Is AI a financial bubble?

Artificial intelligence (AI) has emerged as the defining investment theme of the decade. From chipmakers and cloud giants to startups promising revolutionary breakthroughs, the AI boom has propelled equity markets to record highs and reshaped investor portfolios. But as valuations soar and capital floods into the sector, a pressing question looms: is AI a financial bubble?

The rise of the ‘Magnificent Seven’

Sector valuations have certainly surged. In the first nine months of the year, gains in the US equity market have been led by a handful of high-performing sectors. Communication services and information technology continue to dominate, driven by explosive growth in AI, cloud computing, and semiconductors. These sectors now account for more than 40% of the S&P 500’s weighting.

Capital has been pouring into the sector, with both retail and professional investors eager to buy into the AI narrative. This year, markets have been notably influenced by the rising power of the US retail investor. A study by JP Morgan showed that retail investors owned 36% of the US equity market in late April, following a sharp correction triggered by Donald Trump’s tariff announcements. Over the past decade, the average has hovered around 12%, but year-to-date, it’s closer to 21%.

This surge in retail ownership helps explain the rapid turnaround in market dynamics and the swift rebound of Wall Street indices to record highs. Retail investors have been actively hunting for bargains – particularly in AI – and their enthusiasm first helped stabilise the market after the tariff shock, then lifted indices from their lows at breakneck speed.

These investors are following the lead of the Magnificent Seven tech giants—Apple, Amazon, Alphabet (Google), Meta Platforms, Microsoft, Nvidia, and Tesla—which are collectively investing between $385bn and $400bn annually into AI infrastructure and related technologies as of 2025.

Investing in each other

Then comes the emergence of circular deals – a troubling sign in the current AI investment landscape.

“Circular deals” refer to a pattern of interdependent investments and purchases among a small group of companies, where money flows in a loop. This can inflate valuations and perceived growth without necessarily reflecting underlying economic fundamentals.

A prime example is Nvidia’s $100bn investment in OpenAI. OpenAI uses that capital to buy Nvidia’s chips, boosting Nvidia’s revenue and market value. Nvidia also holds stakes in CoreWeave, which supplies infrastructure to OpenAI, while Oracle – another partner – buys Nvidia chips to support OpenAI’s data centres. These arrangements resemble the vendor-financing tactics of the dot-com bubble, where companies bought each other’s services to inflate growth metrics.

The productivity puzzle

Much of the AI hype is built on the promise of transformational productivity gains – boosting corporate profitability and a country’s economic growth, ultimately creating richer societies. Optimists foresee a revolution akin to the post-war reconstruction era. Pessimists, like MIT economist Daron Acemoglu, predict more modest returns – just 0.7% annual productivity growth over a decade.

This potential disconnect between expectations and reality is a classic symptom of a financial bubble. The internet revolution’s impact on the UK economy offers a cautionary tale: while it did increase productivity, the gains were uneven and complex. A 2025 evidence review commissioned by the UK government found that the internet enabled broad knowledge sharing, remote collaboration, and automation, which improved efficiency for many UK businesses. However, the benefits varied significantly depending on the type of technology, sector, and company capabilities.

Despite these advances, the UK has experienced a notable productivity slowdown compared to countries such as France, Germany, and the US. The review concluded that structural issues – including skills gaps, management practices, and investment levels – limited the full potential of digital technologies. Moreover, the uneven diffusion of technology across sectors and regions meant that not all parts of the economy benefited equally. These skills challenges will be especially critical in the AI era.

AI’s impact on jobs

As AI evolves, it is reshaping the world of work in ways that are both exhilarating and unsettling. The impact on employment – and therefore on broader economic growth and national GDP – is deeply uncertain.

AI is often described as a double-edged sword. On one side, it promises to boost productivity, streamline operations, and unlock new economic opportunities. On the other, it threatens to displace millions of workers, particularly in roles involving routine or repetitive tasks.

Of course, AI is also expected to create new roles – many of which don’t yet exist. But as we saw with the internet revolution, the impact is unlikely to be evenly distributed. High-skill, high-wage professions – such as software engineering, data science, and AI ethics – are likely to thrive. So too are roles requiring emotional intelligence, creativity, or complex decision-making: think therapists, teachers, and strategic leaders.

