First, the obvious one: we do not have or want government guarantees for OpenAI datacenters. We believe that governments should not pick winners or losers, and that taxpayers should not bail out companies that make bad business decisions or…
The Illusion of the AI Bubble: Sam Altman’s High-Stakes Bet on the Future
Sam Altman’s recent reflections on X (formerly Twitter) are more than a defense of OpenAI’s spending—they are a manifesto for the future of civilization. Without ever uttering the word “bubble,” Altman implicitly dismisses the accusation that the artificial intelligence boom is another speculative mania. Instead, he positions OpenAI’s trillion-dollar expansion as a rational response to the tectonic economic and technological transformation underway.
At stake is not just OpenAI’s valuation or Nvidia’s next earnings call—it’s whether humanity is overbuilding a dream or underpreparing for destiny.
Altman’s Argument: Betting on the Infinite Game
Altman’s post paints OpenAI as the architect of a coming “AI-powered economy.” He justifies an eye-watering $1.4 trillion in infrastructure commitments over the next eight years, backed by projections of exponential revenue growth—from over $20 billion in annualized run rate today to hundreds of billions by 2030.
The logic is simple but audacious: if intelligence is the ultimate production function, investing in compute—the new oil—will yield compounding returns across every industry. He envisions AI spilling beyond text generation into enterprise tools, robotics, AI-powered hardware, and scientific discovery, where machines accelerate the pace of human knowledge itself.
Altman’s message echoes the industrialists of previous eras—Ford, Edison, Jobs—each mocked for building too much, too fast. “The greater risk,” he implies, “is not excess but insufficiency.” If humanity underbuilds, the shortage of compute, data, and electricity could throttle innovation for decades.
This is not bubble talk—it’s infrastructure talk. He argues that we’re laying the digital equivalent of railroads across the global economy. And just as no one today calls the railroad boom of the 19th century a bubble, Altman suggests that history will view the AI era as a foundational overbuild—necessary, inevitable, and transformative.
The Bubble Thesis: Echoes of Tulips and Dot-Coms
Yet critics see a different story—one more tulip, less transistor. The skeptics argue that the AI frenzy has classic bubble traits:
Runaway capital flows:
AI infrastructure spending has reached levels 17 times greater than the dot-com boom and four times higher than the subprime crisis. Venture capitalists, sovereign funds, and Big Tech giants are all flooding the same sector, often in circular arrangements—Nvidia funds startups that then buy Nvidia’s chips, creating what one analyst called “the world’s most sophisticated self-licking ice cream cone.”
Limited real-world adoption:
Beyond a few dazzling demos, many AI tools remain novelties. Productivity gains are marginal, enterprise adoption is slower than expected, and small businesses find few reliable use cases. As one economist quipped, “If you subtract AI from the U.S. economy, GDP growth is flat.”
Ecological and social strain:
Data centers devour water and electricity, drawing community protests from Arizona to Ireland. If the hype collapses, society could be left with ghost factories of compute—monuments to digital excess.
Concentration risk:
The entire ecosystem hinges on a handful of players—Nvidia, Microsoft, OpenAI, Anthropic. If one falters, contagion could ripple through markets, just as dot-com overbuilds led to the telecom bankruptcies of 2001.
The imagery is eerily familiar: lavish valuations, vaporware startups, and speculative capital chasing exponential promises. Over half of investors in recent surveys believe AI is already in a bubble. The skeptics warn that even a temporary cooling could wipe out trillions in market value.
The Counterargument: This Time Is (Partly) Different
But anti-bubble advocates—Altman among them—argue that comparing AI to tulips or Pets.com misses the point. Unlike past speculative frenzies, AI is already reshaping the economic landscape.
Real revenue: Microsoft, Amazon, and Google have reported double-digit growth in AI-related cloud services. AI is no longer a promise; it’s a product.
Structural demand: Every major corporation is retooling workflows for automation, analytics, and co-pilots. AI is not an optional luxury—it’s the new electricity.
Scientific revolutions: From protein folding to materials discovery, AI is accelerating frontiers of science that could redefine medicine, energy, and agriculture.
