The AI Bubble Is Coming: We've Seen This Movie Before, We Know How It Ends

A few days ago, I was listening to the first episode of Andrew Neil's new podcast, interviewing Andrew Ross Sorkin, The New York Times financial journalist and author of 1929. What stopped me wasn't anything entirely new. It was how precisely Sorkin placed today's AI bubble inside a pattern going back nearly two centuries, a pattern we convince ourselves is different every single time it appears. It never is. This piece works through what that pattern actually says about jobs, private credit, and the $750 billion currently propping up American GDP.

Is AI a Bubble? What History Says

Yes, and the history is unambiguous about it. Andrew Neil's interview with Andrew Ross Sorkin placed today's AI investment cycle inside a pattern that's repeated at least four times before: the Railway Mania of the 1840s, Radio Corporation of America in the 1920s, and the dotcom boom of the late 1990s. In every case, the underlying technology was genuinely transformative. In every case, the investment bubble around it ran far ahead of any credible near-term revenue case. The AI bubble vs dotcom bubble comparison isn't a stretch. It's the same mechanism, wearing different clothes.

Why the AI Boom Looks Like the Industrial Revolution, And Why That's the Problem

AI is not hype. It's the most significant technological transformation since the Industrial Revolution, and the productivity gains are real. That's exactly why the comparison should worry us rather than reassure us. The Industrial Revolution created the most brutal economic dislocations of the 19th century. Whole categories of work were destroyed. Communities built around specific trades were hollowed out. New work eventually emerged, but the transition was neither smooth nor fast, and the people who bore the cost of it were rarely the people who gained from it.

That asymmetry is worth dwelling on, because it's the part most AI commentary skips past on the way to talking about productivity gains. The Industrial Revolution didn't just displace individual workers. It displaced entire towns built around a single trade, weaving, milling, smithing, and those towns rarely recovered on the same timeline the national economy did. The people who owned capital during that transition captured the upside of mechanisation almost immediately. The people who supplied labour captured the benefits decades later, and largely because of sustained political and organised pressure, not because prosperity trickled down on its own.

We've already asked whether the deeper cost runs further than jobs alone, we've asked whether AI is quietly eroding original thought itself, and the jobs question sits right alongside that concern rather than separate from it.

The Jobs Case: What Dario Amodei and Andrew Neil Are Both Warning About

The Amodei Warning

Dario Amodei, CEO of Anthropic, has spoken openly about the possibility of 20 to 30 per cent unemployment in America as AI scales. Dario Amodei's warning to Axios wasn't dismissed by Sorkin on the podcast. What he said instead was more unsettling: if AI succeeds on its own terms, if the productivity gains are real and today's valuations are justified, then by definition those companies will need to eliminate enormous numbers of jobs. That's how the productivity gains get realised. That's how the investment case stacks up. The bull case for AI and the AI bubble jobs impact aren't in tension. They're the same argument.

Which sectors carry the most immediate exposure is worth being specific about, rather than treating displacement as an abstract, economy-wide fog. Customer service, first-line legal review, entry-level coding, and large parts of financial analysis are already showing measurable task automation, not job automation yet, but the distinction between the two tends to collapse faster than employers initially expect once a task-level tool proves reliable enough to remove the human step entirely. That's usually the point where a role gets restructured rather than simply assisted.

The Neil Precedent: 80% of Jobs We Couldn't Yet Name

Andrew Neil made a point in the interview I found genuinely arresting, drawn from his own experience rather than theory. As editor of The Sunday Times in the 1980s, he recalled 16 to 20 pages of job vacancies advertised in print every week, and noticed roughly 80 per cent of those job titles hadn't existed twenty years earlier. "We just didn't know," he said. Economies with strong growth do generate jobs nobody can currently name. But the gap between jobs destroyed and jobs created is rarely short or painless, and it falls hardest on people with the fewest options to retrain quickly or the capital to ride out disruption. History is unambiguous here too: capital captures the gains first, and labour catches up later, usually only after considerable political pressure.

