Introduction
Recent discourse on the US economy has centered on a supposed AI-driven bubble. Nearly $1.5tn is being poured into data centers, microchips, and R&D in 2025, despite vague, circular business models. This has led regulators and experts (e.g., Michael Burry betting against Nvidia [NASDAQ: NVDA] and Palantir [NASDAQ: PLTR]) to sound the alarm that the massive gains in public markets aren’t reflective of actual intrinsic value, or, in other words, that a bubble exists. It is quasi-impossible to predict or detect bubbles until they pop. There is a famous joke in finance regarding this fact that has evolved throughout the years:
“Wall Street indexes predicted nine out of the last five recessions!” – Paul Samuelson (1966)
At its core, this joke, recently adapted for Michael Burry (“Dr Burry has predicted twenty out of the last two recessions”), points to two key pillars of public markets: irrationality (not to say private markets are always rational) and its decoupled nature. Asset prices in public markets are dependent on supply and demand. Regardless of a stock, bond, or commodity’s intrinsic value, its price is determined by buyers and sellers. Of course, purchasing behaviour should, in theory, be based on financial fundamentals, but this is not always the case. Market participants, such as the 18-year-old retail investor armed with pocket money and a dream, or a highly skilled portfolio manager, can all be irrational at times. This irrationality can be informed or oblivious. If informed irrationality sounds strange, imagine this: Palantir has kept an abnormally high P/E since 2023, currently at 385.51 as of 28/11/2025. This unusually high P/E raises the obvious question: Who in their right mind is buying this stock? On any trade, the fundamental question is simple: What return can I get at what risk? Palantir has risen roughly 150% in the past year. Regardless of any lack of fundamentals (immense growth potential exists, but not at this value), if rational investors believe demand will be high (often driven by irrational buyers) and that they can achieve massive returns at low risk, they will invest. Palantir may have been a stupid trade a year ago, but it sure was profitable. This irrational rationality (beyond merely creating fertile ground for bubbles) muddies the ability to discern whether a bubble exists or when it will pop. This core fiatesque function of the modern stock market allows, and many would argue, incentivizes this very form of speculation and fuels bubbles.
This article will delve into the three main categories of bubbles and their underlying rationales, offering a new perspective on the recent supposed AI bubble. We will explore how occurrences or even perceptions of rapid innovation or financial euphoria tend to lead to mispricing, exaggerated expectations, and runaway narratives. Exploring past cycles and phenomena will help in determining the potential effects of an AI bubble burst.
Financial-Engineering Bubbles
Financial engineering is the application of logic, mathematics, economic theory, computer science, and statistics to create financial products, strategies, and structures. This is often seen as creating parasitic value-extracting solutions and tools that inevitably hurt most stakeholders through predatory schemes and, in recent cases, bubbles. While this does occur, one must remember that financial engineering has ameliorated and lubricated the activity of the world economy. The invention of futures on currencies or commodities has created protections for agents such as farmers or international firms. Risk-management models have increased awareness and control over risk mitigation, enabling more efficient markets. Structured products such as mortgage-backed securities (MBS) have raised liquidity in the housing industry while also providing diversification for investors. Financial Engineering bubbles occur when newly minted financial products, often complex and misunderstood, are released into the market, where investors seeking to maximise returns and minimise risk begin pushing prices far beyond economic fundamentals. These products tend to rely on complex mathematical models and tangled structures, allowing for perversion of the original asset and expanding the original scope of the bubble. The unique aspect of these bubbles is investors’ reliance on the asset’s engineering to limit risk. This claim of risk elimination is often false, as products hide risk behind models and structures rather than eliminating it. Financial engineering bubbles pop because the engineered structures rely on models and assumptions that fail under stress, leading to hidden risks caused by amplified leverage and liquidity mismatches that unwind violently when market conditions shift.
The Global Financial Crisis (GFC) of 2008:
“Only when the tide goes out do you discover who’s been swimming naked.” – Warren Buffett (1994)
In 1977, Lewis Ranieri helped securitize and create the first private-label MBS. An MBS is a debt instrument that has rights to the cash flows from a pool of mortgages (highly similar to traditional bonds). Rather than being backed by federal agencies, these new securities were supported by banks. The idea was, frankly, genius because it unlocked an immense, previously illiquid asset class. Before 1977, banks were forced to hold low-yielding, long-term mortgages on their balance sheets. With an MBS, banks could securitize mortgages and sell them to investors, freeing up balance sheets. This sale created a cascade of fixed-income securities, generating huge fees for banks as they had to structure each MBS, increasing their income significantly compared to the basic underlying mortgages.