Compute scarcity: Paradoxically, the very shortages of chips and GPUs suggest underinvestment, not excess. If this were a bubble, supply would be glutted and demand tepid. Instead, it’s the reverse.
Even Federal Reserve Chair Jerome Powell has distinguished AI from the dot-com era, calling it a “real-economy transformation” rather than speculative exuberance.
In short: AI may be overheated, but it is not hollow. The steam comes from engines that actually turn.
A Tale of Two Economies: Speculation and Substance
To understand the paradox, think of AI as a double helix of speculation and substance. One strand is financial—the frenzy of funding, valuation, and narrative. The other is technological—the slow, irreversible diffusion of capability. These strands twist around each other, creating both volatility and vitality.
Yes, there are frothy segments—startup valuations untethered from revenue, circular investments, and “AI-washing” by companies desperate to ride the trend. But there is also deep substance: the quiet embedding of AI into logistics, law, education, and healthcare, in ways that will outlast market cycles.
Every great technological leap has gone through this cycle. The dot-com crash destroyed billions but birthed Amazon and Google. The railway mania bankrupted investors but built the arteries of modern commerce. Even the electrification bubble of the 1890s looked wasteful—until the lights stayed on.
AI’s current overbuild may look reckless in quarterly earnings reports, but in historical hindsight, it may prove to be civilization’s most necessary overreach.
The Metaphors of Momentum: From Steam Engines to Neural Nets
The tension between overbuilding and underinvesting is as old as progress itself. The Victorians built steam railways faster than they could populate towns; NASA built rockets before having anywhere to go; Silicon Valley builds models before society is ready to use them.
Altman’s trillion-dollar bet is part of that lineage—an act of faith that the infrastructure of intelligence must precede the age of intelligence. His calculus is Promethean: even if the fire burns a few hands, humanity must still light it.
To dismiss AI as a bubble is to mistake early turbulence for terminal failure. The Wright brothers didn’t prove aviation sustainable by showing a profit; they proved it by staying in the air.
Conclusion: The Necessary Overbuild
So, is AI in a bubble? The answer depends on your time horizon. In the short term, yes—there will be corrections, bankruptcies, and hubris punctured by reality. Some of today’s “AI unicorns” will evaporate as quickly as the dot-coms did.
But in the long term, AI is not a tulip or a mortgage-backed illusion. It is the next substrate of civilization, a general-purpose technology as foundational as electricity or the internet.
Altman’s trillion-dollar ambition may sound reckless, but history often rewards the reckless who build the future rather than those who fear it. The real question is not whether AI is a bubble—it’s whether humanity can afford not to overbuild the mind of its next industrial age.
Like the cathedrals of medieval Europe, the great AI infrastructures of today are monuments to faith—faith that intelligence, once ignited, will illuminate the world rather than consume it.
In the late 1990s, the Internet was real—astonishingly real. It was already changing how humans communicated, learned, and traded ideas. By 1994, early adopters were sending emails and building websites. By 1996, search engines were mapping the digital frontier. By 1998, Amazon and Google were born. By 1999, e-commerce had arrived. By 2000, the dot-com boom had turned into mania. And by 2001, it crashed. Hard.
But the Internet didn’t die. Pets.com did. The infrastructure remained; the potential was intact. After the “nuclear winter” of the early 2000s, the Internet roared back—stronger, more efficient, and foundational to everything that followed.
Today, we are in a similar moment with artificial intelligence.
The AI Moment Is Real—Bigger Than the Internet
AI is not a fad, not a passing storm. It is a platform shift—a new electricity. The same way the Internet transformed communication, AI is transforming cognition itself. It will not merely change how we use computers; it will change what computers are.
AI can already write, see, listen, summarize, reason, translate, and code. It is already embedded in search, healthcare diagnostics, logistics, design, and education. The generative layer is just the beginning; autonomous systems, multimodal reasoning, and embedded intelligence will follow.
If the Internet was about connecting information, AI is about connecting intelligence.
