Forget 2008 Leverage: AI’s Crash Will Be About Overcapacity
Nobody rang a bell in 2007. The factory floors kept humming, the mortgage brokers kept closing deals, and the air conditioning plants kept shipping units right up until the moment they didn’t. We tend to imagine financial collapses as sudden events. They’re not. They’re slow leaks that turn into ruptures, and right now we’re listening to the hiss.
The AI investment boom has poured more capital into data centres, chips, and model training than almost any technology cycle in history. Depending on who you ask, that’s either the foundation of the next productivity miracle or the most expensive mistake corporate America has made since the dot-com bubble. Both things can be true at once. Bubbles and breakthroughs are not mutually exclusive. The railroads were real. So was the wreckage.
This piece walks through why the coming AI correction could outpace the 2008 financial crisis in terms of scale, speed, and social cost, even though the mechanics look completely different on paper. We’ll dig into the debt structure, the labour market spiral already in motion, who actually eats the losses, and what history tells us about what comes after the dust settles.
Two Crashes, Two Completely Different Engines
The 2008 crisis was a textbook Hyman Minsky moment. Private debt fueled asset purchases, asset prices inflated, and when they cracked, credit reversed violently. We went from a flood of positive credit to a wave of negative credit almost overnight. That whiplash is what caused the downturn, not the housing itself. Housing was just the trigger.
The AI buildout looks different on the surface. A huge share of the spending isn’t financed through traditional bank lending. Companies like Google, Microsoft, and Amazon are largely self-funding through cash reserves and operating profit. That matters. It means the AI boom isn’t, strictly speaking, a classic debt-driven bubble in the Minsky sense.
But here’s the catch. Self-funded doesn’t mean risk-free. It means the risk has migrated somewhere else, and that somewhere else is private credit, off-balance-sheet financing vehicles, and increasingly, retirement and insurance portfolios. According to Oliver Wyman’s analysis, the dependence on debt financing for large-scale AI projects creates concentrated, lumpy exposure that’s particularly vulnerable to idiosyncratic shocks. One failed hyperscaler bet can ripple through counterparties nobody was watching.
The Schumpeterian Read
There’s a more useful lens here than Minsky, and it comes from Joseph Schumpeter. Schumpeter argued that technological revolutions create boom-bust cycles driven by overinvestment, not leverage. Everyone piles into the new shiny thing, capacity massively overshoots demand, prices crash, and the weak players go bankrupt while one or two survivors absorb the wreckage and dominate the next phase.
That pattern fits AI almost perfectly. We’ve seen this movie before with railroads, with telecom fibre in the late 1990s, and with the dot-com crash itself. Massive infrastructure gets built ahead of actual demand. Most of the builders don’t survive. But the infrastructure doesn’t disappear. It gets bought up cheaply and integrated into the rest of the economy during the slump that follows.
So I think what we’re seeing with AI is a classic Schumpeterian bubble. You’ve had a massive investment, and now, of course, it’s not true that those companies had to borrow money to do it, but because there are so many well-funded players racing each other, the overbuild happens anyway.
Why “No Debt” Doesn’t Mean “No Crisis”
Plenty of commentators point to the lack of obvious leverage and conclude AI poses no systemic risk. That’s a comforting story, and it’s also incomplete. Risk in modern finance doesn’t need a balance sheet to travel. It needs a counterparty.
Consider the layers involved in a single hyperscale data centre build. There’s the chip manufacturer, the construction financing, the power utility commitments, the cloud customer contracts, and increasingly, private credit funds backing the build-out in exchange for yield. Each layer assumes the others will perform. Pull one thread and the whole tapestry of assumptions starts to fray, even without a single subprime mortgage in sight.
| Factor | 2008 Global Financial Crisis | Potential AI Correction |
|---|---|---|
| Primary fuel | Household mortgage debt | Corporate capex, private credit |
| Leverage source | Banks, shadow banking | Private credit funds, off-balance-sheet vehicles |
| Trigger | Housing price collapse | AI revenue failing to justify capex |
| Speed of contagion | Fast, systemic, banking-wide | Slower, concentrated, sector-specific initially |
| Who absorbs losses | Banks, then taxpayers via bailout | Tech equity holders, private credit investors, and possibly pensions |
The Oliver Wyman team is blunt about what regulators and lenders need to be doing right now: running scenario analysis under both equity and debt collapse conditions, stress-testing for a 30 to 50 per cent drop in equities, and mapping their indirect exposure to AI-linked counterparties before, not after, the correction hits. Most institutions are not there yet.
