This article is sourced from: Xu Gong, North of Gelong City.
Data support: Gougu Big Data
The AI bubble is becoming the most divisive consensus in the global market. Dalio says the bubble is already very high, while Huang says the opportunity is just beginning; one sees an overheated capital market, while the other sees the beginning of a productivity revolution.
The real question isn't whether there's an AI bubble, but what will remain after the bubble bursts. The dot-com bubble of 2000 caused the Nasdaq to plummet, companies to go bankrupt, and wealth to evaporate, but it also left behind the infrastructure of submarine cables, broadband networks, and cloud computing, which ultimately supported Amazon, Netflix, YouTube, and mobile internet.
Today's AI stands in a similar position. On one hand, hundreds of billions of dollars are being poured into data centers, power, liquid cooling, optical modules, and GPUs; on the other hand, there's a huge gap between the current situation and the fully realized application revenue. A bubble clearly exists, but the underlying productivity is genuine. When token costs plummet and intelligence begins to be used like water and electricity, AI will no longer be just a chat tool, but will enter the real workflows of code, healthcare, finance, law, manufacturing, and scientific research. The market will weed out shell companies and PPT-based entrepreneurs, but it won't reverse the direction of AI+. The bubble will burst, but the industry will remain. Enjoy:
In recent days, the market has experienced sharp fluctuations, and the "AI bubble theory" has become increasingly prevalent.
Bridgewater Associates founder Ray Dalio said that the AI market is in a bubble, and that 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 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 can live without the internet? Today, the value of the internet industry has far exceeded that of its bubble period.
The AI bubble, at least superficially, presents a similar picture. The bubble existing in the capital markets cannot prevent almost every industry in society from actively being empowered by AI.
AI+ is an inevitable trend. Just as all industries cannot function without the internet now, all industries will be inseparable from AI in the future.
01 The "Intelligence Tax" That Innovation Must Pay
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.
Back then, those once-famous names, such as software company MicroStrategy, saw their 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 (the pioneer of fresh food e-commerce) went bankrupt on the spot... In the panic, almost everyone accused the internet of being a scam.
However, the physical infrastructure built up through excessive spending by speculative capital often nurtures the next generation of super giants at extremely low cost. The bursting of the bubble is not a problem with internet technology itself, but rather that the pace of physical infrastructure construction could not keep up with 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.
Without the massive, forward-thinking global investment in telecommunications infrastructure around 2000, there would have been no explosion of video streaming on YouTube, let alone the later development of cloud computing infrastructure.
The most typical example is Amazon. Its 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," was in line 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 overinvestment and the formation of bubbles. This is the intellectual tax that innovation must pay. But when the bubble bursts, what remains will be a more robust and indestructible advanced productive force.
02 Why are corporate AI spending increasing instead of decreasing?
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 total investment in AI infrastructure is expected to reach $5.3 trillion by 2030. Of this, only about 25% will be spent on GPUs, while the remaining 75% will be 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, Anthropic, Cohere, Mistral, and Perplexity, is not expected to exceed $40 billion in 2026.
Nearly 700 billion was poured into the infrastructure layer, while only tens of billions were recouped at the application layer. This severe asymmetry—what is it if not a bubble?
We cannot draw such a conclusion so simplistically. There is a key 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 each million tokens for models with the same level of intelligence plummeted to $0.1-$0.15.
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.
Following traditional linear thinking, with costs plummeting, companies should reduce their AI spending. However, the reality is that enterprise AI cloud spending tripled between 2024 and 2025.
Why?
As the marginal cost of "intelligence" approaches zero, AI is no longer just a simple text summarizer or chatbot; it has entered a new era of intelligent agents and multimodal enhanced retrieval. Companies are starting to have AI agents automatically run thousands of tasks in a loop, writing code, scanning millions of legal contracts, and simulating biological experiments.
Cheap tokens have unlocked 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. Their ecosystems are extremely similar, but their underlying financial health is vastly different.
(A hard-core financial comparison between Nvidia and Cisco)
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 sobered 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 increases exponentially.
This is precisely why AI can gradually be embedded into almost all traditional industries. Just as all industries have been pursuing "Internet Plus" over the past two decades, from SaaS software to biomedicine, and then to advanced manufacturing robots driven by embodied intelligence, in 2026, almost every industry is embracing AI Plus. No one is discussing "whether we should use AI," but rather worrying about "whether our data has been properly cleaned, whether our API call limits are sufficient, and whether our RAG architecture is optimal."
There is indeed a bubble in the AI industry right now. But for companies, if you don't embrace the bubble, you'll be crushed by the times. This has already been proven by the internet era over the past two decades.
03 The profound evolution of the market: from infrastructure to applications
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. A few upstarts could easily raise money by writing dozens of pages of PowerPoint presentations and wrapping themselves in an OpenAI 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 bursting bubble. But this is only the surface. The deeper logic of the market is undergoing three profound evolutions:
First, the value transfer from CapEx to OpEx.
Currently, the profits are mostly going to those selling the tools—Nvidia, TSMC, and companies selling optical modules and server liquid cooling equipment have reaped the lion's share of the gains. However, as computing power gradually becomes "infrastructure," like water and electricity, the real excess profits will gradually shift to the application layer. This means 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 doesn't necessarily mean a collapse. In many cases, rapid profit growth will gradually absorb the high valuations through a "time-for-space" approach. 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 involved in every micro-level asset pricing process.
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 boast over 1 billion active users, a significant portion of whom use them as alternatives to handle demanding mental tasks. This includes you and me. All of this is a fact that is undeniable and visible to everyone.
04 Conclusion
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, hoping to earn ten dollars back the day after investing one dollar. When nearly $700 billion in infrastructure investment cannot be fully converted into profits in the short term, the market will inevitably undergo a brutal reshuffle. Those speculative shell companies that rely solely on PowerPoint presentations will be eliminated, leaving behind those with genuine technological foundations and real-world application scenarios.
After the reshuffling, those cheap and massive computing centers and highly optimized model algorithms will serve thousands of industries at extremely low prices.
Since 2000, humanity has entered a 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.
