If the AI bubble is already bursting, who will truly remain?
TechFlow Selected TechFlow Selected
If the AI bubble is already bursting, who will truly remain?
The sharp decline in computing power costs is accelerating AI’s transformation of industries across the board; after this major reshuffling, what remains will be an irreversible, genuine productivity revolution.
Navigating Web3 tides with focused insightsSource: Gelonghui, Chengbei Xugong
Data support: Gougu Big Data
The AI bubble has become the most polarizing consensus in global markets. Ray Dalio says the bubble is already high, while Jensen Huang says the opportunity has only just begun—one sees overheating in capital markets, the other sees the dawn of a productivity revolution.
The real question isn’t whether AI has a bubble, but what remains after it bursts. The dot-com bubble of 2000 triggered a massive Nasdaq crash, company bankruptcies, and evaporated wealth—but it also left behind undersea fiber-optic cables, broadband networks, and cloud computing infrastructure, ultimately enabling Amazon, Netflix, YouTube, and the mobile internet.
Today’s AI stands at a similar inflection point. On one side lies hundreds of billions of dollars poured into data centers, power systems, liquid cooling, optical modules, and GPUs; on the other, a vast gap between infrastructure investment and realized application revenue. A bubble clearly exists—but the underlying productivity gains are real. As token costs plummet and intelligence becomes as readily accessible as water or electricity, AI will evolve beyond chat tools to power real-world workflows in coding, healthcare, finance, law, manufacturing, and scientific research. Markets will weed out shell companies and PowerPoint-only startups—but they won’t reverse the AI+ trend. Bubbles burst; industries endure. Below, enjoy:
Markets have swung violently in recent days, and “AI bubble” talk is rampant.
- Ray Dalio, founder of Bridgewater Associates, says: “There is an AI bubble—and it’s ‘relatively high.’”
- Jensen Huang, CEO of NVIDIA, says: “AI presents enormous opportunities—compute demand has only just begun to explode.”
Whom should we believe?
Both are right.
Does the AI industry have a bubble? Absolutely.
Yet bubbles in technology are often society’s only way of paying tribute to disruptive, advanced productivity—and are not purely pejorative terms.
In the long run, such bubbles are inevitable when transformative productivity first emerges.
Many compare today’s situation to the 2000 dot-com bubble, with deep concern. That bubble did indeed cause the Nasdaq to plunge nearly 78%, wiping out over $5 trillion in wealth.
But two decades later, which industry can function without the internet? Today, the internet’s value far exceeds its peak bubble valuation.
The AI bubble—at least superficially—mirrors this pattern. While capital markets exhibit clear froth, virtually every sector across society is actively embracing AI-driven transformation.
AI+ is inevitable. Just as every industry now depends on the internet, every industry will soon depend on AI.
01 The “IQ Tax” Innovation Must Pay
Back when any company with a .com in its name could go public and raise money, the Nasdaq surged nearly 600% between 1995 and 2000—followed by a two-and-a-half-year financial storm.
High-profile names collapsed overnight: software firm MicroStrategy, brought down by accounting scandals and overblown claims, plunged 62% in a single day; Pets.com (an online pet food retailer) and Webvan (a pioneer in grocery e-commerce) went bust outright… In the panic, nearly everyone blamed the internet itself for being a scam.
Yet the physical infrastructure built up amid speculative excess often ends up nurturing tomorrow’s super-giants—at extremely low cost. Bubbles burst not because internet technology was flawed, but because physical infrastructure construction lagged behind market expectations.
For instance, telecom giants like WorldCom and Global Crossing invested heavily in global undersea fiber-optic cables and dense wavelength-division multiplexing networks—bankrupting themselves, yet laying the groundwork for Netflix, Zoom, and the mobile internet boom. These cheap “information highways” became the perfect incubator for next-generation services.
Without the global telecom infrastructure overinvestment around 2000, YouTube’s video streaming explosion—and later, cloud computing infrastructure—would never have happened.
Amazon is the quintessential example. Its stock price crashed from a 1999 peak of $107 to just $7 in 2001—a drop exceeding 90%. Yet it survived, because its core business logic—“reconstructing retail via the internet”—aligned with the direction of advanced productivity.
This exemplifies Amara’s Law: We overestimate the short-term impact of new technologies and severely underestimate their long-term effects. At the outset of technological revolutions, speculative capital inevitably triggers overinvestment—creating bubbles. This is the “IQ tax” innovation must pay. But once the bubble dissipates, what remains is more resilient, advanced productivity.
02 Why Are Enterprise AI Expenditures Rising—Not Falling?
Fast-forward to 2026, and the AI bubble appears even larger.
Just five major cloud providers—Amazon, Google, Meta, Microsoft, and Oracle—are projected to spend $690 billion on capital expenditures in 2026 alone. Total AI infrastructure investment through 2030 is expected to reach $5.3 trillion. Of that, only ~25% goes toward GPUs—the remaining 75% funds physical infrastructure: liquid cooling systems, power transmission, network switches, optical modules, and land acquisition.
On the revenue side, all leading pure-AI firms—including OpenAI, Anthropic, Cohere, Mistral, and Perplexity—are projected to generate less than $40 billion in combined 2026 revenue.
Nearly $700 billion poured into foundational layers versus just hundreds of millions recovered at the application layer. What else could this asymmetry be—if not a bubble?