After the Dot-Com bubble (discussed later in this article) and 9/11, rates dropped sharply, prompting investors to seek safe, high-yielding assets, which led them to MBS. Mortgages were seen as safe and perennially lucrative. With this surge in demand for MBS, lenders began loosening underwriting standards to issue more mortgages, introducing “NINJA” (no income, no job, no assets) loans, and teaser rates (mortgages with initially low rates that increase after a set period). The decline in loan quality was irrelevant to the banks, as they sourced the loans and sold them into the public markets immediately. Investors, on the other hand, were unaware of the complex structuring of these products and, in many cases, were misled. An AAA MBS tranche could be created from a mortgage pool with risky loans, and, due to market sentiment and opaque structures, remain unaware of this. Rating agencies should have reported these risks, but due to misaligned incentives (issuers of MBS paid rating agencies the assets they were going to sell) and a complete lack of regulation, they didn’t. When teaser rates reset in 2006, defaults skyrocketed, and mortgage prices began falling. The fun was over. Massive losses on the MBS markets spread like wildfire.
Several factors amplified the scale of the spread. MBS were treated as virtually riskless and used as collateral in the repo markets. This enabled banks to overleverage their balance sheets, ultimately leading to the failure of various financial institutions, including Bear Stearns and Lehman Brothers. Creative financial engineering created securities such as the Collateralized Debt Obligation (CDO), which repackaged risky loans together to diversify risk, or CDO-squared, which was a CDO backed by other CDOs, and ultimately the synthetic CDO, which, through credit default swaps, gave investors exposure without any underlying asset. Ultimately, this bubble was a travesty unseen in recent history. Billions evaporated, and even countries such as Ireland almost failed. This bubble and the eventual recession were purely man-made, driven by greed and carelessness. Aside from the onslaught of regulations regarding capital requirements, transparency, and risk retention, little was materially gained from this housing market crash.
“Volmageddon” – The Short Volatility Crash of 2018:
Following years of quantitative easing and the European debt crisis, global markets entered a period of relative tranquillity, during which the VIX (volatility index of S&P 500) frequently remained below 12, and volatility spikes were short-lived. Investors seeking high yields began shorting volatility through exchange-traded notes (ETNs) such as the VelocityShares Daily Inverse VIX Short-Term ETN (XIV) issued by Credit Suisse. The XIV provided the inverse of daily VIX futures performance; when the VIX futures index declined, XIV rose, and vice versa. From 2012 to 2017, this worked fantastically. In 2017, the VIX hit its lowest annual average ever, which led to explosive growth in XIV and other short-volatility strategies. These products became so large that their rebalancing operations impacted the VIX futures market itself. To understand this, one must grasp the actual execution of the XIV.
The XIV did not track the VIX itself, but the inverse of a short-term VIX futures index. Each day, XIV adjusted its notional exposure to maintain a –1× relationship with that index. In calm periods, this produced consistent gains. However, even a moderate volatility spike could trigger a destructive chain reaction: during a volatility surge, XIV would be forced, under its daily reset rule, to buy VIX futures in a rising market to reduce its inverse exposure. When in a calm period, this strategy was effective, but even a moderate volatility spike could begin a destructive chain reaction. At the worst possible time, during a volatility spike, the XIV would be forced to buy VIX futures to reduce its exposure. Importantly, the XIV and other ETNs of its type had a clause allowing the issuers to redeem the product (end its operation) if its value fell by more than 80% in a single day.