The Private Credit Blind Spot Nobody Is Watching

Why This Is Structural, Not Incidental

Post-2008 regulation made banks safer and inadvertently pushed lending into the shadows. Private equity credit funds now do the work banks once did, without prospectuses, mandatory filings, or regulatory oversight. As Sorkin put it on the podcast: "Banks hardly lend the money anymore. Now it's basically private equity firms creating these funds and lending that money out. We have no transparency, no real look into what is going on inside these firms. And some of these guys are already running into trouble."

The private credit AI risk connection makes this structural rather than incidental. Several of the largest private credit funds have leaned heavily into software and AI businesses, some partnering directly with hyperscalers including Meta. If AI valuations correct, the damage won't stay contained to equity markets. It will travel through private credit structures that are, by design, invisible to the regulators who'd need to see it coming.

The Opacity Problem

The opacity itself is the risk, arguably more than the leverage. A regulator watching bank balance sheets can at least see concentration building in real time and intervene before it metastasises. A private credit fund lending against AI infrastructure or software revenue has no equivalent public reporting requirement, which means the first sign of trouble for most observers will be a fund freezing redemptions or a high-profile default, not a gradual, visible build-up anyone could have flagged months earlier.

The 1920s Bankers Were the Tech Bros of Their Day

Andrew Neil asked Sorkin directly whether today's AI founders and CEOs are the modern equivalents of the colourful characters who drove the 1920s boom. Sorkin's answer was unambiguous: "Absolutely. As I was writing the book, I was thinking about each of my characters back in 1929 as their modern-day equivalents."

Consider the parallels. "Sunshine Charlie" Mitchell, who ran what became Citibank, was the great democratiser of his era, opening equity markets to ordinary Americans and telling the public that prosperity was theirs for the taking. John Raskob, GM's CFO, founded the General Motors Acceptance Corporation in 1919, letting GM dealers offer instalment credit directly to everyday consumers. He published an article in 1929 literally titled "Everybody Ought to Be Rich," then went on to build the Empire State Building. In Sorkin's words, Raskob was the Elon Musk of his time. These weren't stupid or venal men. They were visionaries who'd been right for so long they'd stopped being able to imagine being wrong.

A hundred years later, in 2026, JPMorgan Chase CEO Jamie Dimon championed the "democratisation of finance" while hosting prominent investors at the SpaceX IPO roadshow. The language changes. The psychology doesn't. What's genuinely notable is how little the actual rhetorical toolkit has changed. Both eras lean on the same three words: access, democratisation, and inevitability. Access frames speculative participation as empowerment rather than risk. Democratisation frames exclusivity as unfairness being corrected. Inevitability frames scepticism as simply being behind the times. None of the three claims are inherently false. All three happen to be exactly what a speculative promoter would say regardless of whether the underlying asset is sound.

Railroads, Radios, Dotcoms, AI: The Anatomy of Every Bubble

The AI boom is at least the fourth time transformative technology has generated a speculative cycle vastly outrunning the revenue that could justify it.

Era	Technology	Peak Behaviour	Investor Outcome
1840s	Railway Mania	Investment from all walks of life, disconnected from commercial reality	Bust of 1847 was brutal; railways went on to build the modern industrial economy
1920s	Radio Corporation of America	Stock rose to extraordinary heights on a genuine communications revolution	Fell from $530 to $3 in the crash that followed
Late 1990s	Dotcom boom	Real transformation, trillions in notional value created	Trillions destroyed 2000-2002; Amazon and Google survived on the bones of the crash
2026	AI	$750 billion in expected US investment	Unproven at the scale current valuations require

What's striking laid out this way isn't any single row. It's how consistent the pattern is across nearly two centuries and four entirely different technologies, railways, radio, the internet, and now AI. The technology won in every prior case. The investors who financed the bubble around it largely did not. The railways bankrupted a huge share of their original financiers before becoming the backbone of industrial logistics for a century. RCA's shareholders lost effectively everything before broadcast radio and television became permanent fixtures of daily life. The dotcom crash wiped out trillions before the survivors became the most valuable companies on earth. In every case, being early and being right about the technology were not the same as being right about the price you paid for it.

What $750 Billion in AI Spend Is Really Propping Up

$750 billion is expected to be invested in AI in America in 2026. Harvard economist Jason Furman's GDP calculationfound that if you strip that spend out of US GDP, American growth falls to approximately 0.1 per cent, effectively flat. The entire growth story of the world's largest economy is currently being sustained by a single, concentrated bet on a technology whose revenue model, at the scale required to justify current valuations, remains unproven.