The Coming “Mini-Crashes”
However, the path forward is not a straight line.
The Internet’s dot-com crash wiped out thousands of startups with no real business model. Most didn’t fail because the Internet wasn’t real—they failed because their businesses weren’t real. Pets.com, the poster child of that era, was selling dog food online with no viable logistics model and no profits.
AI will go through the same pruning process. Some companies are building enduring technology and infrastructure. Others are riding hype. “Pets.AI” startups—those that exist only because the word “AI” attracts capital—will collapse.
Many will raise huge sums, make viral demos, and vanish within 24 months. There will be rounds of layoffs, rebrandings, and pivots. Investors will lament an “AI winter.” But the real story will be quiet and steady—AI embedding itself into every workflow, device, and decision.
Fundamentals Never Change
Every technological revolution feels like a suspension of economic gravity. But gravity always returns.
Businesses must make money. They must create value greater than their costs. Venture capital can buy time but not immortality. Hype can amplify early growth but cannot sustain it. The companies that survive will do so for the same reasons Google, Amazon, and Apple survived: product-market fit, revenue, adaptability, and execution.
AI will be no different. The winners will build things that people actually need—tools that save time, reduce costs, improve decisions, or create joy. The losers will build shiny demos without a path to profit.
The Pets.AI Warning
The phrase “Pets.AI” will soon become shorthand for hype cycles gone wrong. For every OpenAI or Anthropic, there will be hundreds of startups promising “AI for everything” without solving anything.
History doesn’t repeat, but it rhymes:
1999: “Everyone needs a website.”
2025: “Everyone needs an AI model.”
In both eras, the claim is partly true—but the value lies not in having technology, but in using it meaningfully.
A company deploying AI to reinvent logistics, diagnostics, or design may thrive. But one building “AI for AI’s sake” will burn out fast.
The Real Gold Rush Is Still Ahead
AI’s true impact will emerge after the hype has cooled. Once infrastructure is stable and capital has retreated, enduring builders will remain. The next Google, Amazon, or Salesforce of the AI era is still being born—likely in some small lab, research group, or startup garage.
AI is bigger than the Internet because it is not a new network—it’s a new nervous system. It won’t merely connect people; it will connect ideas, decisions, and intelligence itself.
The dot-com crash was not the end of the Internet. It was the Internet growing up. Likewise, the coming correction in AI will not mark its demise—it will mark its maturity.
Conclusion: Real Technology, Real Discipline
AI is real. The hype is also real. The difference lies in discipline.
The future will reward those who treat AI not as a lottery ticket, but as infrastructure—who focus on building, serving, solving, and sustaining. The rest will join the graveyard of Pets.AI: companies that mistook temporary excitement for permanent transformation.
The Internet didn’t die in 2001. It conquered the world by 2005.
The Rise and Fall of Pets.com: When America’s Love for Dogs Met the Internet Gold Rush
America has always loved its pets. Dogs and cats are not just animals—they are family. To millions of Americans, a pet is a child, a confidant, a companion. You can’t buy dog meat in America because the very idea feels unthinkable. The dog, in many ways, is the American cow—sacred not in religion but in sentiment. This cultural truth sits deep in the national psyche.
And then came the Internet—the biggest technological revolution since electricity. For the first time in history, anyone could sell anything to anyone, anywhere. The dot-com era of the late 1990s was the digital gold rush, and it created a perfect storm of emotion and innovation.
At the heart of that storm sat Pets.com, a company that combined America’s love for animals with the world’s excitement about the Internet. It was, on paper, an unbeatable combination. But in reality, it became the most famous crash of the early Internet age—a cautionary tale that still echoes today in every tech bubble, including AI.
The Perfect Storm of Hype
In 1998, Pets.com launched with a simple idea: sell pet supplies online. Food, toys, leashes, collars—anything for your dog or cat, delivered right to your door. For pet lovers, it was a dream. For investors, it was destiny.
The timing was ideal. America’s pet industry was booming, the Internet was expanding, and venture capital was flowing freely. Pets.com quickly became a media darling. It had a cute logo, a catchy domain name, and a sock-puppet mascot that starred in Super Bowl commercials.