On the 5th of February 2018, the S&P 500 dropped 4%. This would usually be an insignificant drop in the history of this index; however, due to the influx of counter-volatility strategies, a cycle emerged. As volatility increased, the XIV and other ETNs lost value, forcing them to buy VIX futures to rebalance. This further increased the price on VIX futures, further increasing the loss for the XIV. Falling past the 80% trigger, Credit Suisse closed the XIV, wiping billions from the market. These products failed due to their complexity and non-linear exposure. Linking them to the futures rather than to the spot VIX created unexpected sensitivities. Small changes in volatility created disproportionate portfolio impacts. Daily rebalancing created a vicious positive feedback loop with no circuit-braking systems in place. Ultimately, most investors didn’t understand the mechanics of these ETNs and believed that the XIV was solely a bet against fear rather than a complex financial instrument. Thankfully, due to low exposure on banks’ balance sheets, limited use as collateral, and limited synthetic replication, this crisis remained largely contained. It does serve as a good lesson on how complex products can cause a crisis even when the underlying market is behaving normally.
Asset Scarcity Bubbles
‘Asset-Scarcity’ bubbles are a type of cycle in which the price of assets decouples from their fundamental value and climbs unjustifiably on the premise of perceived rarity or irreplaceability. This constrained supply can occur naturally (e.g., in the price of commodities like gold) or be caused by regulatory environments or market participants’ perceptions. The mechanism that turned this perceived supply bottleneck, be it authentic or not, into a bubble is the feedback loop it conjures. As prices rise, the fixed supply forces higher demand, which in turn fuels further price escalation. Common catalysts for this are new markets (foreshadowing the AI “bubble”), novel assets, and technological or cultural shifts. As excitement builds, the scarcity narrative plays on the bias colloquially known as FOMO (fear of missing out), which overrides rational market behaviour and valuation metrics.
Tulip Mania:
One of the oldest and most well-known examples of this category of bubble is the Dutch Empire’s Tulip mania. Tulips became a status symbol during the empire’s golden age, and due to their rarity, tulip bulbs (essentially the flower’s seed) became a storehouse of wealth. Because they were scarce and culturally sought after, market participants engaged in bidding frenzies. Futures contract-like instruments enabled leveraged speculation far beyond the physical bulb supply. This expansion of forward contracts drove a synthetic demand and pulled capital from diverse sources. Throughout the 1630s, participation in this market swelled beyond elite collectors into broader market investors, as the trading system lacked enforceable margin rules that further magnified volatility, until the market eventually collapsed in February 1637. The triggers that ultimately caused the market to unwind were failed settlements and a sudden drop in buying interest as contract prices collapsed amid evaporated market confidence.
Silver Thursday:
Throughout 1979-80, the brothers Nelson and William Hunt attempted, and to a degree succeeded, in cornering the silver market through a massive accumulation of physical and futures positions. Inflation fears at the time diminished consumer confidence and prompted people to buy up physical store holdings of wealth, leading to a rally in silver. As market participants realized the brothers’ plan, they tried to beat them to the chase, hoping to benefit from their expected cornering of the market and turn the rally into a bubble. Resembling the tulip bulb bubble, futures activity created leverage far more than the market’s actual depth. The Hunt brothers financed these futures primarily through debt. At their peak, they held a third of the world’s privately traded silver.
The psychology behind this bubble was the belief that silver would be a fail-safe hedge against monetary instability and that the Hunt Brothers’ rally would continue indefinitely. Leveraged futures and large physical asset purchases siphoned institutional capital into one crowned trade. The unwind trigger was the New York Commodities Exchange) which, upon realizing the emerging systemic risk posed by such a volatile market, implemented the “Silver Rule 7” on January 7th, 1980. The regulation sharply limited margin-based purchases of silver; traders could purchase silver only with essentially a 100% upfront payment. As new margin calls were disallowed, the synthetic demand that had been boosting prices diminished, and the Hunt brothers and other major shareholders faced massive and cash-heavy margin calls. On March 27th, 1980, they missed a $100 million margin call, and that day would henceforth be known as Silver Thursday. As the price dropped and people’s confidence in the asset’s continued rise faltered, options traders lost massive sums, and the bubble burst.
Japanese Real Estate:
Throughout the 1980s, the Japanese government engaged in aggressive monetary easing, primarily through lower interest rates, spurring a credit expansion. At the time, land was viewed as uniquely scarce in Japan, fostering the belief that asset prices could never fall and that any investment, no matter its debt burden, would be sound. “Land never goes down” became a national narrative. Banks responded by lending heavily against rising collateral values, creating a circular loop in which higher prices justified more lending, in turn pushing prices even further. Since real estate is one of the most common forms of collateral, the increase in lending spread far beyond the asset class. Corporate crossholdings further reinforced this cycle, as firms used inflated asset values to finance additional real estate deals.
