The four words that precede every crash: “This time is different.”

Except this time, the person warning about a bubble is Sam Altman, the CEO most responsible for creating it. When OpenAI’s chief executive warned last week that investors were “overexcited” about AI, markets reacted immediately. Nvidia fell 3.5%, Palantir dropped nearly 10%, and the selloff spread globally.

The warning came amid a cascade of seemingly supporting data. That same week, MIT researchers published findings that 95% of companies investing in generative AI were seeing no measurable returns. Apollo Global Management’s chief economist warned that current valuations exceeded even dot-com bubble peaks. And Federal Reserve data showed AI investment consuming more than half of America’s total capital expenditure.

The numbers tell the story. Anthropic raised $450 million at a $4.1 billion valuation despite negligible revenue. Character.AI hit $1 billion in valuation with 1.7 million monthly users—roughly $588 per user. Inflection AI raised $1.3 billion before essentially acqui-hiring its team to Microsoft, leaving investors with an empty shell.

Ray Dalio, founder of Bridgewater Associates, told the Financial Times the current environment mirrors 1998-1999, warning that while AI will certainly transform the economy, investors are “confusing that with the investments being successful.”

Sound familiar? It should. This exact sequence—revolutionary technology, abundant capital, speculative frenzy, then sudden reality checks—has played out with remarkable consistency for over 180 years. Railway Mania in the 1840s. Radio stocks in the 1920s. Dot-com fever in the 1990s. Each time, the technology was real. Each time, the speculation was unsustainable. Each time, the overbuilding became tomorrow’s foundation.

The Original Tech Bubble: When Britain Went Mad for Railways

To understand what’s happening with AI today, we need to travel back to 1840s Britain, where the world’s first true technology bubble was taking shape around the railway.

The Railway Mania of the 1840s makes today’s AI frenzy look restrained. Between 1843 and 1846, Parliament approved 263 Acts for new railway companies proposing 9,500 miles of track, nearly matching today’s entire UK railway network. The speculation democratized investing in a way never seen before. Clerks, shopkeepers, and domestic servants, people who had never owned stocks before, mortgaged their homes and borrowed money to buy railway shares. The mania pulled in everyone from farmers to factory workers, all convinced they were investing in the future.

The mania started with genuine success. The Liverpool-Manchester Railway had proven the concept, reducing the London-Glasgow journey from days to 24 hours. This wasn’t just faster transportation but a compression of time and space that seemed magical to people who had never traveled faster than a horse could run.

Economic conditions enabled speculation. Low interest rates pushed Britain’s middle class toward railway stocks. Shares could be purchased with just a 10% deposit, with the balance paid later through capital calls. This leverage meant ordinary people could control far more stock than they could afford—imagine buying $10,000 of stock with $1,000 down. By 1846, railway companies comprised 71% of total stock market value, up from 23% eight years earlier.

Then reality intruded. In October 1845, the Bank of England raised rates. Suddenly, leveraged investors faced capital calls they couldn’t meet. Families who thought they were making modest investments discovered they owed thousands of pounds they didn’t have. By 1850, railway shares had lost 85% of their peak values. Over 200 companies went bankrupt. The Commercial Crisis of 1847 became one of Britain’s worst financial disasters.

But while investors lost fortunes, Britain gained invaluable infrastructure. All that speculative investment had built a railway network that became the backbone of the Industrial Revolution. The overbuilt, redundant tracks that seemed like waste in 1847 enabled Britain’s industrial dominance for the next century.

The Dot-Com Bubble: When the Internet Broke Economics

Fast-forward 150 years, and the pattern repeated with eerie precision. The late 1990s internet bubble demonstrated how compelling narratives can completely detach valuations from any semblance of business reality.

The numbers were staggering. The NASDAQ rose 800% from 1995 to its March 2000 peak of 5,048. By traditional measures, the insanity was obvious: the index’s price-to-earnings (P/E) ratio reached 200. In plain English, investors were paying $200 for every $1 of actual profit these companies generated. For context, a healthy market typically sees P/E ratios of 15-20. Even Japan’s infamous 1989 bubble peaked at 80. But dot-com investors had convinced themselves that earnings didn’t matter anymore. What mattered was “eyeballs” and future potential. Many of these companies had no earnings at all, making their P/E ratios technically infinite. This mathematical impossibility should have been a warning sign. Investors had completely abandoned fundamental analysis in favor of pure speculation about future potential, convinced that traditional metrics like revenue, profit margins, and cash flow were obsolete in the ‘new economy.