It wasn’t selling technology—it was selling love.
But underneath the glossy branding and national ad campaigns was a business that didn’t make sense.
When Marketing Outran Math
Pets.com spent tens of millions of dollars on marketing—celebrity endorsements, cross-country tours, and high-profile ad spots—before proving it could make a profit. Its costs were astronomical: shipping 40-pound bags of dog food across the country for less than the store price, all while offering discounts and free delivery.
The more it sold, the more money it lost.
Investors didn’t care—at least not yet. In the fever of the dot-com boom, eyeballs mattered more than earnings. Growth was the only metric that counted. Pets.com went public in February 2000 with massive hype. But within nine months, it was bankrupt.
The company’s stock went from $11 a share to 22 cents. The sock puppet was silenced.
The Deeper Lesson: Emotion Isn’t a Business Model
Why did Pets.com fail so spectacularly?
Because it mistook emotion for economics.
America’s affection for pets was real. The Internet was real. But the connection between those two realities was not a sustainable business. You cannot ship bulk pet food at a loss forever and expect to make it up on volume. The dream was beautiful—but the math was brutal.
The collapse of Pets.com became the defining symbol of the dot-com bubble, teaching a generation of entrepreneurs that branding and buzzwords cannot replace business fundamentals.
The Cultural Collision
Pets.com wasn’t just a company. It was a cultural collision—between a country’s emotional values and a new technological frontier.
The Internet promised to democratize commerce. Pet culture promised endless love and loyalty. But business requires something else entirely: profitability.
In the end, America’s love for pets couldn’t save Pets.com from the cold logic of the market.
Why It Still Matters — The “Pets.AI” Parallel
Fast forward to today, and history is repeating itself in another form. The new gold rush is AI. Every startup wants to add “AI” to its name, raise millions, and promise disruption. Just as “dot com” once guaranteed excitement, “.AI” now guarantees attention.
But, as with Pets.com, many of these ventures are chasing hype, not value. They mistake cultural fascination (AI as magic) for economic viability.
AI is real—just as the Internet was real.
But “Pets.AI” startups—those built on marketing buzz instead of business fundamentals—are heading for the same crash.
The Enduring Truth
The story of Pets.com is not about dogs or data. It’s about discipline.
Technology can amplify emotion, but it cannot replace sound judgment. Consumers can love your brand, but they must also need—and pay for—what you sell.
The Internet didn’t die after the dot-com crash. It matured.
AI won’t die after its coming corrections. It will evolve.
But in every era, one rule remains unbroken: Love your product all you want—but make sure it loves you back on the balance sheet.
The Coming AI Glut: When Abundance Meets a World Built on Scarcity
In every technological revolution, there are the Pets.coms—the overhyped ventures that burn bright and vanish—and there are the Ciscos, Lucents, and undersea cables—the invisible infrastructure builders that survive the storm and shape the next age.
During the dot-com boom of the late 1990s, the world overbuilt the Internet. Fiber-optic cables wrapped the planet. Data centers mushroomed. Equipment manufacturers couldn’t keep up with demand. For a brief moment, there was a glut—too much capacity chasing too few users. But within a decade, that “excess” became woefully insufficient for the rise of YouTube, Facebook, cloud computing, and streaming.
History doesn’t repeat, but it rhymes. The same pattern is forming with artificial intelligence.
The Birth of the AI Glut
The world is in the middle of an AI infrastructure arms race. Tech giants are ordering GPUs by the millions. Data centers are expanding like new cities. Electricity demand is spiking. Nations are building sovereign compute reserves. The numbers are staggering—tens of billions of dollars invested every quarter in chips, models, and data pipelines.
To an outside observer, this looks like overbuilding—too much, too fast. And in the short term, it may well be. There will be idle clusters, half-trained models, and power-hungry servers waiting for real workloads.
