Original title: The AI bubble is already bursting.
Original author: Xu Gong from the North of the City, Gelong
In recent days, the market has experienced dramatic fluctuations, and the "AI bubble theory" has become increasingly prevalent.
Bridgewater Associates founder Ray Dalio said: There is a bubble in the AI market, and the level is "relatively high".
Nvidia CEO Jensen Huang said: There is a huge opportunity in AI, and the demand for computing power is just beginning to explode.
Whom should we believe?
They are both right.
Is there a bubble in the AI industry? Absolutely.
However, technological bubbles are often the only way society can pay tribute to disruptive advanced productivity.
It is not simply a derogatory term.
In the long run, this is a phenomenon that is bound to occur at the beginning of the emergence of advanced productive forces.
Many people are comparing the current situation to the dot-com bubble of 2000 and are deeply worried.
The dot-com bubble of that year did indeed cause the Nasdaq to plummet by nearly 78%, wiping out more than $5 trillion in wealth.
But twenty years from now, which industry will be able to live without the internet?
Today, the value of the internet industry has far exceeded that of its bubble period.
The AI bubble, at least on the surface, appears to be a similar situation.
The bubbles that exist in the capital market cannot stop almost all industries in society from being proactively empowered by AI.
AI+ is an inevitable trend.
Just as all industries today cannot do without the internet, all industries in the future will also be inseparable from AI.
01
In an era when any company could go public and raise money simply by having ".com" in its name, the Nasdaq surged nearly 600% between 1995 and 2000. What followed was a financial crisis that lasted for two and a half years.
Those once-famous names, such as software company MicroStrategy, saw its stock price plummet by 62% in a single day due to accounting scandals and overblown claims; Pets.com (an online dog food seller) and Webvan (a pioneer in fresh food e-commerce) went bankrupt on the spot.
...
In the panic, almost everyone was accusing the internet of being a scam.
However, the physical infrastructure accumulated through the excessive squandering of speculative capital often nurtures the next generation of super giants at extremely low cost.
The bursting of the bubble was not due to a problem with internet technology itself, but rather because the physical construction of infrastructure could not keep up with the pace of the market.
For example, those once-dominant telecommunications companies (such as WorldCom and Global Crossing) invested heavily in laying global submarine optical cables and optical density wavelength division multiplexing networks. Although this led to their own bankruptcy, these inexpensive "information superhighways" became the perfect breeding ground for the rise of Netflix, Zoom, and the mobile internet in the future.
Without the massive, forward-thinking global investment in telecommunications infrastructure around 2000, there would have been no YouTube video streaming explosion, let alone the later cloud computing infrastructure.
The most typical example is Amazon.
The stock price plummeted from a high of $107 in 1999 to $7 in 2001, a drop of more than 90%.
But it survived because its underlying business logic, " reconstructing retail with the internet ," aligns with the direction of advanced productivity.
This is the classic Amara's Law : overestimating the short-term impact of a new technology while severely underestimating its long-term impact.
In the early stages of a technological revolution, the frenzy of speculative capital inevitably leads to over-investment and the formation of bubbles.
This is the intellectual tax that innovation must pay.
But when the bubble bursts, what remains will be an even more indestructible advanced productive force.
02
Looking back to 2026, the AI industry bubble appears even bigger.
The five largest cloud service providers alone—Amazon, Google, Meta, Microsoft, and Oracle—are projected to spend $690 billion on capital expenditures by 2026, and their total investment in AI infrastructure is expected to reach $5.3 trillion by 2030.
Of these, only about 25% were spent on GPUs, while the remaining 75% was poured into physical infrastructure: liquid cooling systems, power transmission, network switches, optical modules, and land.
In terms of revenue, the combined total revenue of all the leading pure AI companies, including OpenAI, AnthropicCohere, Mistral , and Perplexity, is not expected to exceed $40 billion in 2026.
Nearly 700 billion yuan was invested in the foundational layer, while tens of billions were recovered in the application layer.
What else could this severe asymmetry be but a bubble?
We cannot draw such a conclusion so simplistically and hastily.
There is one crucial point that cannot be ignored.
When OpenAI released GPT-4 in March 2023, the mixing cost per million token inputs was approximately $30.
By April 2025, with the optimization of the model architecture and the improvement of inference computing power, the price of a model with the same level of intelligence will plummet to $0.1-$0.15 per million tokens.
According to Stanford University's AI Index Report and data from TokenCost, the cost of AI inference has fallen by more than 99.7% in the past two years.
According to traditional linear thinking, with costs plummeting, companies should reduce their AI spending.
But in reality, enterprise AI cloud spending tripled between 2024 and 2025.
Why?