The Wall Street Journal, that bastion of financial sobriety, suggested investors “re-think” the “quaint idea” that companies should be profitable.

As with railways, the underlying technology was genuinely revolutionary. The internet was indeed going to change everything—just not as quickly or as profitably as investors assumed. And just like the railway boom, perfect economic conditions enabled the speculation. The Federal Reserve had cut rates after the 1998 Long-Term Capital Management collapse, flooding the system with cheap money. The 1997 Taxpayer Relief Act lowered capital gains taxes, making speculation more attractive.

But it was the psychology that made the dot-com bubble truly extraordinary. Any company that added “.com” to its name could attract investment. VA Linux Systems gained 698% on its IPO day. Pets.com, which sold pet food online at a loss, reached a $300 million market capitalization. TheGlobe.com went public with no revenue and saw its stock price rise 606% in a single day. The democratization of investing through online brokers meant that by spring 1999, one in twelve Americans claimed to be starting a business. Individual investors poured $260 billion into equity funds in 2000, with margin debt peaking at $300 billion. Day trading became a cultural phenomenon. CNBC became appointment television.

The philosophy of the era was captured perfectly by a Kleiner Perkins partner who said, “In the old days, you needed a business model. Now you need a business concept.” Revenue was yesterday’s metric; “eyeballs” and “mindshare” were what mattered. Traditional metrics were dismissed as relics of the industrial age.

The crash, when it came, was swift and merciless. From March 2000 to October 2002, the NASDAQ fell 78%. Total market capitalization losses reached $5 trillion, roughly half of US GDP at the time. But these staggering numbers don’t capture the human cost to everyday people. The dot-com crash eliminated 200,000 jobs in Silicon Valley alone, while millions of ordinary investors watched their retirement accounts and college funds evaporate. The same middle-class Americans who had been told they were foolish not to participate in the ‘new economy’ now faced financial ruin. Teachers’ pension funds were halved. Family savings meant for homes and education vanished. Web designers who commanded six figures found themselves competing for $15/hour contracts. Communities built around tech hubs hollowed out as workers fled to cheaper cities. Of the 7,000 to 10,000 internet companies launched in the late 1990s, only 48% survived past 2004.

Creative destruction is brutal math. The capital? Gone. Completely vaporized. But infrastructure isn’t stock certificates. Those fiber optic cables didn’t vanish when Pets.com did. The data centers kept humming after Webvan went dark. All that ‘wasted’ investment had already transformed into something physical. The pipes, servers, and networks that would become the foundation for Google, Facebook, Amazon Web Services, and the digital transformation that actually did change everything. The bubble’s victims unknowingly funded the future. They just paid a decade too early.

No Country Is Immune

The bubble pattern isn’t uniquely Anglo-American. Japan’s 1980s asset price bubble saw the Nikkei rise 300% in five years before losing 60% of its value. The grounds of the Imperial Palace in Tokyo were theoretically worth more than all the real estate in California. When it crashed, Japan entered a “lost decade” of economic stagnation.

China has perfected the art of managed bubbles. Its 2007 stock market bubble saw the Shanghai Composite rise 480% in two years before crashing 72%. Its 2015 bubble was even more dramatic: a 150% rise in six months followed by a 43% crash in just two months. Each time, state intervention prevented total collapse but couldn’t prevent massive wealth destruction.

Today’s AI bubble is distinctly global. China races to match US capabilities, pouring state resources into domestic champions. The EU struggles to regulate while trying not to fall behind. Saudi Arabia commits $40 billion to AI investments through its sovereign wealth fund. This synchronized global speculation might amplify both the boom and any eventual bust.

The Pattern We Can’t Escape

Step back from the specifics, and a pattern emerges across centuries:

1. Genuine breakthrough creates legitimate excitement. Railways, electricity, internet, AI—each represented real technological leaps, not just incremental improvements.

2. Early success stories make believers out of skeptics. The Liverpool-Manchester Railway shrank days to hours. Amazon survived when 90% of dot-coms died. ChatGPT hit 100 million users in two months. These proof points silence doubters just long enough for speculation to take hold.

3. Capital markets provide the fuel. Low interest rates, new investment vehicles, democratized access—different mechanisms, same result: too much money chasing the future.

