The A.I. Bubble: Why this time is actually different (and why that still won’t save everyone)
“What we learn from history is that people don’t learn from history.”
- Warren Buffet
The Ghost of Bubbles Past
Have you seen it? The knowing glances exchanged between veterans of the dot-com era as they watch valuations soar beyond reason. Circular deals proliferating along with promises of revolutionary change to justify the astronomical spending. We’ve seen this movie before – and we remember how it ended.
Yet this time, everyone insists, it’s different.
Welcome to 2025, and artificial intelligence has captured the world's imagination (and capital) in ways that make the internet boom look quaint. Venture capitalists have poured approximately $193 billion into AI startups this year alone[1], birthing a new generation of unicorns (481 to be exact) with a collective valuation of $1 trillion[2]. At the center of this maelstrom sits OpenAI, a company that didn't exist in its current form five years ago, now valued at $500 billion despite burning through $8 billion annually[3],[4] and projecting cumulative losses of $44 billion through 2028[5].
OpenAI's CEO, Sam Altman, has become the face of a revolution that promises to transform everything from how we work to how we think. His company's ChatGPT has captured 500 million weekly users faster than any application in history[6], each interaction burning through expensive computing power that keeps the company deeply unprofitable. Yet investors keep writing checks, each one bigger than the last, in a spectacular bet that artificial intelligence will reshape the global economy.
The parallels to another time, another boom, are impossible to ignore. Trust me, I was there.
Back to the Future
Dig if you will this picture…. It’s the late 90s when the internet was young, full of promise, and everyone wanted a piece of the “digital frontier”. The Telecom Act of 1996 had just unleashed a gold rush as companies raced to lay fiber optic cables and point to multi-point microwave radios that would carry humanity into the information age. Over the next five years, telecom companies would invest more than $500 billion - mostly borrowed money - into building what they believed would be the nervous system of the new economy[7].
The believers had compelling math on their side. Industry leaders, including WorldCom's executives, claimed internet traffic was doubling every 100 days. At that rate, no amount of infrastructure could be too much. Companies like Global Crossing and Level 3 borrowed billions to wrap the planet in fiber optic cables, and Advanced Radio Telecom Corp and Winstar poured hundreds of millions into point to multi-point wireless networks, their stock prices soaring as investors bought into the vision of infinite demand[8].
But the math was wrong. Catastrophically wrong.
Internet traffic was actually only doubling once a year, not every hundred days. By the time reality set in, companies had laid enough fiber to circle the Earth thousands of times over. When the music stopped in 2001, the carnage was swift and brutal. WorldCom filed for what was then the largest bankruptcy in U.S. history. Global Crossing, which had been worth $47 billion at its peak, collapsed into bankruptcy within weeks of posting a $3.4 billion quarterly loss[9].
By 2004, an estimated 85% to 95% of the fiber laid during the boom sat unused - "dark fiber," they called it - a monument to technological and financial hedonism.
The New Gold Rush
Fast forward to today, and the AI boom makes those telecom investments look almost modest. Microsoft, Amazon, Google, and Meta alone are committing close to $350 billion this year to data center construction[10]. The scale is breathtaking: U.S. data center investment is running at roughly $200 billion annually, transforming rural counties and small towns into crucial nodes in the AI revolution[11].
But here's where the two stories diverge. During the fiber optic frenzy, companies installed capacity far ahead of demand, betting on future growth that never materialized. Today's tech giants face the opposite problem: they can't build fast enough. Amazon, Microsoft, and Google all report that demand for computing power is outstripping their ability to supply it. Every Monday morning, OpenAI's CFO Sarah Friar joins a call with Microsoft with the same urgent request: more compute, always more compute[12].
The numbers tell a story of genuine adoption, not just speculation. OpenAI is projecting $12.7 billion in revenue for 2025, up 243% from the previous year[5]. Unlike the dot-com era's "eyeball" metrics and vague promises of future monetization, AI companies are generating real revenue from real customers solving real problems.
The question now is, will the ultimate revenues actually make the capital investment a return driven endeavor?
The Oracles Speak
Not everyone is convinced this boom is different. Warren Buffett's longtime partner Charlie Munger, before his passing, warned about the "mania" surrounding AI investments. Ray Dalio, the hedge fund titan who predicted the 2008 financial crisis, has drawn explicit parallels between today's AI exuberance and the dot-com days[12]. Even Sam Altman himself acknowledged in August that we might be seeing the "beginnings of a bubble"[13].
The warning signs are hard to ignore. Consider the bizarre circularity of recent deals: OpenAI is taking a 10% stake in AMD while simultaneously buying their chips. Nvidia is investing $100 billion in OpenAI while OpenAI purchases billions in Nvidia hardware. Microsoft funds OpenAI, which buys computing power from Microsoft's Azure, which runs on chips that Microsoft helps finance[14]. It's a dizzying web of interconnected bets, each company simultaneously customer, supplier, and investor to the others.