But the mistake would be to confuse short-term saturation with long-term futility. Just as the Internet’s fiber glut of 2000 became the foundation for the digital explosion of 2010, today’s AI glut will one day look tragically inadequate for the demands of the 2030s.
The real risk is not in overbuilding AI capacity. It is in underthinking what AI means for civilization itself.
The Unasked Questions
AI is not just another wave of automation or efficiency. It challenges the core logic of our economic and political systems.
The industrial and digital revolutions expanded human capacity but kept the basic framework intact: scarcity. Goods, labor, and opportunity remained limited; value came from managing that scarcity efficiently.
AI breaks that logic. It promises abundance—of knowledge, design, computation, and creativity. A single person with AI tools can now do the work of a hundred. Entire industries can be automated at near-zero marginal cost. The question is no longer, “How do we produce more?” but “What happens when production is no longer the constraint?”
Our systems—economic, legal, political—are not built for that world.
A World Built for Scarcity
The global economy still runs on scarcity economics.
Scarcity gives money meaning. It gives jobs necessity. It gives governments power.
But AI inverts all that.
When information, creativity, and even intelligence itself become infinitely reproducible, traditional notions of ownership and control start to fracture.
Today, we treat AI like another commodity market—data centers, chips, and cloud credits. But that is like treating the early Internet as just a collection of phone lines. We are building abundance infrastructure within scarcity institutions.
That is where the collision is coming.
The WTO Analogy
When the World Trade Organization (WTO) was formed in 1995, it reflected the world as it was then:
a system of nations trading goods across borders.
But today, power and productivity no longer sit neatly within nation-states.
A handful of companies—OpenAI, Google, Anthropic, NVIDIA, Amazon, Tencent, Baidu—already wield influence equal to or greater than many governments.
If you were to design a global coordination system for AI today, it wouldn’t just be an agreement between countries.
It would have to include companies, individuals, and algorithms themselves—because power has decentralized that far.
AI is not just reshaping the economy; it is redefining governance.
The Real Challenge
The danger is not that AI will run out of money or momentum. The danger is that we will use it to reinforce old systems rather than build new ones.
We are pouring trillions into GPU farms, but how much thought are we giving to:
What happens to work when most labor becomes optional?
How should wealth be distributed when productivity is near-infinite?
What rights should algorithms have, if they act autonomously on our behalf?
How do we build global coordination when borders no longer define power?
We are investing in compute, not philosophy. In power, not purpose.
Abundance vs. Scarcity
AI’s promise is abundance. But humanity still behaves as if trapped in a scarcity economy.
We hoard data. We gate access. We monetize attention.
Abundance means there is more than enough intelligence, creativity, and possibility to go around.
Scarcity economics says someone must always lose for another to win.
As long as we cling to that zero-sum mindset, AI will magnify inequality rather than eliminate it.
The winners of this age will not be those who own the most GPUs,
but those who reimagine the systems of value and governance that can sustain abundance.
The Glut We Need
An AI glut is inevitable—and even necessary.
Like the fiber-optic cables that once lay dark under the oceans, today’s GPU clusters will form the neural backbone of the next civilization.
But infrastructure alone is not wisdom.
If we build abundance without reforming the systems that still reward scarcity, we will create not a new enlightenment—but a new imbalance.
The question is not how much AI we can build,
but what kind of world we will build with it.
That, not the number of data centers, will decide whether this AI revolution ends in collapse—or in collective awakening.
The Real AI Glut: When Abundance Meets Scarcity’s Final Battle
It is not true that the world is building too much AI infrastructure. In fact, even at the current pace, the expansion is likely insufficient for what the next decade will demand. But an AI glut is still coming—not because the physical capacity will exceed need, but because that capacity will collide head-on with our existing scarcity-based institutions and paradigms.
Fiber-optic cables, GPUs, and data centers are not the problem. The real bottleneck lies in the software of civilization: our economic, political, and social operating systems, all of which are built on the assumption that scarcity is permanent.
The Misdiagnosis of Overbuild
Critics warn that the world is overbuilding AI—too many chips, too many data centers, too much compute. But this argument mistakes short-term utilization for long-term necessity. Every great technological leap—from railways to electricity to the Internet—looked like overbuilding at first. The infrastructure always outpaces the imagination.