The Dot-Com Precedent Nobody Wants to Repeat
We’ve actually run this experiment before, and recently enough that the data is clean. Following the 1990s internet boom, the NASDAQ Composite crashed nearly 80 per cent from its March 2000 high to its October 2002 low, while the S&P 500 fell roughly 50 per cent. Around $6 trillion in equity valuation vanished, equivalent to about 60 per cent of GDP at the time. Unemployment peaked near 6.3 per cent, and it took 47 months for the labour market to recover and seven years for the S&P 500 to claw back its losses.
Former IMF chief economist Gita Gopinath has warned, writing for The Economist, that an AI bubble collapse on that scale could carry severe global consequences, potentially wiping out more than $20 trillion in wealth for American households and another $15 trillion for foreign investors. That’s not a regional event. That’s a global balance sheet shock.
The Labour Market Spiral That’s Already Underway
Here’s where this cycle genuinely diverges from anything we’ve seen before, and it’s the part that should worry people more than the capital markets angle. Past bubbles destroyed paper wealth and then receded. This one is simultaneously reshaping the labour market while it inflates, which means the bust and the structural disruption arrive at the same time.
Some analysts describe this as an “intelligence displacement spiral.” The mechanics go roughly like this: AI tools replace white-collar tasks, displaced workers cut spending, businesses respond to weaker demand by automating further, and the cycle reinforces itself because the technology keeps getting cheaper and more capable rather than plateauing. Traditional recessions are self-correcting because the triggering shock eventually fades. This one doesn’t fade. It compounds.
Four Phases Worth Watching
- Software collapse: Companies use AI coding tools to build internal software instead of paying for SaaS subscriptions, triggering pricing pressure and layoffs across the software sector.
- Zero-friction economy: AI agents start handling shopping, insurance, taxes, and travel booking directly, shrinking fees and commissions and squeezing traditional intermediaries.
- Wage compression: Displaced high-income professionals flood into lower-paying service roles, pushing wages down broadly since top earners drive a disproportionate share of consumer spending.
- Private credit stress: Tech and software debt issued during the growth years deteriorates as AI disrupts the business models that debt was supposed to support.
That last phase is the one that should keep risk officers up at night. A meaningful chunk of private credit financing structures arose, directly or indirectly, from insurance reserves and retirement assets. If those structures sour, the people holding the bag aren’t hedge funds with risk appetite built for it. They’re pension beneficiaries who never chose to be exposed to AI capex in the first place.
Who Actually Eats the Losses
This is the question that separates an interesting macro story from a genuine crisis, and it’s where the AI scenario starts to rhyme uncomfortably with 2008. In the financial crisis, losses that originated with subprime borrowers eventually landed on taxpayers, because the banking system was too interconnected to let fail. The bailout privatised the upside during the boom and socialised the downside during the bust.
According to research published through the Bulletin of the Atomic Scientists, if the AI bubble pops, the expected playbook is similar: the Federal Reserve steps in with liquidity injections to stabilise the broader financial system, much like it did after 2008. The concern raised by economist Andrew Odlyzko is that another bailout cycle would mean another significant jump in national debt and a further widening of wealth inequality, since the recovery in corporate profits and share prices tends to benefit the wealthiest disproportionately.
