What Makes an Investment Bubble?

What the price is asking you to believe

When is an industry merely doing well? When is it getting expensive, overheating, or caught in a dangerous bubble? I want a way to distinguish those conditions that gives us more than another adjective.

An asset price makes a claim about the future. The buyer is paying on the assumption that future cash flows will justify the outlay. To examine that claim, I ask whether the required future can occur, whether the capital being spent to produce it can earn an adequate return, and who will bear the loss if it doesn't. A large price increase doesn't answer those questions. A high valuation might be justified. And dismissing the technology doesn't help: technologies associated with past bubbles often became useful parts of the economy.

That gives me three layers to examine. Valuation asks what profits, revenue, and end demand the market price requires, and how those requirements compare with historical experience. Investment asks whether new capital can earn more than its cost and whether revenue comes from paying end customers. Financing asks whether the spending is funded by operating cash flow or outside money, and who might be forced to sell if conditions deteriorate.

The layers matter because the damage can spread. A valuation problem hurts the holders of the asset. Poor investment returns can hurt the industry and its suppliers. Fragile financing can pass the loss to creditors and the wider financial system.

My four categories describe an accumulation of those vulnerabilities. In a healthy boom, demand and earnings are growing and the three layers look reasonably sound. In a stretched market, the price requires something close to an optimistic outcome, although the underlying investment and financing still look viable. In an overheated market, capital is also pouring into projects whose marginal returns approach or fall below their cost, creating excess capacity. A speculative frenzy adds vulnerable financing: leverage, outside funding, or circular arrangements help prices and capital flows push each other higher, while valuation discipline is discarded.

The size of the rally isn't what determines the category.

The unit of analysis matters too. By an investment theme, I mean capital drawn together by a common narrative, such as AI. Chips, electricity, cloud infrastructure, models, and applications can occupy different stages at the same time. Each needs its own examination; a single score for "AI" conceals too much.

1. A rally is a clue, not a diagnosis

The strongest objection comes from Eugene Fama's challenge: if bubbles could be identified in advance, why wouldn't prices already have adjusted? I think the objection needs to be taken seriously, but it doesn't settle every question about risk.

In their 2019 paper Bubbles for Fama, Robin Greenwood, Andrei Shleifer, and Yang You distinguish average returns from crash risk. A sharp industry rally need not predict unusually low subsequent average returns. It can nevertheless predict a higher probability of a crash. Those are different results, and confusing them changes the conclusion.

The study also finds that volatility, turnover, issuance, and the path of the rally help distinguish subsequent outcomes. Price increases contain information, but they don't make every rally a bubble. I want to examine the accompanying conditions.

There is another serious objection to a purely valuation-based diagnosis. In their model published in 2009, Lubos Pastor and Pietro Veronesi show how technological change can generate prices that look like bubbles afterward even when investors behave rationally. Uncertainty and learning about adoption matter. Expensive alone isn't a definition of irrationality.

2. The same return can conceal different risks

My framework concentrates on three ingredients: prices that outrun fundamentals, price increases that generate further buying, and fragile financing.

Overvaluation alone is a weak basis for declaring a bubble because it can overlook growth and uncertainty. Fragile credit isn't a necessary condition for every bubble either. The latter two ingredients help explain how a boom feeds on itself and how costly a reversal might become.

Each judgment should answer a question we can investigate. Two assets can both double in price while presenting very different risks. In one, earnings and cash flows may support the move. In the other, borrowed money may be keeping the process going.

3. Work backward from the market value

Start with the market capitalization and work backward. What profit does it require? What revenue would produce that profit? How much of the relevant industry's spending would have to become end demand for the product? Compare the required growth and its duration with what similar businesses have achieved.

An expensive asset may grow into its valuation. A price that requires an implausible economic outcome presents a different problem. The task is to find out how much room there is between a demanding assumption and one with no credible path to delivery.

Comparing an index's earnings yield with Treasury yields can help start the inquiry. But that spread isn't a complete measure of the equity risk premium. Growth, the earnings measure, and risk still need to be considered.

4. Spending ahead of revenue can be sensible

During infrastructure development, investment commonly arrives before the revenue it is meant to support. Railways, electricity networks, and cloud computing have all required that sequence. A large gap between spending and current revenue doesn't by itself establish a bubble.

Ask instead whether each additional dollar can earn a return above the cost of capital over the asset's useful life. Useful life is crucial. Railway infrastructure can last decades. Servers and accelerators can need replacement much sooner, so they have less time to earn back the initial investment.

Then examine who pays. Is revenue coming from end customers, purchases among related industry participants, investor subsidies, or supplier financing? If a significant share of orders comes from customers financed by the vendor or its investors, those sales deserve closer scrutiny. During the 1999–2000 telecom expansion, equipment suppliers including Cisco and Lucent financed customers who used that money to purchase equipment. Sales looked healthier than the demand that could survive without that funding. When customers ran short of cash, orders disappeared too.

Supply also responds. Attractive returns draw new capacity. Constraints in power or advanced packaging can delay excess supply, but their removal can allow a substantial amount of capacity to arrive together.

5. Find out where the money comes from

The same investment program can have very different consequences depending on how it is financed. If a company pays from operating cash flow, shareholders bear much of the direct investment loss. Debt, off-balance-sheet arrangements, and private credit can transmit losses to creditors and financial institutions.