That's not inherently disqualifying on its own. The same was true of the railways in 1845 and the internet in 1998. What it tells us is that we're standing in exactly the same structural position as every major bubble that preceded this one. As Jeff Bezos told Sorkin just weeks before the podcast: "A bubble may well burst, but it is a necessary component of innovation."

There's a second, less discussed implication in Furman's number too. If AI capital expenditure is functionally the entire US growth story right now, then any slowdown in that specific spending category, whether from a genuine correction, a credit event in private markets, or simply hyperscalers pausing to digest existing infrastructure, doesn't just affect tech valuations. It shows up directly in headline GDP, which means the AI bubble isn't purely a financial markets story anymore. It's a macroeconomic one, with consequences for interest rate policy and employment data well beyond the sector itself.

"This Time Is Different": The Four Most Dangerous Words in Finance

What connects the railways, RCA, the dotcom boom and AI is the same phrase, deployed in every era by intelligent people who should have known better: this time is different. The technology is different, so the old rules don't apply. The scale is different, so the old valuations don't matter. The opportunity is different, so old scepticism is misplaced.

The technology is always genuinely different, and the scale is often different too. But the human psychology driving the investment cycle never changes. The mechanism repeats with remarkable consistency across two centuries of financial history: scepticism eroded by sustained success, leverage building in places regulators can't see, ordinary savers drawn in last, with the least information and the least protection.

It's worth noting what "this time is different" actually accomplishes psychologically, because it isn't simple denial. It's a genuinely persuasive argument in the moment, made by people who are usually right about the technology itself and wrong only about what that means for the price. Railway engineers weren't wrong that railways would reshape the economy. RCA wasn't wrong that radio would reshape communication. The internet bulls weren't wrong that the internet would reshape commerce. Being right about the technology and being right about the valuation are two entirely separate claims, and history shows they get conflated almost every single time.

What This Means for Business Leaders and Investors Right Now

AI will change the world, probably more profoundly than any technology since the original Industrial Revolution. It will also destroy jobs at a scale and speed we're not yet prepared for, and generate speculative losses that fall hardest on those who arrive last and know least. Both things are simultaneously true, and pretending otherwise doesn't make either one less real.

For anyone building strategy around AI right now, whether in automotive finance or anywhere else, the practical takeaway is the same one we've made in a different context before: an industry already learning that speed can outrun scale tends to discover the gap the hard way, after the capital's already committed rather than before. And on the human side of this transition, trust, not just technology, is what will determine who wins this transition, because the firms and leaders who survive the correction, whenever it comes, will be the ones who built credibility with customers and regulators before they needed it, not after.

In the long run, the technology always wins. More often than not, the investors in the bubble do not.

Talk to Madox Square if you want to think through what this cycle means for your own exposure, whether that's capital allocation, private credit exposure, or workforce planning.

Frequently Asked Questions

1.Is the AI boom a bubble like the dotcom crash? 

The pattern matches closely. Real, transformative technology combined with valuations running far ahead of proven revenue models is exactly what preceded the Railway Mania, RCA's 1920s collapse, and the dotcom crash.

2.What did Dario Amodei say about AI and unemployment? 

Amodei, CEO of Anthropic, has said AI scaling could plausibly produce 20 to 30 per cent unemployment in America, and has framed this as a direct consequence of AI succeeding on its own commercial terms, not a side effect of it failing.

3.How exposed is private credit to an AI correction? 

Significantly, and largely invisibly. Private credit funds now perform much of the lending banks used to do, without the same regulatory oversight, and several of the largest have leaned heavily into software and AI businesses.

4.What is Jason Furman's calculation about AI and US GDP? 

Furman found that stripping AI-related investment out of 2026 US GDP growth leaves the figure at roughly 0.1 per cent, meaning America's entire growth story currently rests on a single concentrated technology bet.

5.Why do people keep saying "this time is different" before every bubble bursts? 

Because sustained success erodes scepticism faster than it builds caution, a pattern that has repeated across every major speculative technology cycle for nearly two centuries, from railways to radio to the dotcom era.