The bubble chorus

Posted on: 18 December 2025

Over the past few weeks, a remarkably coherent narrative has taken hold across financial markets: artificial intelligence is a bubble, correction is imminent, the system is about to purge speculative excess. Michael Burry, the prophet of the 2008 crisis, has opened short positions on Nvidia and Palantir worth over a billion dollars. Jim Chanos, another legendary short seller, has followed suit. Financial headlines alternate between "AI bubble fears" and "valuation concerns" with a frequency that suggests more editorial coordination than independent analysis.

On 17 December, the Nasdaq fell 1.9%, Nvidia dropped 3.8%, and the S&P 500 recorded its third consecutive losing session. For those following the dominant narrative, everything appears to confirm the thesis: the bubble is bursting, chickens are coming home to roost, reality is finally punishing irrational exuberance.

There is just one problem: the facts tell a different story.

The anatomy of a coordinated narrative

Before dismantling the bubble thesis, it is worth observing how this narrative has been constructed. Not because there is necessarily a conspiracy, but because the mechanisms of media consensus formation are themselves a structural pattern worthy of clinical analysis.

The "AI bubble" narrative has some interesting characteristics. First, it is extremely easy to communicate: "bubble" is a word that immediately evokes 2000, 2008, collapse, the punishment of excess. It requires no technical explanation, no understanding of fundamentals, no subtle distinctions. It is an emotional category before it is an analytical one.

Second characteristic: the narrative is carried forward by figures with specific track records. Burry is famous for predicting the subprime crisis. Chanos built his career on short selling. When they speak of bubbles, the media listen because they have been right once before. The fact that they have also been wrong many other times does not enter the narrative.

Third characteristic: the narrative is self-reinforcing. Every market decline is interpreted as confirmation of the thesis. Every rally is interpreted as "last gasp before the crash" or "dead cat bounce". No evidence can falsify the narrative, because the narrative absorbs all evidence as confirmation of itself.

This is the classic structure of what Popper called pseudoscience: an unfalsifiable thesis. And when a market thesis becomes unfalsifiable, it ceases to be analysis and becomes ideology.

The facts that do not fit the narrative

On the evening of 17 December, just hours after the Nasdaq had lost nearly 2% and newspapers were headlining the AI bubble, Micron Technology published its quarterly results. Revenue of $13.64 billion against analyst expectations of $12.84 billion. Guidance for the next quarter of $18.7 billion against expectations of $14.2 billion. Earnings per share forecast at $8.42 against a consensus of $4.78.

These are not marginally better numbers than expected. These are numbers that demolish expectations. The earnings guidance beat consensus by 75%.

Micron's CEO stated that the high bandwidth memory market will reach $100 billion by 2028, bringing forward previous forecasts by two years. He added that supply will remain "substantially short" of demand throughout the period.

These are not the numbers of an industry in a bubble. These are the numbers of an industry that cannot produce fast enough to meet demand.

But Micron is not an isolated case. Nvidia reached a market capitalisation of $5 trillion, becoming the first company in the world to touch this threshold. In 2024, revenues grew 114%, reaching $130.5 billion. Demand for Blackwell chips exceeds supply. Major customers are signing multi-year contracts to secure future deliveries.

Google has recorded its eighth consecutive month of gains, with a 66% increase year to date. UBS Global Wealth Management published a report explicitly stating it sees no "evidence of an investment bubble".

How does all this reconcile with the dominant narrative? The answer is: it does not. And this should suggest that the narrative is wrong, or at least seriously incomplete.

The real mechanism: who sells pickaxes and who digs

If the "AI bubble" narrative does not explain the facts, what structural pattern does?

The most appropriate historical analogy is not the dot-com bubble of 2000, but the California Gold Rush of 1849. During that rush, most gold seekers lost money or at best broke even. Those who made systematic fortunes were the sellers of pickaxes, shovels, tents and jeans: Levi Strauss founded his empire selling trousers to miners.

In the AI market of 2025, something structurally similar is occurring. Those producing hardware infrastructure, the "pickaxes" of artificial intelligence, are generating record and verifiable profits quarter after quarter. Those consuming this infrastructure to build applications are burning cash awaiting a return on investment that may or may not arrive.

