The psychology of market bubbles

Alexander Joshi, London UK, Head of Behavioural Finance

This article is technical in nature and may require some background knowledge and experience in investing to understand the themes that we explore below.

All data referenced in this article is sourced from Bloomberg unless otherwise stated, and is accurate at the time of publishing.

Concerns are mounting over whether equity markets are experiencing another bubble. A handful of artificial intelligence (AI) stocks have been responsible for a disproportionate share of the S&P 500’s returns, as analysed in Beyond the AI boom: Rebalancing portfolios for a more sustainable cycle. But just how long can such elevated spending and valuations be sustained?

Bubbles have occurred throughout history, creating surging asset values, typically followed by just as dramatic downfalls. Memories of the dotcom bubble, again involving technology stocks, in the late 1990s and the global financial crisis of the 2000s shape investors’ current anxiety.

They attract a lot of attention (see chart) as they are notoriously hard to identify in real time, and how large the stakes are. Capitalising on a bubble in its early stages can be extremely profitable, missing out can be costly, but being caught out can be disastrous. The wider contagion effects can impact even those not exposed to the relevant companies.

Tracking the number of times the term AI bubble is mentioned in Bloomberg since 2023 has surged in the second half of 2025

Source: Bloomberg, Barclays Private Bank, October 2025

An asset bubble occurs when prices rise far above intrinsic value, fuelled by speculation and enthusiasm rather than fundamentals. The result is unsustainable pricing, typically ending in a sharp correction.

There’s no universally agreed definition for bubbles, but speculative demand, psychological drivers and amplifiers are markers of the phenomenon.

Valuation parameters can help to spot a bubble, but a psychological diagnosis is essential. Bubbles reflect irrational exuberance, putting companies on a pedestal, fear of missing out and in many cases a belief that there is no price too high.

Whilst each one is different, speculative manias share similarities. As the quote goes: ‘history doesn’t repeat itself, but it often rhymes’. Economist Charles P Kindleberger’s framework outlines five key stages:

The stages show that bubbles aren’t just about excess liquidity, they involve collective biases brought out by seductive narratives:

Bubbles often accompany new developments, and AI’s rapid rise fits the pattern. The Magnificent 7 companies contributed more than 20% of global equity gains in 2025, and over 40% of the S&P 500 returns. These companies command rich valuations.

The dramatic rise in AI company spending, rather than share prices alone, is a particular concern. In the telecom bubble at the turn of the century, an exciting new technology justified unprecedented levels of investment. Eventually, supply outstripped demand, and many companies never earned a return on investment.

Since OpenAI’s ChatGPT’s release in late 2022, hyperscalers’ annual capex has more than doubled (see chart), as they bet heavily on the infrastructure needed to train and run ever-larger models. Meta, Amazon, Microsoft, Alphabet and Oracle’s combined capex is projected to rise 64% year-on-year to over $370 billion in 2025 and $500 billion in 2026.

A concern for some is also the circular nature of deals between AI model makers, compute providers, and chipmakers involved. For example, Nvidia’s $100 billion investment in OpenAI to help fund its data-centre buildout, announced in September, commits to filling those sites with Nvidia chips.

The amount of annual capex spending by hyperscalers has more than doubled since 2023

There is also macroeconomic dependence. Estimates place AI data centres among the largest infrastructure build-outs in modern history. AI data centres now drive a significant portion of US GDP growth. While not inherently bad, this dependence is risky if AI momentum stalls. If expectations break, the snapback could be severe.

The phrase “this time is different” is dangerous in investing, but there are distinctions from the dotcom era, with which many comparisons are being made.

AI is a revolutionary technology in a way in which the internet cannot compete, and today’s leading tech firms are far stronger than their predecessors, with scale, market dominance, extraordinary cash flows and above-average profit margins. Generative AI revenues are accelerating rapidly, and we are in just the beginning of enterprise adoption. The vast spending has thus far mainly been financed from cash flows.

High tech stock prices reflect a belief in their persistence. Investors are treating leading players as if they’ll dominate for decades, extrapolating today’s success far into the future. But history favours change over persistence. When companies are priced to perfection, even good news can disappoint. The damage from negative surprises can be profound.

The varied reaction to the most recent earnings and spending plans of Big Tech underscores the sensitivity of investors to how quickly the AI build-out can deliver revenue. Additional capital outlays now need to be justified, especially after the sums already deployed.

Investors will want proof that the payoff isn’t perpetually just over the horizon. Because the useful life of graphic processing units for frontier applications such as model training is low (perhaps three years), returns will be needed in a few years, not decades.

A market boom can resemble a bubble early on, with rising valuations and accelerating investment. But fundamentals − underlying cash flows, productivity gains, demand growth and profitability − eventually catch up with the optimism. Overshoots can still occur, but durable industries and lasting value emerge.