Because when the marginal cost of "intelligence" approaches zero, AI is no longer just a simple text summary or chatbot, but has entered a new era of intelligent agents and multimodal enhanced retrieval.
Companies are starting to have AI agents run tasks automatically thousands of times a day, such as writing code, scanning millions of legal contracts, and simulating biological experiments.
Cheap tokens unlock a massive amount of long-tail demand that was previously unable to be commercialized due to cost constraints.
This can be seen by comparing Nvidia in 2026 with Cisco, the network hardware giant in 2000.
The two have extremely similar ecological niches, but their underlying financial health is vastly different.
This perfectly illustrates the "Jeves Paradox" in economics: technological progress improves energy efficiency, but instead of reducing energy consumption, it leads to greater demand due to lower costs.
Even after the so-called "DeepSeek moment" at the beginning of last year, the market quickly woke up in the following months: the more optimized the algorithm, the lower the threshold for enterprises to adopt AI, and the total consumption of computing power actually increased exponentially.
This is precisely why AI has the potential to gradually be embedded in almost all traditional industries.
Just like how all industries have been embracing the "Internet Plus" strategy over the past two decades.
From SaaS software to biomedicine, and then to advanced manufacturing robots driven by embodied intelligence, almost every industry is embracing AI+ in 2026.
No one discusses "Should we use AI?" but rather worries about "Has our data been properly cleaned? Are our API call limits sufficient? Is our RAG architecture optimal?"
There is indeed a bubble in the AI industry at present.
But for businesses, if they don't embrace the bubble, they will be crushed by the times.
This has been proven by the Internet era over the past two decades.
03
Currently, we are undoubtedly at a critical juncture in the technology life cycle: on the eve of the "trough of disillusionment" on Gartner's Hype Cycle, or a turning point in the theory of "Technological Revolution and Financial Capital".
The AI bubble has already burst, but many people haven't realized it.
In the past few years, a large number of venture capitalists have developed a fear of taking action.
A few upstarts could easily raise money by writing dozens of pages of PPTs and wrapping themselves in OpenAI's API. Now, as the tide recedes, these companies with no moat and only concepts are dying out in large numbers.
This is the market undergoing self-purification, and also a sign of a bubble bursting.
But this is only the surface.
The underlying logic of the market is undergoing three profound evolutions:
First, the value transfer from CapEx to OpEx.
Currently, the money has been earned by those selling shovels. Nvidia, TSMC, and companies that sell optical modules and server liquid cooling equipment have reaped most of the profits.
However, as computing power gradually becomes "infrastructure," like water and electricity, the real excess profits will gradually shift to the application layer.
These are AI-native companies that can use extremely low-cost tokens to truly solve pain points in vertical industries and reshape business processes (OpEx optimization).
Second, valuation multiple compression and earnings digestion.
The market's high valuation of AI infrastructure does not necessarily mean it will collapse.
In many cases, a company's rapid profit growth will gradually absorb the high valuation by "trading time for space".
As long as the revenue growth of cloud computing giants keeps pace with the depreciation of capital expenditures, this game of musical chairs can evolve into an unprecedented industrial upgrade.
For example, global automotive and chip giants have shortened the R&D to mass production cycle of new products by 35% and improved the overall efficiency of the entire production line by 18% by introducing end-to-end AI twin technology.
For example, in the financial industry, by 2026, quantitative trading, risk control, and credit assessment will be entirely dominated by multimodal agents. AI is not only processing macroeconomic expectations with microsecond-level timestamps, but also deeply participating in every micro-level asset pricing.
In industries such as law, healthcare, and auditing, which heavily rely on senior professional knowledge, AI has already transformed from a "junior assistant" to a "partner-level expert."
ChatGPT, Gemini, and Claude have over 1 billion active users, a significant portion of whom use them as alternatives for their daily high-intensity mental work.
Including you and me.
All of the above are actual events that happened, and everyone can see them.
04
Looking back at the magnificent history of science and technology, Schumpeter 's concept of "creative destruction" is always playing out.
The capital market is always impatient, always hoping to make 10 yuan back the day after investing 1 yuan today.
When the nearly $700 billion in infrastructure investment cannot be fully converted into profits in the short term, the market will inevitably face a brutal reshuffle.
Eliminate those opportunistic shell companies that just rely on PowerPoint presentations to get by, and keep those with real technical expertise and practical applications.
After the reshuffling, those cheap and massive computing centers and highly optimized model algorithms will serve thousands of industries at extremely low prices.
After 2000, humanity ushered in the digital age where all industries are inseparable from the internet.
Today, we are irreversibly heading towards a golden age of intelligence where all industries are governed and empowered by AI.
Amidst the clamor of the bubble, the underlying productive potential is genuine and without any exaggeration.
