP growth.
▶ The 60-second explainer (script)
Is AI a bubble? The honest answer is conditional — and the condition is measurable. First, definitions. A bubble isn't prices going up a lot. It's prices detached from the cash those assets will ever produce. That's why this is hard. Nobody disputes the spending is happening. They dispute what it'll earn. And future earnings aren't a fact — they're a forecast. So anyone stating flatly that AI is or isn't a bubble is dressing up a prediction as an observation. What you can measure is how it's being paid for. In January 2026, the Bank for International Settlements looked at exactly this. Historically, big tech financed investment out of operating cash flow — money the business made. The BIS found the scale is now so vast that firms are shifting from cash flow to debt, with private credit expanding fast. Why does that matter? Asymmetry. Build a data center with your own cash and it disappoints — you had a bad year. Build it with borrowed money and it disappoints — you still owe the money. Debt turns disappointment into distress. The numbers: private credit to AI firms went from near zero to over two hundred billion dollars, from under one percent of loan volumes to almost eight. Meanwhile AI investment is about one percent of US GDP and has driven almost half of recent US GDP growth. This isn't a tech story anymore. It's a GDP story. And the sharpest tell the BIS names: equity prices have run far ahead of debt market pricing. Stock investors and bond investors are staring at the same companies and disagreeing about risk. Both can't be right. But here's the nuance people miss. Bubbles and real infrastructure aren't opposites. The 1990s telecom bust destroyed fortunes and left behind the fiber that carried the internet. The data centers may be genuinely useful and the people who financed them may still lose their shirts. The BIS calls current stability risks moderate — but says it all hinges on AI firms hitting high earnings expectations. That's the whole ballgame. This is not investment advice.
What authoritative sources say
People also ask
What would actually prove AI is a bubble?
Earnings failing to arrive. The BIS frames the boom's sustainability as hinging on AI firms meeting high earnings expectations, so the test isn't valuation levels — it's whether the revenue shows up to justify them.
Why does the shift from cash to debt matter so much?
It changes what happens if returns disappoint. Cash-financed investment that underperforms produces a bad year. Debt-financed investment that underperforms still has to be repaid, which is how disappointment becomes distress.
Can AI be a bubble and still be transformative?
Yes, and that's the historical pattern. The 1990s telecom bust destroyed enormous capital while leaving behind fiber that carried the internet for decades. Useful infrastructure and ruined investors are compatible outcomes.
What does 'equity prices ran ahead of debt pricing' mean?
Stock markets and bond markets are valuing the same companies with different risk assumptions — bond investors are pricing more caution than equity investors. The BIS flags this disagreement as underscoring the tension in the boom.
How exposed is the broader economy?
More than most people assume. AI-related investment is roughly 1% of US GDP and has driven almost half of US GDP growth in recent quarters, so a sharp pullback wouldn't stay contained to tech. Past investment booms ended with average GDP growth slowdowns above 1 percentage point.
The same question, asked other ways
- Is there an AI bubble?880/mo