Inevitably, as inflation pressures rose and asset prices spiralled away from fundamentals, both the Bank of Japan (BoJ) and the Ministry of Finance stepped in. The BoJ raised interest rates from 2.5% in 1988 to 6% in 1990. The Ministry of Finance imposed controls on lending, instructing banks to cap the growth of their real estate lending to the rate of their total loan book. Once lending growth slowed, collateral values fell, forcing asset sales and, eventually, returning real-estate prices to a more intrinsic value.
Technological & Industrial Expansion Bubbles
“Industrial/Technological Expansion Bubbles” generally consist of a speculative boom which arises when a new technology or industrial paradigm enters the market and is expected to deliver transformative productivity gains, which leads the market to price in a scale the speed of value creation that far exceeds a technology’s future of free cash flow growth, which can realistically be achieved. Rather than being driven by a fixed supply of a valuable asset or through financial engineering, these bubbles are anchored in a growth narrative in which investors and corporates convince themselves that a fresh technological “revolution” whether it be in railways, electrification, information technology and more recently on the field of artificial intelligence can constitute a structural break which can realign previous expectations on productivity levels, margins, and growth potential.
However, the defining feature of this bubble type is the light, self-reinforcing linkage between markets and real-economy investment. Early productivity gains and rapid adoption draw in capital, then rising valuations lower the cost of funding, and ultimately, cheap capital enables companies to pursue aggressive CAPEX, rapid capacity buildouts, ambitious M&A, and loss-making land-grab strategies that would be economically unacceptable under different conditions. These expansions temporarily inflate growth metrics and create the appearance of validation, thus attracting further thematic inflows. Ultimately, the sector becomes highly overlapping, with multiple players pursuing similar strategies, often resulting in duplicated infrastructure and chronic overcapacity. However, as demand growth cools, adoption slows, pricing power disappoints, and capital conditions tighten, this feedback loop unwinds as funding windows close. Valuations mean-revert towards levels more consistent with realistic free cash flow expectations. Despite the chaotic nature of this type of bubble, throughout history we have seen that both in the cases of the railway boom of the 1840s and the dot-com bubble of the early 2000s, valuations collapsed during the burst, however, the technologies spearheading the bubble indeed revolutionized the global economy and the role of their respective technologies is still relevant in the modern day, in addition, the leading players in the bubble far survived the burst and ultimately became some the most prominent corporate players in the post-bubble financial markets as we will explore in our upcoming deep dive of both the Railway Mania of the 1840s and the Dot-Com bubble of the 2000s.
Railway Mania:
Railway Mania, which took place in Britain during the 1840s, stands out as a leading example of an industrial bubble. Diving into the backdrop, at the time of the bubble, Britain was undergoing a drastic technological transformation fueled by the fruits of the First Industrial revolution and the innovations that followed, one of them being the railway, which at the time was rapidly expanding in terms of its kilometers of train tracks as well as the regions it was covering and what it meant for rapid transportation and productivity amid the rapid industrialization of the country. The expansion of railways track gathered an even more accelerated pace from 1843 to 1847, with 1846 alone accounting for over 260 Acts of Parliament which authorized new railway companies and routes totaling 9,500 miles and capital surged from low double-digit millions of pounds per year to an investment peak in 1847 equivalent to a material share of British GDP, while an index of railway shares more than doubled between 1843 and 1845, in addition, most shares were only partially paid (usually at an ownership margin of no more than 10%) which allowed middle-class households to take significant exposures with limited capital, which further magnified speculation over the British railways sector.
The feedback mechanism of this bubble was classic as rising prices encouraged speculative players to buy more shares, pushing valuations higher and validating overoptimistic profit projections. When the Bank of England raised rates and safer yields became more attractive, investors started to doubt the validity of companies in the railway sector and the possibility that they could achieve the returns they had been advertising, and they faced mounting cash strains. Ultimately, confidence cracked, and secondary market share prices fell sharply. In context, railway indexes fell by roughly half between the mid-1850s and 1850 with hundreds of companies filing for bankruptcy and ceasing operations as CAPEX plans were impossible to handle, in addition led to smaller companies being absorbed by bigger rivals such as the Midland Railway company, Lancashire & Yorkshire Railway and London & North Western Railway which consolidated as the new post-crisis players through aggressive takeovers. However, despite the chaos the bubble’s collapse brought, it left behind a vastly integrated rail network in the UK, with thousands of track miles under construction, which materially reduced transportation costs and spurred significant industrial gains for the UK throughout the remainder of the 19th century.