We do not have too much compute; we have too few new institutions to make full use of it. We are still trying to fit infinite intelligence inside finite economic models.
The Real Collision: Abundance vs. Scarcity
AI represents abundance: of knowledge, creativity, insight, and production. With AI, marginal costs approach zero. A single individual can now do the work of hundreds; a small firm can operate at global scale.
But our institutions—governments, corporations, labor markets—exist to manage scarcity. They assume limited goods, limited opportunities, and limited control. Their hierarchies depend on constraint.
The result is inevitable tension: abundance infrastructure colliding with scarcity institutions.
For example:
Education systems still ration learning through degrees, even as AI can teach every child individually.
Economies still tie income to jobs, even as AI automates labor.
Politics still treats information as power, even as open models can democratize knowledge.
AI is not overbuilt; society is under-redesigned.
The Coming Glut
The “AI glut” will appear not in compute capacity but in blocked potential. We will have more intelligence, more data, and more automation than our economic and political systems can process.
Imagine data centers running at half capacity while millions remain unemployed—not because the AI isn’t capable, but because laws, markets, and institutions can’t adapt fast enough to let abundance flow.
This mismatch—between what AI can produce and what the system allows—will look like oversupply. It will feel like stagnation. But it will actually be a crisis of imagination, not of engineering.
The Last Stand of Scarcity
Scarcity paradigms will not surrender easily. The entire logic of taxation, ownership, wages, and even identity is rooted in limitation. Every established power structure—corporate, political, financial—depends on scarcity to justify its existence.
So, as AI pushes toward abundance, expect resistance:
Legal fights over data access and model ownership.
Political backlash against automation and digital citizenship.
Economic friction as elites try to re-monetize abundance through artificial scarcity—subscriptions, patents, or walled gardens.
Scarcity will lose eventually, but not without a fight. And that fight will define the next decade.
After the Clash
The end of scarcity institutions will not come through collapse but through obsolescence. Once abundance becomes undeniable, the frameworks of limitation will fade naturally. New systems—open, decentralized, participatory—will rise to manage shared intelligence rather than restricted property.
The transition will be chaotic but creative. It will resemble the shift from monarchies to democracies, or from print to digital: painful for the old order, liberating for everyone else.
The Takeaway
The world is not overbuilding AI. It is under-preparing for abundance.
The real glut will not be in silicon, but in possibility—too much intelligence for a world still clinging to artificial scarcity.
And when abundance finally breaks free from those old constraints, the so-called AI glut will reveal itself for what it truly is: the birth pain of a post-scarcity civilization.
OpenAI–AMD: The Most Circular Deal in AI History — Bold Innovation or Bubble 2.0?
Overview
The recently announced OpenAI–AMD deal (October 6, 2025) represents one of the most ambitious—and potentially destabilizing—arrangements in the emerging trillion-dollar AI economy. Under the agreement, OpenAI commits to purchasing hundreds of thousands of AMD Instinct AI accelerators to power up to 6 gigawatts (GW) of data-center capacity—enough energy to run multiple hyperscale clusters rivaling Microsoft or Google.
In exchange, AMD has issued OpenAI warrants for up to 160 million shares—roughly 10% of AMD’s common stock—at a nominal price of one cent per share, vesting in phases tied to deployment milestones beginning with the first gigawatt of compute.
This structure has stunned analysts for its scale, risk symmetry, and financial circularity. It deepens AMD’s challenge to Nvidia’s dominance while giving OpenAI unprecedented leverage in chip supply—yet it also raises the specter of dot-com-style “circular financing”, where capital appears to multiply without corresponding value creation.
The Circular Dynamic: How the Loop Works
1. Mutual Dependency and the Value Loop
The relationship forms a self-reinforcing cycle:
OpenAI’s demand for chips inflates AMD’s sales and valuation.
AMD’s rising stock makes OpenAI’s warrants more valuable.