| Stakeholder | Exposure type | Likely outcome in a downturn |
|---|---|---|
| Hyperscalers (Google, Microsoft, Amazon) | Direct capex, equity | Survive, consolidate market share |
| Smaller AI startups | Venture capital, debt | Mass bankruptcies, asset fire sales |
| Private credit funds | Direct lending to AI infrastructure | Defaults, redemption pressure |
| Pension and insurance holders | Indirect, via fund allocations | Reduced returns, possible benefit-risk |
| Displaced workers | Labor income | Wage compression, underemployment |
| Taxpayers | Fiscal backstop | Higher national debt post-bailout |
Stranded Assets Nobody Budgeted For
There’s also a physical dimension to this that gets less attention than the financial engineering. The current buildout has bet enormous sums on energy-hungry data centres and specialised AI chips that depreciate fast and have limited alternative uses if demand doesn’t materialise as projected. If utilisation rates disappoint, those facilities don’t gracefully convert to something else. They become stranded assets, sitting on balance sheets as write-downs waiting to happen.
That’s a meaningfully different kind of risk than a mortgage-backed security. A house retains some value even in a downturn. A purpose-built GPU cluster optimised for a specific generation of model training has a much narrower resale market, and obsolescence in chip technology moves fast.
The Part That’s Genuinely Different This Time, and Not in a Good Way
Every previous boom-bust cycle eventually produced a silver lining. After the dot-com crash, all that overbuilt fibre-optic infrastructure became dirt cheap, and it became the backbone of the broadband internet we use today. After the railroad bubbles of the 19th century, the overbuilt rail network became the foundation of national logistics. Excess capacity from a bust gets absorbed cheaply by the rest of the economy during the slump, and that absorption is usually what seeds the next growth phase.
There’s reason to expect something similar with AI compute. Once prices crash, that capacity gets cheap, and other industries integrate the technology far faster than they would have during the expensive boom years. That’s actually the optimistic case buried inside the pessimistic forecast.
What’s different is the labour market timing. Previous tech bubbles didn’t simultaneously automate the jobs of the people who’d need to absorb the bust. This one does. The same technology causing the overinvestment is also displacing the workers who’d normally cushion a downturn by finding new employment elsewhere. That’s the part economists keep flagging as genuinely novel, and it’s why some forecasts describe this less as a typical recession and more as a structural labour market reset happening under recession conditions.
How Long Does the Bubble Have Left to Run?
Nobody has a reliable timeline for this, and anyone claiming certainty is selling something. What we can say is that the conditions for delay are unusually favourable now. The largest AI players are generating real profits, not just promises, which buys credibility. Political incentives currently favouring markets are buoyant. And global capital is still searching for somewhere productive to park itself, which keeps demand for AI-linked assets elevated even as valuations stretch.
Commentary from The Guardian’s economics desk captures the mood well: the bubble has further to run precisely because the incentives to delay the reckoning are stacked in its favour, even though a correction is treated as essentially inevitable by most serious observers. There isn’t a crystal ball that identifies the exact trigger. There rarely is, until after the fact.
What Could Actually Trigger It
- A major hyperscaler reporting AI revenue growth that meaningfully undershoots capex guidance
- A high-profile private credit default tied to data centre financing
- A sharp tightening in global credit conditions from central banks responding to inflation
- Energy price spikes that make data centre economics suddenly unworkable
- A geopolitical shock disrupting chip supply chains, like renewed tension around Taiwan semiconductor production
Any one of these alone might just cause a correction. Two or three hits lost together is what tends to turn a correction into a crisis. That’s how 2008 actually unfolded, too. Housing weakness alone wouldn’t have done it. It was housing weakness p, plus securitisation, plus thin bank capital buffers arriving together.
What This Means If You’re Not a Central Banker
Most readers aren’t running stress tests at a systemically important bank. But the practical implications still matter for individual decision-making, whether you’re an investor, an employee in a tech-adjacent role, or just someone trying to read the economic weather correctly.
For Investors
Concentration risk is the obvious concern. If a meaningful share of a portfolio sits in the small handful of companies driving AI capex, that’s effectively a bet on this specific cycle resolving smoothly. Diversification away from pure-play AI exposure, paying attention to valuation multiples relative to actual free cash flow, and understanding private credit exposure inside any bond or fixed-income allocation are all reasonable steps. This isn’t investment advice, just basic risk hygiene that applies in any late-cycle environment.