Dot-Com Bubble:
Following the late 1990s dot-com bubble, which serves as our example of a technological bubble, the dot-com bubble provides a modern parallel, this time around the commercial internet and the field of information technology. As the World Wide Web moved from an academic novelty to a mainstream platform, investors, corporations, and policymakers converged on the idea of a “new economy” led by online businesses and digital infrastructure, something expected to transform commerce, media, and communications radically. As the decade progressed, capital markets and investment banks responded with an unprecedented rerating of technology and internet-related stocks, as the tech-heavy Nasdaq Composite Index roughly quintupled between 1995 and its March 2000 peak. On this backdrop, the number of companies engaging in highly successful IPOs ballooned, with companies such as Netscape, VA Linux, and theGlobe.com experiencing skyrocketing valuations that eventually collapsed in most cases after the bubble burst. In this backdrop, loss-making startups achieved peak multi-billion-dollar market valuations based on metrics such as page views and user growth rather than traditional free cash flow metrics. Investors entered the sector as venture capital poured into dot-com startups that, in many cases, lacked a profitability plan or a specific product. In addition, and adding more speculation, retail investors joined via online brokerages and margin accounts, thereby boosting market valuations. In this case, cheap capital enabled aggressive spending on marketing, acquisitions, and, most visibly, network build-outs and “get big fast” strategies, which temporarily boosted top-line metrics to justify high valuations and encourage the banks to bring more listings into the public markets. However, below the surface, business models were often fragile, with customer acquisition costs unsustainably high, weak unit economics, and a crowded competitive landscape. As interest rates rose between June 1999 and May 2000, several high-profile companies missed expectations. They failed outright, prompting analysts to question their business models and the realism of their growth expectations. As the bubble burst in the first half of 2000, the Nasdaq fell by almost 80% between 2000 and 2002, most dot-com companies went bankrupt or were acquired at fire-sale prices, yet similarly to the bursting of the Railway bubble of the 1840s, the aftermath paved the way for growth and consolidation in the internet sector and its infrastructure. As a positive, the bubble accelerated the installation of fiber-optic networks, data infrastructure, and software capabilities, and it trained a generation of engineers and entrepreneurs who defined the new era of technological companies, with Amazon being a leading player that survived and thrived in the post-dot-com era. Now, during the 2000s, investment in the technology sector took place with much greater discipline, leading to continuous technological growth and paving the way for the high adoption and large-scale role of the internet we see today.
Common Traits across Bubbles
Throughout the analysis of asset scarcity, technological/industrial expansion, and financial engineering bubbles, a few common key traits have emerged that are worth noting. First, every bubble has a sense of inevitable improvement, in which rising prices are seen as confirmation of a new paradigm, driven by viral narratives, and as indicative of future price appreciation. As seen in the Japanese real estate bubble, the sentiment of “land always appreciates” drove an irrational disregard for intrinsic valuation. Another recurring theme is the overestimation of total addressable markets (TAM). Investors project universal adoption, unlimited demand, or physical constraints that permanently favour higher prices. While equally irrational as using past prices as indicators of the future, TAM overestimates are not unique to bubbles and creep up wherever human optimism takes hold. The flip side of this is an underestimation of compression risk, in which prolonged stretches of price stability/improvement erode realistic risk assessments. As historical data is used to extrapolate into the future, volatility and tail events, particularly in the face of a cultural fad, are mis-priced, and models become overly optimistic, forcing a feedback loop. This is typically worsened by excessive leverage enabled through high credit availability and low-margin derivatives. While this allows investors to control large positions with minimal capital, it creates artificial demand. Even slight downward price moves may trigger margin calls, exacerbating both the expansion and the collapse of the bubble. Another characteristic of this cycle is an extreme concentration of capital, as diverse banks, funds, and retail investors pile into the same trade. This reduction in market diversity magnifies risk and creates a feedback loop: disproportionate inflows push prices up, mirroring attractive growth, attracting more investors, and magnifying risk. Finally, and perhaps most importantly, a bubble would not be a bubble without the sentiment that “this time is different”. Time and time again, market participants are of the strongest convictions that, for whatever reason, historical constraints don’t apply to this specific rally. However, as we’ve seen, this is sadly never the case.