"It looks eerily like the circular financing arrangements of the late 1990s," notes Jeffrey Sonnenfeld of Yale, who's studied market bubbles for decades[14].
When everyone is simultaneously a customer, supplier, and investor, you're not building a market—you're building a house of cards.
But The Economics Don't Add Up – Yet…
Perhaps the most troubling aspect of the AI boom is the fundamental economics. Traditional software companies follow a well-known trajectory: high upfront costs give way to massive margins as they scale. Microsoft Windows costs the same to develop whether it sells one copy or one billion. Facebook's costs actually decreased as it grew, spreading fixed expenses across an ever-larger user base.
AI breaks this model completely.
Every ChatGPT query costs money in computing power. Every new model requires massive investments in training. OpenAI's gross margins hover around 40%, constrained by the variable costs of computation[15]. The company burns through $700 million monthly on computing resources alone15. Unlike software, where the millionth user costs nothing to serve, AI's millionth user costs almost as much as the first.
This reality has led to some spectacular financial gymnastics. OpenAI, despite its $500 billion valuation, expects to lose $14 billion by 2026[5]. The company's recent $40 billion funding round—the largest in startup history - included $30 billion contingent on converting from a nonprofit to a for-profit corporation by year's end[15]. Investors are essentially betting that OpenAI can completely restructure its fundamental economics before the money runs out.
The Bulls' Clap Back
Yet for all the skepticism, the bulls have compelling arguments of their own. Goldman Sachs economists estimate that AI could unlock $8 trillion in productivity gains for the U.S. economy alone, with plausible estimates ranging from $5 trillion to $19 trillion[16]. If even the low end of these projections proves accurate, today's investments could look prescient rather than profligate.
"You can't look at AI as a bubble, though some of these things may be in a bubble. In total, it'll probably pay off," JPMorgan CEO Jamie Dimon said at a recent conference. His argument: transformative technologies always create bubbles, but the infrastructure they leave behind changes the world[17].
The competitive dynamics also differ fundamentally from previous bubbles. The telecom boom saw dozens of companies racing to build identical infrastructure, each believing they'd capture market share from incumbents. Today's AI race is more akin to a nuclear arms race among superpowers. Microsoft, Google, Amazon, and Meta aren't just investing for profit—they're investing for survival. As one executive put it: "The cost of being wrong about AI is that we exist. The cost of overinvesting is some wasted capital. Which would you choose?"
The Smart Money vs The Dead Money
As the dust settles, patterns are emerging about who will likely triumph and who will become cautionary tales. The infrastructure providers - Amazon, Google, and Microsoft - appear best positioned, controlling both the cloud infrastructure and the customer relationships that AI companies need. They're selling picks and shovels in a gold rush where they also own the mines.
Nvidia, despite concerns about its lofty valuation, has built what may be the strongest moat in technology. With 80% to 95% market share in AI chips and software that locks in customers, the company has become almost synonymous with AI itself. Emerging players like Broadcom are carving out crucial niches in specialized processors and networking equipment, potentially worth more than Amazon and Palantir combined by 2030[18].
The losers are already becoming visible. Legacy technology companies that moved too slowly - Adobe and Intel come to mind - risk having their business models devoured by AI-native competitors. Pure-play AI startups with no path to profitability face extinction once the funding music stops. Companies like SoundHound AI, despite their technical prowess, trade at price-to-sales ratios of 56 as of October 2025, with no profits in sight.
But perhaps the biggest losers will be the last investors in, those buying at peak valuations with nowhere to go but down. As one veteran venture capitalist observed, "In every bubble, the earliest investors make fortunes, the middle investors break even, and the last investors hold the bag."
Dark Fiber's Legacy – A Silver Lining
Here's the twist that bubble skeptics often miss: even catastrophic bubbles can leave behind transformative infrastructure. Those miles of dark fiber that symbolized the telecom crash eventually became the backbone of our modern internet. YouTube, Netflix, cloud computing – none would exist without the massive overcapacity created by the telecom bubble. Companies that didn't exist during the boom harvested the infrastructure left behind by its casualties.
The same pattern may emerge from AI's infrastructure binge. Today's "overbuilt" data centers and GPU clusters could enable applications we can't yet imagine. Just as no one in 2001 could have predicted that unused fiber would one day stream 4K videos to billions of smartphones, we may not yet understand what cheap, abundant AI computation will enable.
"The infrastructure outlives the bubbles," notes technology historian Andrew Odlyzko, who studied both railway manias and telecom booms. "The speculators lose their shirts, but society gets the rails, the fiber, the foundations of the next economy"[19].