OpenAI’s gains can fund additional chip purchases—fueling AMD’s growth further.
This is not hypothetical: following the announcement, AMD shares surged over 30% in two days, briefly adding more than $80 billion in market capitalization, and turning OpenAI’s warrants—still unvested—into a potential windfall worth several billion dollars.
2. Incentive Alignment with Hidden Volatility
The deal aligns both companies toward rapid deployment. But this alignment also magnifies downside risk. If AI model monetization slows, AMD’s stock could plummet, wiping out OpenAI’s paper gains and constraining its purchasing ability—creating a feedback spiral of contraction.
Circular Finance Across the AI Industry
The OpenAI–AMD pact is not an isolated case. Across the AI sector, “circular investments” are multiplying—where chipmakers, cloud providers, and AI labs invest in one another in exchange for exclusive purchasing commitments.
Below are the most prominent examples:
Example
Key Details
Circular Logic
OpenAI–Nvidia (Sept 2025)
OpenAI to deploy Nvidia systems for up to 10 GW; Nvidia invests up to $100 billion in OpenAI.
$4 billion stake; Anthropic partly hosted on Google Cloud.
Google’s equity grows with Anthropic’s spend.
Nvidia–xAI (Musk)
$2 billion equity/debt; xAI uses Nvidia GPUs for Colossus 2.
Nvidia invests → sells its own chips back → revenues inflate.
Nvidia–CoreWeave
Nvidia owns a stake; CoreWeave rents Nvidia GPUs to OpenAI.
Nvidia profits twice—via ownership and sales.
OpenAI–Oracle–Nvidia Triad
OpenAI trains on Oracle Cloud; Oracle buys Nvidia chips; Nvidia invests in OpenAI.
A full three-way financial loop.
These arrangements have accelerated innovation but also drawn comparisons to the late-1990s telecom and dot-com vendor-financing boom, when suppliers lent to customers to buy their own products—masking real demand and inflating balance sheets.
What Is Circular Financing—and Why It Matters
Circular financing occurs when companies invest in one another or offer financial incentives tied to reciprocal business transactions—creating the illusion of organic growth.
In AI, this often means:
AI labs pre-paying or committing to chip purchases worth billions.
Chipmakers or cloud firms granting equity, warrants, or credit back to those labs.
Both sides recording growth—even though no external cash or customer demand enters the system.
This technique turbocharges expansion but risks synthetic growth—capital spinning within a closed loop of the same players.
Risks Embedded in the AI Circular Economy
1. Bubble Formation and Overvaluation
AI valuations are now reminiscent of the Cisco-era bubble of 2000.
Example: Nvidia’s market cap briefly surpassed $4.5 trillion, while total AI circular deal volume exceeded $1 trillion—despite uncertain monetization.
OpenAI alone is projected to burn through $115 billion by 2029, a rate unsustainable without perpetual capital recycling.
2. Systemic Fragility
The AI ecosystem now resembles a financial network—highly leveraged through interdependence.
If one link (say, a delay in OpenAI’s next model) breaks, it could reverberate through Nvidia’s revenues, AMD’s stock, Oracle’s margins, and even global data-center energy planning.
3. Distorted Market Signals
Because purchases are incentivized by equity, not demand, investors cannot accurately gauge real adoption. This can lead to over-production of GPUs and overbuilding of data centers, echoing the “dark fiber” glut after 2001.
4. Regulatory Red Flags
As the “Magnificent 7” dominate both AI compute and investment flows, FTC and DOJ scrutiny is increasing. Regulators are probing potential conflicts of interest, market concentration, and accounting opacity.
5. Operational Constraints
Even if financially sound, the ecosystem faces physical limits:
Power shortages in Texas, Virginia, and Ireland.
Cooling and land constraints near major cloud hubs.
Escalating water usage—each GW data center consumes ~1 billion liters annually.
Dรฉjร Vu: Lessons from the Dot-Com and Telecom Bubbles
The late-1990s internet boom offers an instructive analogy.