4. Social proof overwhelms skepticism. When your neighbor gets rich on railway stocks or your coworker quits to join a startup, FOMO beats rational analysis every time.

5. Reality reasserts itself. Capital calls come due. Revenue fails to materialize. The future arrives, just slower and differently than promised.

6. Infrastructure remains after speculators flee. Britain’s rail network, America’s fiber optic cables, today’s GPU clusters—overbuilding becomes tomorrow’s competitive advantage.

This pattern held true across different continents, centuries, and technologies. The actors change but the script remains remarkably consistent. Which brings us to today’s AI boom and an unprecedented twist: what happens when everyone can see the pattern while living it?

Why Now: The Bubble That Knows It’s a Bubble

Here’s what’s never happened before: everyone knows the script.

CEOs are calling their own sectors overheated. Researchers are publishing failure rates in real time. Even casual investors track AI valuations. This isn’t insider knowledge anymore, it’s mainstream discourse.

Compare that to history. Railway Mania? Information traveled at horse speed. The dot-com crash? CNBC existed, but Twitter didn’t. Most people learned about the bubble after their 401k got vaporized.

Now we have instant global commentary on every funding round, every overhyped demo, every cautionary study. MIT publishes data showing failure rates, and it’s trending on LinkedIn within hours. Bloomberg runs “Is This a Bubble?” segments weekly. Retail investors share articles about historical bubble patterns.

The modern version of “adding .com to your name” is even more brazen. Companies append “AI” to their descriptions and watch valuations soar. Buzzfeed stock jumped 300% after announcing AI-generated content. C3.ai trades at 15 times revenue despite losing money on every dollar of sales. Even established companies play this game: consulting firms that once sold “digital transformation” now sell “AI transformation” at premium prices.

Nvidia’s numbers capture the mania perfectly. The company trades at a P/E ratio floating between 58 and 72, more than triple the S&P 500 average. Its market cap exceeds $4 trillion based largely on selling shovels to AI gold miners. Microsoft commits $13 billion to OpenAI. Amazon pledges $4 billion to Anthropic. Google races to match with Gemini investments. The capital expenditure on AI infrastructure already exceeds the entire dot-com bubble when adjusted for inflation. To put this in perspective, AI infrastructure spending is approaching hundreds of billions annually, a level of concentrated technological investment not seen since the space race.

This creates a paradox: the most transparent bubble in history might also be the most inevitable. Because knowing you’re in a bubble doesn’t stop you from participating. Ask anyone who bought GameStop during the meme stock frenzy. They knew it was insane. They did it anyway.

The question isn’t whether awareness prevents stupidity. It’s whether it changes the flavor of stupidity we’re about to witness.

Why AI Might Actually Be Different This Time

Maybe AI will break the mold. The optimists aren’t completely delusional. There are genuine reasons why this technology could defy historical patterns:

Speed of deployment changes everything. Railways took decades to build. Even software requires downloads, integrations, and training. But AI? ChatGPT reached 100 million users in two months. When infrastructure is cloud-based and products are accessible through browsers, adoption curves go vertical in ways physical products never could.

The recursive improvement possibility. Every previous technology improved through human innovation. AI might be the first technology capable of improving itself. If AI can enhance AI development, we could see exponential rather than linear progress. This isn’t science fiction. Current models already help train next-generation models.

Network effects on steroids. More users make social networks more valuable, but more users make AI smarter. Every prompt teaches the system. Every correction improves performance. The company that achieves AI dominance might have an insurmountable advantage because their lead compounds daily.

Why the giants might create their own competition. Here’s the twist: the massive training costs and compute requirements that make today’s AI leaders seem unassailable might actually democratize the technology. Once foundation models exist, smaller companies can build specialized applications without bearing the upfront costs. OpenAI spends billions so a startup can spend thousands. But unlike previous platform monopolies, AI might be uniquely resistant to moats. When your competitor’s model improves, you can often incorporate those advances. When open-source alternatives emerge, they benefit from collective innovation. The result? Instead of winner-take-all dynamics, we might see an explosion of AI-powered businesses that collectively grow the pie faster than any bubble can inflate. The very nature of the technology (reproducible, composable, improvable) could create more value than it destroys.