AI Cycle: Is this time different?
All the prior analysis leads us to this question: Is the current AI cycle any different from the bubbles we saw throughout the article? However, the response could not be more nuanced, as we can see that the current stage of AI combines aspects of industrial/technological bubbles with its role in the new productivity and innovation rate, and some aspects of asset scarcity bubbles through its high barriers of entry and complexity of setting up data center infrastructure. On the one hand, we see an already substantial base of real revenues, with hyperscalers and leading chipmakers reporting tens of billions of dollars tied directly to AI-related compute, cloud services, and enterprise software, as well as existing multi-year contracts that specify set CAPEX numbers that companies are investing. On the other hand, equity markets are not only capitalizing on existing cash flows but also making very aggressive assumptions about long-term sector dominance, network effects, and monetization, which can be questionable given that most AI ventures are currently in an intense cash-burning stage.
What distinguishes this cycle is the CAPEX intensity of building and operating. AI infrastructure, which demands enormous upfront outlays on data centers, specialized semiconductors, power, and networking. In terms of its spending profile, it can resemble that of the railway boom, where vast sums of capital went to build infrastructure and were dominated by a few players through high entry barriers, operating similarly to the present, as a small cadre of platforms, chip vendors, and cloud providers have captured high returns and market expectations. At the same time, and more similarly to the dot-com bubble, we also have a large group of smaller startups raising large amounts of venture capital as they seek to capture AI growth by providing AI platform-as-a-service.
However, on the AI growth, the elephant in the room is that geopolitics play a leading role as the AI growth dynamic is reinforced by a strong layer of support from the national governments, particularly in the U.S. and China as both countries face each other on a new AI arms race where AI and data centers and chips infrastructure being defining factors in each economy’s respective productivity, automation. More importantly, military capabilities in warfare have become a crucial aspect where AI is used. In such a framing, the question becomes less whether individual projects will meet conventional hurdle rates and more whether a country can afford not to invest. Given that, the technological race and financial visibility sometimes become secondary concerns. This stance can ultimately inflate a bubble by keeping capital flowing into the AI theme long after commercial signals would have called for greater discipline. Still, it also can make the cycle more productive on the run as accelerated deployment of computing power and infrastructure will lead to the consolidation of a core infrastructure and an ecosystem which given the fact that AI is here to stay, will pave the way for an economy where AI becomes even more integrated, similarly to the internet after the burst of the Dot-com bubble which not only retained its role but further expanded to become arguably, one of the most crucial aspects of the modern economy.
The key analytical question, therefore, is not whether there is speculative excess in AI investing and expectations, which there clearly is, but more of how much of today’s CAPEX and operating spend is building durable infrastructure versus transient business models if the balance skews to the former, history suggests that even a painful correction for investors could still leave the countries which are doubling down on AI investment as well as corporate early movers with a unique strategic advantage that would endure and solidify itself as a pillar of the economy.
Conclusion
Bubbles have a terrible connotation. Much of this is rightfully due. Financial institutions have, and most likely will in the future, propagate crises, causing widespread damage for no apparent reason. This view is a tad myopic, however. While many financial bubbles (specifically those based on financial engineering) yield no future benefit (except for regulatory measures to prevent further crises), other bubbles result in growth. In nature, populations of R-Species such as rodents and insects will expand and contract violently. No ecologist would say this cycle is wrong; instead, they would say it is nature. The fundamental structural reason behind bubbles might be human nature, which, unless we all disappear, is unavoidable. We, as a society, tend to overprice the short-term costs and underprice the long-term impacts of bubbles. Even with the risk of an AI bubble, it might be beneficial to inflate it. This technology is useful. Overinvestment now, while potentially damaging in the short term, will reap dividends for years to come. The price we pay for innovation and an efficient economy might have to be the occasional bubble.
0 Comments