The Moment of Truth
As 2025 draws to a close, the AI boom stands at an inflection point. The optimists see a technology revolution in its early innings, with today's investments laying groundwork for decades of innovation. They point to real revenue, actual products, and genuine productivity gains. AI isn't just promises - it's already changing how millions work, create, and think.
The pessimists see classic bubble dynamics: circular financing, impossible valuations, and economics that don't pencil out. They note that 95% of corporate AI pilots fail to deliver meaningful results[20], that the biggest players are losing billions, and that the concentration of value in a handful of stocks has made the entire market vulnerable to a correction.
The truth, as it often does, likely lies somewhere in between. We are almost certainly witnessing some degree of overinvestment, some amount of Seinfeld-esque irrational exuberance. Companies will fail, fortunes will evaporate, and historians will shake their heads at our collective delusion. That's the nature of transformative technologies - they create bubbles as surely as summer creates thunderstorms.
But this bubble, if that's what we're living through, feels different from its predecessors. The demand is real. The revenue is real. The productivity gains, while early, are measurable. Unlike the telecom boom's bet on future demand that never materialized, AI is struggling to meet demand that already exists.
History’s Lesson
Standing here in the final months of 2025, watching trillions of dollars flow into artificial intelligence, we're reminded that the most important technological revolutions often arrive wearing the disguise of speculative bubbles. The railroad bubble of the 1840s left behind the transportation network that built modern America. The electricity bubble of the 1920s powered the twentieth century. The telecom bubble gave us the internet age.
Hemant Taneja, CEO of venture firm General Catalyst, perhaps put it best: "Of course there's a bubble. Bubbles are good. Bubbles align capital and talent in a new trend, and that creates some carnage, but it also creates enduring, new businesses that change the world"[21].
The question isn't whether AI represents a bubble - by most measures, it does. The question is whether this bubble, like those before it, will leave behind infrastructure and innovations that justify the madness. Whether OpenAI's $500 billion valuation proves prophetic or preposterous, whether today's investors are visionaries or fools, one thing seems certain: the world after this bubble will be fundamentally different from the world before it.
For those of us living through it, watching fortunes made and lost at unprecedented scale, the challenge isn't predicting when the music stops. It's positioning ourselves to thrive in whatever world emerges from the wreckage and wonder of humanity's latest leap into the unknown.
The ghost of bubbles past may haunt Silicon Valley's corridors, but it's accompanied by the spirit of innovation’s future. And sometimes, just sometimes, that's enough to justify the madness.
"Of course it's a bubble. But bubbles fund the future. The trick is being on the right side of the pop."
#ONWARD
Footnotes:
[1] https://aibusinessweekly.net/p/ai-startups-capture-51-percent-venture-funding-2025
[2] https://fortune.com/2025/08/13/ai-creating-billionaire-boom-record-pace-now-498-ai-unicorns-worth-2-7-trillion/
[3] https://www.cnbc.com/2025/10/02/openai-share-sale-500-billion-valuation.html
[4] https://en.wikipedia.org/wiki/OpenAI
[5] https://taptwicedigital.com/stats/openai
[6] https://www.saastr.com/openai-crosses-12-billion-arr-the-3-year-sprint-that-redefined-whats-possible-in-scaling-software/
[7] https://en.wikipedia.org/wiki/Dot-com_bubble
[8] https://www.thebubblebubble.com/dot-com-bubble/
[9] https://www.fabricatedknowledge.com/p/lessons-from-history-the-rise-and
[10] https://www.reuters.com/business/autos-transportation/companies-pouring-billions-advance-ai-infrastructure-2025-10-06/
[11] https://www.evli.com/en/articles/ai-race-winners-and-losers
[12] https://www.cnbc.com/2025/10/21/are-we-in-an-ai-bubble.html
[13] https://www.weforum.org/stories/2025/10/artificial-intelligence-bubble-dot-com-tulip-mania/
[14] https://insights.som.yale.edu/insights/this-is-how-the-ai-bubble-bursts
[15] https://sacra.com/c/openai/
[16] https://fortune.com/2025/10/16/ai-bubble-will-unlock-an-8-trillion-opportunity-goldman-sachs/
[17] https://fortune.com/article/jpmorgan-ceo-jamie-dimon-artificial-intelligence-trump-tariffs-fortune-500-titans-disruptors-industry/
[18] https://www.fool.com/investing/2025/10/22/prediction-1-artificial-intelligence-ai-stock-will/
[19] https://www-users.cse.umn.edu/~odlyzko/talks/gmu2008.pdf
[20] https://fortune.com/2025/08/21/an-mit-report-that-95-of-ai-pilots-fail-spooked-investors-but-the-reason-why-those-pilots-failed-is-what-should-make-the-c-suite-anxious/
[21] https://finance.yahoo.com/news/don-t-fear-ai-bubble-105043701.html