We shouldn’t jump to such conclusions too hastily. One critical factor cannot be overlooked:
- In March 2023, when OpenAI launched GPT-4, the blended cost per million tokens was roughly $30.
- By April 2025, thanks to model architecture optimizations and improved inference compute efficiency, the cost for models delivering equivalent intelligence had plummeted to $0.10–$0.15 per million tokens.
According to Stanford University’s AI Index Report and TokenCost data, AI inference costs have fallen by over 99.7% in the past two years.
Under traditional linear logic, plunging costs should reduce enterprise AI spending. Yet reality shows enterprise AI cloud expenditures tripled between 2024 and 2025.
Why?
Because as the marginal cost of “intelligence” approaches zero, AI evolves beyond simple text summarization or conversational assistants—into intelligent agents and multimodal augmented retrieval systems. Enterprises now deploy AI agents to autonomously execute thousands of tasks: writing code, scanning millions of legal contracts, simulating biological experiments.
Cheap tokens unlock massive, previously uneconomical long-tail demand—demand previously constrained solely by cost.
This dynamic becomes clearer when comparing NVIDIA in 2026 with Cisco—the networking hardware leader in 2000. Their ecosystem positions are strikingly similar—but their underlying financial health differs dramatically.
(NVIDIA vs. Cisco: Hard Financial Comparison)
This precisely illustrates the Jevons Paradox in economics: technological improvements in energy efficiency don’t reduce overall energy consumption—in fact, lower costs spur greater demand.
Even after last year’s so-called “DeepSeek Moment,” markets quickly regained clarity within months: the more optimized algorithms become, the lower the barrier to enterprise AI adoption—and total compute consumption rises exponentially.
It is precisely this dynamic that enables AI to embed itself across virtually every legacy industry—just as the internet+ movement transformed every sector over the past two decades. From SaaS software to biopharma, and from embodied AI-powered advanced manufacturing robots, AI+ is now embraced across industries in 2026. No one debates “whether to adopt AI”—instead, executives fret: “Have we cleaned our data? Is our API quota sufficient? Is our RAG architecture optimal?”
Yes, the AI industry currently features a bubble. But for enterprises, refusing to ride the bubble means being crushed by history—a lesson confirmed by two decades of internet evolution.
03 Deep Market Evolution: From Infrastructure to Applications
We stand unquestionably at a pivotal node in the technology lifecycle: just before the “Trough of Disillusionment” on Gartner’s Hype Cycle—or at the turning point described in Carlota Perez’s Technological Revolutions and Financial Capital.
The AI bubble is already bursting—though many haven’t noticed. A few newcomers, armed with a 50-page PowerPoint deck and a thin wrapper around OpenAI’s API, used to raise funding effortlessly. Now, as the tide recedes, concept-only companies—lacking moats or substance—are collapsing en masse.
This self-purification reflects market discipline—and signals bubble deflation. But it’s merely surface-level. Three profound shifts are reshaping the market’s deeper logic:
First, Value Migration from CapEx to OpEx
Today, shovel-makers reap most rewards—NVIDIA, TSMC, and vendors of optical modules and liquid-cooled servers. But as compute increasingly becomes “infrastructure”—as ubiquitous and commoditized as water or electricity—true supernormal profits will gradually shift to the application layer: AI-native enterprises that leverage ultra-low-cost tokens to solve vertical-industry pain points and reshape business processes (OpEx optimization).
Second, Valuation Compression and Earnings Digestion
Markets’ elevated valuations for AI infrastructure don’t necessarily imply imminent collapse. Often, rapid corporate earnings growth absorbs lofty valuations over time—“buying time with space.” So long as cloud giants’ revenue growth keeps pace with depreciation of their capital expenditures, this relay race transforms into an unprecedented industrial upgrade.
- For example, global automotive and semiconductor manufacturers have adopted end-to-end AI digital twin technology—cutting new product development-to-production cycles by 35% and boosting overall equipment effectiveness by 18%.
- In finance, quantitative trading, risk management, and credit assessment are now fully driven by multimodal AI agents in 2026. AI processes macroeconomic expectations at microsecond timestamps—and deeply participates in every micro-level asset pricing decision.
- In highly specialized fields like law, healthcare, and auditing, AI has evolved from “junior assistant” to “partner-level expert.”
Among over one billion active users of ChatGPT, Gemini, and Claude, a significant portion relies on them daily as substitutes for intensive cognitive labor—including you and me. All of this is already happening—and visible to all.
04 Conclusion
Looking back across the turbulent history of technology, Schumpeter’s “creative destruction” continues unabated.
Capital markets are inherently impatient—always hoping to invest $1 today and earn $10 tomorrow. When nearly $700 billion in infrastructure investment fails to yield immediate application-layer profits, markets inevitably undergo brutal consolidation—eliminating PPT-only speculators and shell companies, preserving only those with genuine technical depth and real-world deployment scenarios.
Post-consolidation, vast, low-cost compute centers and highly optimized model algorithms will serve countless industries at extremely affordable prices.
After 2000, humanity entered the digital era—where no industry functions without the internet. Today, we’re irreversibly advancing toward an era of intelligent abundance—where every industry is governed and empowered by AI.
Amidst the bubble’s noise, the underlying productivity momentum remains utterly real—without a trace of dilution.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