These arguments deserve serious consideration. Maybe this time really is different. But that’s exactly what investors thought about railways (“annihilating distance!”), radio (“wireless changes everything!”), and the internet (“bits not atoms!”). Revolutionary technologies can be both transformative and overvalued. The question isn’t whether AI will change the world. It will. The question isn’t whether AI will change the world. It will. The question is whether current valuations reflect realistic timelines and profit potential, or whether we’re once again paying tomorrow’s prices with today’s money.

And if I’m wrong? If AI really does break every historical pattern? Well, then we’re about to witness the greatest creation of wealth in human history. That’s a bet many find worth taking.

When Will the Music Stop?

History offers clues but no certainties about timing. Railway Mania lasted roughly four years from acceleration to crash (1843-1847). The dot-com bubble’s acute phase ran about five years (1995-2000). The pattern suggests major technology bubbles take 4-6 years from mainstream awareness to collapse.

For AI, if we date mainstream awareness from ChatGPT’s November 2022 launch, we’re now almost three years into the cycle. History suggests we’re approaching the middle innings. The typical progression:

• Year 1-2: Wonder and experimentation (2022-2024)
• Year 2-3: Massive capital deployment and infrastructure building (we are here)
• Year 3-4: Market saturation and disappointing returns become evident
• Year 4-5: The music stops

But several factors could accelerate or delay this timeline. Rising interest rates historically trigger bubble collapses, and we’re already seeing rate increases. Conversely, government support for AI as a national security priority could extend the bubble beyond historical norms.

The most reliable indicator? Watch for the moment when AI companies start acquiring each other with stock instead of cash. That’s historically been the last stage before collapse.

Profiting from the Wreckage

While everyone’s debating whether we’re in a bubble, some investors are asking different questions: What’s getting overbuilt? What will be cheap after the crash? Who’s actually solving problems versus who’s just riding the hype wave?

If history is any guide, here’s what the playbook might look like:

First, follow the infrastructure. During Railway Mania, the smartest move wasn’t buying railway stocks but investing in steel, land, and engineering firms. When dot-com imploded, someone had to buy all those fiber optic cables for pennies on the dollar. Today? Data centers, semiconductor fabs, and power infrastructure are getting massive investment. That hardware doesn’t disappear when valuations crash.

Second, look for real revenue. If the MIT study is right, 95% of AI companies are burning cash on “customer discovery” and “market education.” But somewhere in that mix, 5% are actually charging money for solutions people need today. Companies with boring enterprise contracts might outlast those creating viral demos.

Third, consider betting on necessity, not novelty. Amazon wasn’t sexy during the dot-com crash. It was selling books to people who needed books. Adobe wasn’t exciting during the financial crisis. It was selling tools that designers and businesses couldn’t work without. The AI companies that survive might be those automating tasks people hate doing, not those promising artificial general intelligence.

Fourth, prepare for potential fire sales. Every bubble ends with good companies getting dumped alongside bad ones. In 2002, you could buy Google’s pre-IPO shares for a fraction of their eventual worth. In 2008, Netflix traded at $3 per share while being dismissed as a DVD-by-mail dinosaur. By 2010, those $3 shares were worth $30. By 2020, after stock splits, that $3 investment was worth over $300. The next cycle might serve up similar opportunities for those with cash and patience.

The bubble creates the exact conditions that make post-bubble investing potentially profitable. Massive overinvestment drives down infrastructure costs. Market crashes eliminate weak competitors. Talent becomes available as overfunded startups implode.

History’s lesson isn’t “avoid bubbles.” It’s “position for what comes after.”

Different Tools, Same Choices

As I write this, my LinkedIn feed is flooded with AI funding announcements. Every tech conference features panels on “Why AI Changes Everything.” The majority of pitch decks I see now include an AI angle, whether it fits or not. The pattern is impossible to miss.

If you’ve read this far, these claims probably sound familiar. Not because you’ve heard them about AI before, but because they echo every transformative technology for the past 200 years.

But here’s what’s genuinely different this time: we’re not blind to the patterns anymore. Sam Altman’s bubble warning isn’t just refreshing honesty. It’s evidence that we might be learning from history instead of repeating it exactly.

We have tools previous generations lacked. Start with regulation: Sarbanes-Oxley emerged from dot-com’s ashes, requiring real financial disclosure. The Volcker Rule limits banks from making speculative bets with depositor money. Stress testing means major financial institutions must prove they can survive market shocks.