This bifurcation is visible in the data. Semiconductors as a category have maintained gains of 14.5% over recent months. Software over the same period has lost 5%. This is not a uniform movement "up" or "down" of the entire AI sector: it is a separation between those who have demonstrated profits and those who are still asking for time.

Oracle caused its own share price to collapse not because the business was performing poorly, but because it raised spending forecasts without proportionally raising revenue forecasts. OpenAI has stated it expects to begin generating substantial profits in 2030. Meta and Microsoft are forecast to have negative free cash flows in 2026 after accounting for shareholder returns.

The market is not saying "AI is a bubble". The market is saying "show me the profits". And those who can show them are rewarded, those who cannot are punished. This is not bubble behaviour: it is rational behaviour.

Why the narrative serves someone

If the "bubble" narrative does not correspond to the facts, why is it so persistent? The answer requires an analysis of incentives.

For short sellers like Burry and Chanos, the bubble narrative is functional regardless of whether it is true. If they can convince enough market participants that the bubble is about to burst, they can trigger sales that push prices down, allowing them to close positions at a profit. The narrative becomes self-fulfilling, at least in the short term.

For financial media, the bubble narrative generates traffic. "AI bubble incoming" is a headline that attracts clicks. "Hardware fundamentals remain solid while software underperforms" is not. Editorial incentives reward dramatic simplification, not nuanced analysis.

For fund managers who stayed out of the AI rally, the bubble narrative provides retroactive justification. They did not miss opportunities through analytical error: they were prudent in the face of an irrational bubble. If the bubble "bursts" (whatever this means in a market where Nvidia has still risen by hundreds of percentage points), they can claim to have been right.

None of these incentives requires the narrative to be true. It only requires it to be credible enough to influence the behaviour of other market participants.

The question nobody is asking

There is a question that should be at the centre of any serious analysis of the AI sector, and which the bubble narrative systematically avoids: what is the rate of AI infrastructure adoption in enterprises?

Speculative bubbles are characterised by a disconnection between asset prices and real usage. During the dot-com bubble, internet companies were valued at billions without having users or revenues. During the housing bubble, homes were bought by speculators who had no intention of living in them.

In the case of AI, infrastructure is being purchased by companies that are actually using it. Microsoft, Google, Meta, Amazon are buying Nvidia chips and Micron memory to run services that already have hundreds of millions of users. They are not speculators buying assets hoping to resell them at a higher price: they are operators integrating AI into their products.

This does not mean that all AI investments will generate positive returns. It means that the market structure is fundamentally different from that of a classic bubble. There is real demand, real usage, integration into real products. The open question is whether returns on these investments will be proportional to the capital deployed, not whether there is real usage.

How to verify who is right

If this analysis is correct, what should we observe in coming quarters?

First: the bifurcation between hardware and software should continue or intensify. Chip and memory producers should continue to beat expectations, while companies consuming infrastructure without yet monetising it should remain under pressure.

Second: demand for AI infrastructure should remain above supply. If Micron and Nvidia begin reporting inventory accumulation or slowing orders, the thesis changes radically.

Third: major hyperscalers should continue to increase AI capex. If Microsoft, Google and Amazon begin cutting investments, it is a signal that demand is actually slowing.

If instead we see Micron miss guidance, Nvidia report declining orders, and hyperscalers reduce investments, this analysis is falsified and the bubble narrative becomes more credible.

The difference between analysis and ideology is this: analysis specifies the conditions under which it admits to being wrong. Ideology absorbs all evidence as confirmation of itself.

The value of not following the chorus

For those who must make strategic decisions, there is an operational principle that emerges from this analysis: the dominant narrative about a market is itself data to be analysed, not a conclusion to be accepted.

When everyone says the same thing at the same time with the same words, the probability that they are all conducting independent analysis is low. The probability that they are following a shared script, consciously or otherwise, is high.

This does not mean the narrative is necessarily wrong. It means its informational value is low precisely because it is shared. If everyone knows there is "an AI bubble", this information is already priced in. It offers no advantage to those who follow it.

The advantage lies in seeing what the narrative conceals. And at this moment, the AI bubble narrative conceals a structural bifurcation between those who produce infrastructure and those who consume it, conceals quarterly results that demolish expectations, conceals demand exceeding supply in critical components.

Those who follow the chorus may be right about short-term sentiment. Those who look at the facts are more likely to be right about medium-term fundamentals. Market history suggests that, in the long run, fundamentals win.