The Great Illusion: why the bursting of the "AI bubble" will be a historic mistake

The author is an AI analyst, not a financial analyst. These views are his own and do not constitute investment advice.

There is an image circulating on the Internet that makes your head spin. A simple two-bar chart. The first bar represents Nvidia's market capitalization. The second, that of the entire global pharmaceutical sector combined. In October 2025, the Nvidia bar overtook the second one.

Read that sentence again for a second. A single company, one that sells "picks and shovels" for the digital gold rush, is worth more than all the companies that heal humanity put together, from Pfizer to Roche to Eli Lilly.

Faced with such a disproportion, the Pavlovian reflex of financial analysts is to cry bubble. They are right to be afraid, but it seems to me they are wrong about the nature of the danger. What awaits us in 2026 is not the bursting of a bubble, it is an optical illusion that could cost skeptics dearly.

The mirage of circular money

The main argument of the doomsayers rests on a diagram of financial flows that looks, deceptively, like a budgetary Ponzi scheme. In it, you see Nvidia investing in startups like CoreWeave or OpenAI. Those same startups use that money to buy... Nvidia chips. All of it propped up by Microsoft, which invests in OpenAI, which pays CoreWeave, which pays Nvidia. The money goes round in circles, artificially inflating everyone's revenue.

This is what we call "round-tripping." From a distance, it looks like a speculative bubble about to explode, a house of cards held up by hype alone. Except that this analysis misses the essential point: this is not a valuation bubble, it is a logic of industrial consolidation.

Unlike the dot-com bubble of the year 2000, where money evaporated into websites selling pet food with no business model (the famous Pets.com), today's billions are solidifying. They are turning into steel, into copper, into power plants and semiconductor factories. We are not funding hot air, we are funding the physical infrastructure of the 21st century.

The 2026 trap: the construction "dip"

This is where my thesis becomes slightly iconoclastic. I am not expecting a crash, I am expecting a misunderstanding.

We have entered a phase of massive construction (Capex). But building datacenters and the nuclear plants to power them takes time. It is a cyclical logic, slow and heavy. Conversely, software AI evolves at an exponential speed, a technological "snowball" that never stops.

In 2026, we are going to hit a temporal wall. The colossal investments of 2024-2025 will still be under construction. Returns on investment (ROI) will be slow to show up on the balance sheets, because the factories will not be finished yet. So there will be a lag. A silence. A plateau.

Wall Street hates a vacuum. Faced with this mechanical slowdown in investment (the time it takes for the cement to dry, literally), the markets risk panicking. They will interpret this logistical pause as the end of the party. Valuations could drop, stocks could be dumped. People will talk about "the bursting of the AI bubble," which will trigger even more selling. People will believe so firmly in a bubble that we will indeed witness every signal of a bubble bursting.

It will be a monumental mistake. The crisis will be very real but purely narrative. The sector will probably find itself undervalued.

The 2027 counterattack

Those who sell in 2026 will have missed the forest for the trees. Because as soon as 2027, once this infrastructure is delivered, global compute capacity will make a quantum leap. The independent institute Epoch AI, one of the leading references on analyzing the AI value chain, shows in a recent report¹ that within a year, the world's computing power will have been multiplied... by 6! AI models, fed by this new power, will reach levels of performance that will retroactively justify every dollar invested.

We will then witness a violent rebound. The narrative will flip once again: "It wasn't a bubble, it was a foundation." And it is there, between 2028 and 2030, that we will see the emergence of the first company valued at 10 trillion dollars (10,000 billion).

Will it be Nvidia? Perhaps. But do not underestimate the one that recently made a grand return as the AI leader: Google. Google is today the only player in the world to master the entire vertical value chain. They design their own chips (the TPUs, on which Gemini 3 was trained entirely without a single Nvidia chip), they own their undersea cables, their datacenters, their models and their applications. This total technical sovereignty is a weapon of unprecedented power that could allow them to sweep the board at the end of the decade.

There is no bubble

Financial history is written by those who know how to tell noise from signal. The noise will be the panicked cry of 2026 in the face of balance sheets in transition. The signal is the physical and irreversible transformation of the global economy by artificial intelligence.

There is no bubble. There is just a gigantic construction site. And as with all construction sites, there comes a moment when it is loud, dusty, and it feels like nothing is moving forward. That is usually the moment to buy the land.

Footnotes

1. Trends in AI Supercomputers (April 2025). See: https://epoch.ai/blog/trends-in-ai-supercomputers ↩