Is AI a Bubble? A Framework for Thinking About It
Separating whether AI is a genuinely transformative technology from whether today's prices for it are sustainable is the real question — and history offers a useful, if imperfect, guide.
Two different questions, often confused
When people argue about whether artificial intelligence is a bubble, they're frequently talking past each other because they're answering two different questions at once. The first is whether AI is a real, transformative technology — one that will meaningfully change how goods and services are produced over the coming decades. The second is whether the prices being paid today, for the companies and infrastructure tied to that technology, are sustainable relative to the cash flows those investments are likely to generate. A technology can be entirely real and still be badly overpriced at a given moment, just as a technology can be genuinely useful while remaining underappreciated by markets for years. Confusing these two questions is one of the most common errors in these debates, in either direction.
What history offers, and what it doesn't
Economic historians have long pointed to a recurring pattern in transformative technology cycles: railways in 19th-century Britain, electrification in the early 20th century, and the dot-com boom of the late 1990s all involved genuine, lasting technological change alongside periods of investment that, in hindsight, outran near-term demand. In each case, enormous capital was poured into infrastructure — railway track, generating capacity, fiber-optic cable — well ahead of the revenue that infrastructure would eventually generate, and in each case a painful shakeout followed even though the underlying technology went on to reshape the economy. That pattern is often summarized as 'the technology was real, the returns to early investors were not,' and it's a useful caution. But history is not a precise forecasting tool: every cycle has different financing structures, different starting valuations, and a different mix of players, so pattern-matching to the past can only inform judgment, not replace it.
What analysts actually look at
Rather than declaring a verdict, most serious analysts assess bubble risk by tracking a handful of recurring signals. None of these is decisive on its own, and reasonable people weigh them differently.
- Valuation relative to current and expected earnings — how much of future growth is already priced in, and how far out that growth needs to materialize to justify today's price.
- How the investment is being financed — cycles fueled heavily by debt or by circular arrangements between a small number of large players are generally viewed as more fragile than those funded from existing cash flow.
- Breadth versus concentration — whether gains are spread across many companies and sectors or concentrated in a small number of names, since narrow rallies have historically been more vulnerable to sharp reversals.
- The gap between reported usage and reported revenue — whether the technology is being paid for at a scale consistent with the valuations built on top of it, or whether adoption is still mostly promotional or experimental.
A technology can be exactly as transformative as its most enthusiastic supporters claim and still be a poor investment at the wrong price.
The bull and bear cases, stated fairly
The optimistic case holds that AI is already delivering measurable productivity gains in software development, customer service, and research, that the infrastructure being built now will be used for decades, and that today's spending — while large — is being funded largely by profitable companies reinvesting real cash flow rather than by speculative leverage. The skeptical case holds that a meaningful share of current enthusiasm rests on projected rather than realized returns, that a small number of companies account for an unusually large share of both the spending and the market gains tied to it, and that prior technology cycles suggest today's leaders are not guaranteed to be tomorrow's winners even if the technology itself succeeds. Both views can be argued in good faith from the same set of facts, which is itself informative about how much genuine uncertainty exists.
A framework, not a forecast
Rather than trying to predict whether AI is or isn't a bubble — a question that will only be answerable with confidence in hindsight — a more useful exercise is building a personal framework: understanding how exposed your own portfolio is to a small number of AI-related companies or sectors, recognizing that concentration (however it arose) increases both potential upside and potential downside, and deciding in advance how you'd react to a sharp drawdown rather than deciding in the moment. This article is general educational content, not investment advice, and nothing here should be read as a prediction about future prices or a recommendation to buy, sell, or hold any particular investment; consider your own circumstances or speak with a qualified financial professional before making investment decisions.
This article is educational and general in nature. It isn’t personalized investment, tax, or legal advice — always weigh your own circumstances, or talk to a licensed professional, before making financial decisions.