The grey zone between boom and bubble is where investor exuberance makes it hard to tell if capital is building a new economy or inflating unsustainable prices. It’s hubris to assume AI is immune to bubble dynamics, which is where the interpretation game comes in.

Trying to call a bubble or time the market is challenging, and rarely productive as it can lead to unhelpful behaviours for long-term investors. Markets can remain irrational or ‘toppy’ longer than investors expect.

Predicting the future or forecasting complex events with precision is difficult. Instead, consider a range of scenarios and position for those most likely, without deviating too far from long-term thinking informed by historical data.

For investors waiting for the perfect conditions to make investment decisions, one of the few things we can be confident of is that certainty is fleeting; uncertainty is constant.

Bubbles bursting, as well as general market pullbacks, can occur quickly and violently. This makes it imperative to understand the underlying dynamics (stages four and five of the steps outlined earlier).

The availability heuristic means that investors recall dramatic drawdowns from past bubbles and the long recovery periods then endured; 10 years for the US housing one and 15 for dotcom. The current hyperscaler spending is funded mainly by equity, which should limit contagion. Still, sharp market falls can trigger emotional overreactions, compounding losses.

Loss aversion means that volatile periods matter for investors’ long-term returns, as covered in A behavioural review of 2025 so far.

For 2026, investors should plan how to act, or not act, during any market drawdowns. Volatility is normal: the S&P 500 has averaged a 14% intra-year decline in recent decades. With equity indices unusually dependent on a few large companies, this amplifies sensitivity to stock-specific news and heightens the risk of index-level volatility if these leaders de-rate. Formulating rules for such moments can reduce biased decision-making and create opportunities for contrarians.

Shocks can hit sentiment, which can lead to indiscriminate selling. For those with and the composure to be contrarian, such periods allow quality assets to be picked up at a discount.

Most importantly, for navigating possible challenged market environments, investors should focus on committing to disciplined processes to build and maintain high-quality, well-diversified and resilient portfolios. Having strong financial foundations makes it easier to navigate the behavioural challenges of investing.

Some AI winners today won’t survive tomorrow. Schumpeter’s concept of creative destruction reminds us that capitalism’s evolution is essential for growth. Innovation disrupts industries, reallocates resources to more productive uses, and drives progress. The internet’s transformation of society continued despite the dotcom crash.

Nations embracing capitalism have seen tremendous progress as new industries replace old ones. Whilst hyperscalers might shoulder the capex burden, the real winners of this race may be those controlling models, data and ecosystems. AI’s redrawing of labour markets is already visible, as some roles become redundant and new ones are born.

The importance of (cautious) optimism

Bubbles are stories we tell about the dangers of over-optimism, but optimism is vital for economic growth and that of investment portfolios. Losses hurt and understandably their prospect worries investors.

But free markets and financial systems create progress despite adversity. Flexibility and adaptability are crucial for prosperity; and investors must stay alert to transformative trends. Importantly, they must be participating to be able to benefit.

AI exuberance and lofty valuations may make investors hesitant to increase equity allocations in 2026. For those seeking to protect and grow wealth over the long-term, investing in quality assets, selectivity, and diversification are key.

All-time highs: Markets hitting record valuations doesn’t mean a pullback is on the cards. Equity prices aren’t mean-reverting; markets have spent about 30% of the time since 1990 at all-time highs.

Quality: Momentum and high-volatility stocks have dominated in 2025 as market dynamics shifted from fundamentals. That said, quality has outperformed over the long term, as examined in Testing the limits of concentration and resilience of returns. The return on an investment is a function of the price paid for it. A high starting valuation typically leads to lower returns, and vice versa.

Diversification: Diversification across asset classes, regions and sectors is essential. Opportunities abound for those who look beyond the obvious.

Hedging: Protection through hedging strategies allows participation without overhauling portfolios.

Interpreting signal from noise will be crucial for rational decisions in 2026 aligned to investors’ goals.

This is easier said than done. Humans connect with stories (especially ones easily followed, even if not fully backed by facts), and more emotive or dramatic ones engage the human brain’s intuitive and associative ‘System 1’. This makes them masters of pattern recognition, but also leaves them vulnerable to the same kinds of biases that affect human intuition. Engaging the more deliberate, analytical ‘System 2’ necessary for interpretation and nuance can help investors avoid being swept up by narratives.

Good decision-making allows for contradictory ideas: AI can be revolutionary and create value, but not all AI companies will be good investments.

We believe this more nuanced view will serve investors better than a binary yes/no opinion on the bubble question, which can typically lead to all in/out exposure decisions. A nuanced, bottom-up approach to portfolio construction is also useful, especially amid uncertainty and change.

The latest iteration of AI is likely only at the beginning of its journey, as examined in much detail in our AI Outlook 2026. Whether boom or bubble, the implications for investors in 2026 and beyond will be significant. Measured, thoughtful decisions − grounded in an understanding of both market and investor psychology − will be essential.

After a solid year for financial markets despite trade tensions and a redrawn world order, what next for 2026?