Capital flows can also move prices more than intuition suggests. Xavier Gabaix and Ralph Koijen's inelastic markets hypothesis estimates a market-value increase of roughly $5 for each $1 invested in the aggregate stock market, with the estimate depending on the sample and method. The point is the potential size of the price response to flows, not a multiplier to apply mechanically to every trade. Outflows deserve the same attention as inflows.

Narrow market breadth is another clue that needs interpretation. Profits themselves may be concentrated, so a rally driven by a small group of companies isn't enough to establish a bubble. Compare their share of index market value with their share of index profits. A wide gap can reveal how much depends on those few companies fulfilling expectations.

6. Watch the financing conditions

A rising price strengthens the story, draws in money, and helps raise the price again. That is the kind of feedback Soros describes as reflexivity.

We tend to imagine that a bubble ends when investors recognize it. Often the immediate trigger is more practical: buyers run out of room, funding costs rise, orders disappoint, or new supply arrives. The underlying technology doesn't have to be useless for the financing to stop working.

For someone monitoring a theme, that makes funding conditions at least as important as the next technological announcement.

7. Credit changes who pays for the collapse

The consequences depend on how the boom was funded. In their 2015 study Leveraged Bubbles, Oscar Jorda, Moritz Schularick, and Alan Taylor examine seventeen advanced economies over roughly 140 years. Bubbles accompanied by credit booms tend to produce more severe economic damage when they collapse, with deeper recessions and slower recoveries. Equity booms without comparable credit expansion can have smaller costs.

Calling something a bubble therefore leaves a crucial question unanswered: who will bear the loss this time?

8. Put the questions to the financial statements

The three layers can be applied to the AI industry, but a research framework should be distinguished from a live market assessment. The calculations below are illustrative assumptions, not a claim about today's prices or yields.

Suppose an index is reaching new highs while fewer than half of its constituents are above their 200-day moving averages. Is a small group carrying the index? Compare those companies' shares of market value and earnings. Examine the capital drawn into the common theme instead of treating the entire market as one trade.

For an interest-rate example, assume a ten-year Treasury yield of 5.3%. Its reciprocal is roughly nineteen. That provides a crude reference for comparing the Treasury yield with an equity earnings yield; it isn't a maximum reasonable price-to-earnings ratio. Earnings yield isn't the shareholder's realized total return either. If the companies driving the rally trade above that reference, the case for owning them relies more heavily on future growth. Whether valuations are stretched still depends on earnings quality, the durability of growth, concentration, and financing. Nineteen times earnings cannot settle the question.

When a price already discounts a great deal of progress, examine the next two layers through two groups of financial indicators.

The first concerns incremental returns relative to the cost of capital. The 5.3% in the example is a risk-free benchmark. A company's cost of capital also reflects risk and its financing structure. Companies rarely report a separate return on AI projects, so we need imperfect proxies. One is the relationship between cloud revenue growth and growth in depreciation and amortization. Depreciation growing ahead of revenue during construction isn't unusual. Revenue growth slowing over several quarters while depreciation accelerates deserves investigation. It isn't, by itself, a calculation of project returns. Check remaining performance obligations and customer concentration too. Are customers paying from operating revenue, external funding, or money invested by their suppliers?

The second group concerns capital expenditure relative to operating cash flow and how any gap is funded. A ratio persistently near or above 100% calls for examining cash reserves, asset sales, and outside financing. Look at bond issuance, debt balances, interest burdens, and lease commitments that haven't yet commenced. Read those measures alongside solvency: a strong balance sheet can provide substantial room. Private credit, joint ventures, and project financing may require assembling disclosures from the statements, their notes, and separate financing announcements.

I can specify what would change my assessment before those results arrive. If the gap between revenue and depreciation growth keeps worsening while the capital-spending shortfall becomes increasingly debt-funded, I'd move the risk assessment from stretched toward overheated. Either development alone would call for closer observation rather than an immediate verdict. Conversely, covering capital spending from operating cash flow reduces dependence on outside finance. It doesn't rule out impairment, excess capacity, or a valuation loss.

9. A bubble diagnosis isn't a trading signal

Identifying a bubble doesn't mean you can make money from it. A bubble can persist. Mistaking a durable expansion for one also carries a cost: staying entirely in cash while earnings and cash flows grow can be expensive.

I translate these judgments into exposure decisions rather than automatic buy or sell signals. In a healthy or stretched market, participation requires understanding how much future success is already paid for. In an overheated market, reduce exposure and examine which businesses can cover their investment needs from cash flow. In a speculative frenzy, don't build a plan that depends on selling at the exact top.

The cloud buildout in the 2010s is a counterexample worth considering. My reading is that much of the expansion was closer to a boom than a credit-fueled bubble because large companies financed substantial investment from their own operating cash flows. That structure can reduce credit contagion without implying that the entire industry experienced no credit events. Readers should still examine the evidence for themselves.

For any investment theme, I would finish with three questions:

1. How demanding is the future required by the price, compared with historical growth outcomes?
2. Can the new capital earn more than its cost, and does the revenue come from end customers?
3. Was the money earned or borrowed, and who could be forced to sell if the assumptions fail?

The more those answers deteriorate together, the more fragile the boom becomes.

These are my research judgments, not investment recommendations.