Aspect
Dot-Com Era (1995-2000)
AI Era (2023-2025)
Similarities
Differences
Hype Metrics
“Eyeballs,” page views
“Tokens,” compute
Both value attention, not profits
AI leaders (e.g., Nvidia, MSFT) are profitable
Infrastructure Glut
80 million miles of dark fiber
GW-scale data centers
Overbuilding ahead of demand
Physical constraints limit overbuild
Vendor Financing
Lucent lent $8 B to its customers
Nvidia invests $100 B in its customers
Circular revenue illusion
Equity-based, not debt-based
Revenue Models
Ads, speculative startups
SaaS, cloud, enterprise AI
Speculation
Enterprise integration gives staying power
Crash Outcome
78% Nasdaq fall, $5 T loss
— (pending)
Risk of correction
Stronger balance sheets, real use cases
The dot-com bust wiped out trillions, but its survivors—Amazon, Google, eBay—eventually justified the underlying thesis: the internet was transformative.
AI may follow a similar trajectory—an overheated build-out followed by consolidation and enduring impact.
Could This Time Be Different?
There are reasons to believe the AI economy might weather turbulence better:
Cash-Rich Giants — The top AI players collectively hold $380 billion in reserves, unlike debt-laden dot-coms.
Tangible Productivity Gains — Enterprises are integrating AI copilots, agents, and automation—generating measurable efficiency.
Physical Scarcity as a Brake — Unlike software bubbles, energy and chip constraints naturally cap overinvestment.
Public–Private Synergy — Governments from the U.S. to India are investing in AI infrastructure, cushioning demand cycles.
Still, even “different” cycles can crash if return on investment fails to match capex. The AI industrial complex could face a correction more financial than technological—with valuations compressing even as innovation continues.
The Broader Macro Picture
AI circular financing reveals a deeper shift in capitalism itself:
capital no longer flows linearly from investor to enterprise to market—it loops within ecosystems, reinforcing strategic moats and data monopolies.
In this system:
Value is derived from compute capacity, not traditional profitability.
Equity acts as currency, not just ownership.
Growth is self-referential, feeding on expectations more than outcomes.
Whether this represents a “new financial physics” or merely bubble mechanics 2.0 depends on one question:
Will AI’s trillion-dollar capex yield real economic productivity or just larger GPU stockpiles?
Conclusion: Innovation at the Edge of a Bubble
The OpenAI–AMD partnership exemplifies the new AI economy’s dual nature—visionary yet volatile. It fuses industrial-scale ambition with financial engineering reminiscent of past manias.
If AI delivers sustained breakthroughs—in software agents, energy-efficient models, and enterprise productivity—this could be remembered as the “infrastructure boom before the age of abundance.”
But if returns lag, it may go down as the largest circular bubble since 2000—a mirror reflecting the same human optimism, leveraged through silicon instead of fiber.
Image via WikipediaWatch the first video here. Fred has come around to using my word: froth.
But then I have also inched a little in Fred's direction. I recently used the phrase mini bubble burst.
I think we are both still blind men trying to figure out the elephant. But at least we are trying to figure out. Most others are simply making wild jumps.
Some real wealth creation is taking place. But every new emerging sector will necessarily be accompanied by froth. That is the nature of the beast.
There will be a slew of sob stories in a few years but I don't see any imminent collapse. We are in the first year of a boom decade.
Fred has said repeatedly that what we are seeing is a bubble. First thing I say is this is not a yes no question. Is this a bubble? If you force ask me, my answer is no. This is not a bubble. This is hyperactivity. Will many angel investors lose money? Sure. But that does not make it a bubble. Even a top notch VC like Fred Wilson expects one third of his portfolio to go down under. And these are companies that he did not invest in on day one knowing they will go down. You think you picked a winner, you give them sufficient money and guidance, you go to bat for them, and they still go down. If Fred Wilson is at peace with a 33% failure rate, there are VCs whose failure rates are 66% and 90%. Most VCs fail. Most entrepreneurs fail. By some estimates as many as 90% of new businesses fail within a year of getting launched. Looks like 10% is all capitalism needs to survive.