The investor base has evolved too. Where Railway Mania saw shopkeepers betting their life savings and dot-com drew day traders chasing quick profits, today’s AI investments flow primarily through venture funds and institutional investors. Yes, retail can buy Nvidia, but they can’t access pre-IPO rounds where the real speculation happens. This concentration among professional investors won’t prevent a bubble, but it might prevent the kind of widespread financial devastation that followed previous crashes.

Risk management has transformed. Portfolio theory, derivatives for hedging, and real-time data mean sophisticated investors can protect themselves in ways previous generations couldn’t imagine. The Federal Reserve now has tools like quantitative easing that didn’t exist during past crises. Whether they’ll use them wisely is another question, but the toolkit exists.

Most importantly, we have communication networks that spread both hype and skepticism at equal speed. MIT’s study showing 95% failure rates went viral the same week it published. Every breathless AI announcement triggers immediate fact-checking. This doesn’t stop speculation, but it arms investors with information their predecessors desperately lacked.

The speculation, excitement, and even some degree of overinvestment are probably necessary parts of how society handles revolutionary change. But the catastrophic boom-bust cycles aren’t inevitable features of progress. They’re bugs in our system that we can potentially fix.

The AI revolution is real, transformative, and probably unstoppable. Whether it unfolds through sustainable growth or boom-bust cycles depends largely on the choices we make in the next few years. The early signs (including voices like Altman’s warning about overexcitement) suggest we might actually be learning from history.

The AI bubble’s human impact could be fundamentally different. Previous bubbles destroyed jobs when they burst. AI might destroy jobs while it’s still inflating. If AI actually delivers on its automation promises, we could see the first bubble that eliminates more employment during its rise than its fall.

This creates an unprecedented social risk: a technology bubble that succeeds in its goals might cause more disruption than one that fails. The Railway Mania gave Britain train networks and industrial jobs. The dot-com bubble gave us e-commerce and digital careers. The AI bubble might give us unprecedented productivity and fewer jobs. That’s a social equation we haven’t solved.

What We Know vs. What We’ll Do

After 180 years of spectacular booms and devastating busts, the patterns are clear. The question is whether we can act on them.

History suggests most of us can’t. We’ll participate in the mania despite knowing better, just as railway clerks mortgaged their homes and day traders quit their jobs. That’s human nature. But perhaps knowing the script at least lets us play our parts more consciously.

The truth is, everyone already knows we’re in a bubble. You know AI valuations are insane. You know most of these companies will fail. You know the crash is coming. The question isn’t what to do—it’s whether you’ll actually do it when the time comes.

Because when Nvidia drops 50%, you won’t think “opportunity.” You’ll think “it’s going to zero.” When AI becomes toxic and unfashionable, you won’t remember that the technology is real. You’ll question everything. That’s the psychological trap: understanding bubbles intellectually is easy. Maintaining perspective inside them is nearly impossible.

If you must participate (and let’s be honest, the FOMO is real), history offers some guardrails. The infrastructure builders tend to survive even if their valuations don’t. The companies solving mundane problems outlast those promising revolution. And the best opportunities emerge not during the mania, but in the wreckage that follows.

Keep powder dry. When everyone declares AI “dead,” that’s your moment. Not today, when your barista is giving stock tips. The future always arrives. It just takes longer and looks different than anyone expects.

The most radical act might be patience. Let others fund the infrastructure. Let them discover which use cases actually matter. Let them burn through capital learning hard lessons. Then, when the dust settles and the tourists have fled, see what’s left.

Because something always is. Britain got its railways. We got the internet. And we’ll get AI. The only question is whether you’ll be among those who paid for it or those who profited from it.

Bubbles are wealth transfer mechanisms. Money flows from the impatient to the patient, from the leveraged to the liquid, from the emotional to the analytical. The only real edge is temperament. Can you remain skeptical during euphoria and optimistic during despair? History says probably not. But at least now you know the script you’re following.

The real revolution isn’t in the technology we’re building. It’s in whether we’ve finally learned how to build it without destroying ourselves in the process.

The four words that precede every crash: ‘This time is different.’ But the four words that might prevent one? ‘We’ve seen this before.’

I work at ServiceNow, but these views are entirely my own and don’t represent my employer’s position on AI, bubbles, or anything else. This is definitely not investment advice, just one person’s attempt to make sense of history rhyming once again.