For Workers
The displacement spiral described earlier suggests that white-collar roles once considered automation-proof are increasingly exposed. Building skills that complement AI tools rather than compete directly with them, staying liquid financially, and avoiding overextension on debt during what might be a labour market inflexion point are all sensible hedges regardless of how the broader bubble resolves.
For Business Leaders
Companies that bet their entire cost structure on AI vendor pricing staying cheap and stable could be in for a rude surprise if that pricing shifts during a shakeout. Building contractual flexibility and avoiding single-vendor lock-in with AI infrastructure providers is a reasonable form of insurance against this specific risk.
The Honest Bottom Line
This isn’t a doom forecast dressed up as analysis. It’s an attempt to take the structural differences seriously. The AI investment cycle genuinely doesn’t carry the same leverage profile that caused the 2008 meltdown, and that matters. It means the transmission mechanism for a correction will likely look different and possibly move more slowly at first.
But slower doesn’t mean smaller. The combination of concentrated private credit exposure, stranded physical assets, and a labour market disruption happening in real time, simultaneous with the capital cycle, creates conditions that could produce damage on a comparable or larger scale than 2008, just through a different mechanism and over a longer timeline. Recognising that distinction matters more than arguing over whether it’s “really” a bubble at all.
Spend some time for your future.
To deepen your understanding of today’s evolving financial landscape, we recommend exploring the following articles:
The Real Reason Half of Millennials Still Take Money From Mom and Dad
Why You Overspend When You’re Sad: The Neuroscience of Retail Therapy Nobody Talks About
The Great Wealth Transfer – The Inheritance Wave Wall Street Can’t Ignore
Credit Score Mythology: The 7 Things That Secretly Destroy Your Score That No One Warned You About
Explore these articles to get a grasp on the new changes in the financial world.
Legal Disclaimer
This article is provided for informational and educational purposes only and does not constitute financial, investment, legal, or professional advice. The views and forecasts discussed reflect publicly available commentary and analysis as of the time of writing and are subject to change without notice. Readers should conduct their own research and consult a licensed financial advisor, accountant, or attorney before making investment, business, or financial decisions. The author and publisher accept no liability for losses or damages arising from reliance on the information contained in this article.
References
- O. Wyman, “How an AI Bubble Burst Could Shake Global Financial Markets,” Oliver Wyman Insights, Jan. 2026. [Online]. Available: https://www.oliverwyman.com/our-expertise/insights/2026/jan/impact-ai-bubble-burst-on-global-financial-markets.html
- “When It All Comes Crashing Down: The Aftermath of the AI Boom,” Bulletin of the Atomic Scientists, Dec. 2025. [Online]. Available: https://thebulletin.org/2025/12/when-it-all-comes-crashing-down-the-aftermath-of-the-ai-boom
- “The AI Bubble Has Further to Run Despite the Looming Crash,” The Guardian, Jun. 27, 2026. [Online]. Available: https://www.theguardian.com/business/2026/jun/27/ai-bubble-crash-tech-firms-stock-markets
- C. P. Kindleberger and R. Aliber, “Manias, Panics, and Crashes,” International Monetary Fund, Finance & Development. [Online]. Available: https://www.imf.org/external/pubs/ft/fandd/2009/03/kindleberger.htm
- Board of Governors of the Federal Reserve System, “Federal Reserve Monetary Policy Tools,” 2026. [Online]. Available: https://www.federalreserve.gov/
- Bank for International Settlements, “Global Credit Conditions Reports,” 2026. [Online]. Available: https://www.bis.org/
- U.S. Department of Energy, “Data Centre and Energy Consumption,” 2026. [Online]. Available: https://www.energy.gov/articles/data-centers-and-energy
- Council on Foreign Relations, “China-Taiwan Tensions,” Global Conflict Tracker, 2026. [Online]. Available: https://www.cfr.org/global-conflict-tracker/conflict/china-taiwan