Conversely, jobs involving predictable, rules-based tasks are most at risk. This includes not only factory workers and call centre agents, but also white-collar professionals such as paralegals, radiologists, and journalists. This suggests that innovative and effective government policy to upskill the workforce will be essential to achieving the productivity gains forecast by AI optimists – such as Stanford’s Erik Brynjolfsson, who argues AI could double productivity within a decade.

Investors are already expressing concern about the future of enterprise software companies, data providers, and IT consultants. For example, Adobe, the dominant player in creative design software for over two decades, is now being challenged by AI-powered image and video generation tools from startups such as Midjourney, Runway, and OpenAI, as well as internet giants like Google and Meta.

Also under scrutiny are companies like Gartner and Accenture, as their research and consulting functions may be increasingly in-sourced using AI. But this remains speculative. We simply don’t yet know which business models will be successfully disrupted – and the internet boom reminds us that such disruption is often uneven and patchy, depending on factors beyond the technology itself, such as skills training and organisational adaptability.

If these productivity gains fail to materialise, and AI does not deliver a meaningful impact on global productivity, 2025’s $400bn investment by mega-cap tech companies could come to be seen as excessive.

So, is AI a financial bubble?

It’s time to crunch some numbers – albeit with broad brushstrokes.

If we are in a bubble and it bursts, then directly related stocks could decline by 70%–90% over the following three years, while the broader market might fall by 50%. This projection is based on the experience of two previous bubbles: the dot-com boom and the Global Financial Crisis. 

It’s important to distinguish a bursting bubble from a correction, which is a short-term decline of 10% or more from a recent peak. Corrections are a natural part of market cycles and can affect individual stocks, sectors, or broader indices such as the FTSE 100 or S&P 500.

Under these definitions, there have only been two true bubbles in the past 30 years—making them rare events. Corrections, or periods of style/sector rotation, happen far more frequently and tend to be short-lived. In fact, we’ve seen three corrections in AI stocks over the past 12 months.

These were triggered by DeepSeek, a Chinese company, surprising the industry by launching a new AI model developed at significantly lower cost than Western counterparts and negative sentiment following studies by MIT, which revealed that 95% of generative AI pilots failed to deliver measurable returns, and McKinsey, which found that despite widespread adoption, most companies saw no bottom-line gains from AI investment.

These corrections have since been mostly reversed, suggesting that the burden of proof for declaring an AI bubble is high.

Former New York Fed President Bill Dudley recently argued in a Bloomberg editorial that AI will need to generate $1 trillion in annual revenue to justify the estimated $3 trillion in investment – implying a 10% return on investment at a 30% profit margin.

Currently, AI industry revenues stand at around $50bn, but are growing at an annual rate of 75%–100%. For context, global software spend is approximately $800bn per year. While software assists humans in completing workflows, AI has the potential to replace many human tasks entirely. It’s therefore not unreasonable to suggest that $1 trillion in AI revenues could be achievable within the next five to ten years. It is also important to note that these companies are generating significant revenues, something that was absent from many business ideas developed during the dot-com boom. 

Imagine AI as a gold rush

The "picks and shovels" investment strategy refers to how the real winners of the 19th-century Wild West gold rushes were not the prospectors, but the vendors who sold tools - picks, shovels, and pans – regardless of who struck gold.

In the AI boom, the modern equivalents are companies providing the hardware, software, and services that enable AI. These include chipmakers, cloud providers, and data centre REITs. They are broadly exposed to AI’s development across industries and benefit from its expansion, regardless of which specific models or platforms succeed.

Investing in these companies carries a lower risk profile than backing firms developing new AI applications that may not gain traction. These “enablers” are currently leading the charge in AI valuations, which is reassuring. But what comes next is more challenging for investors: identifying the “adopters” and “disrupters”.

Conclusion: a boom, not yet a bubble

The AI boom is undeniably reshaping markets, industries, and expectations – but it has not yet tipped into full-blown mania. Unlike the internet bubble of the late 1990s, today’s AI investment landscape is marked by substantial capital flows into infrastructure, cautious IPO activity, and a more measured pace of adoption. Most of the increase in market valuations is being led by those that provide the infrastructure, which is a good way to play this developing technology but, as we have seen in the last year, they are not immune to corrections. 

While there are signs of froth – circular deals, lofty valuations, and speculative forecasts – the absence of widespread irrational exuberance suggests we are still in the early innings of a transformative technological cycle. Whether AI delivers the productivity gains its champions promise or falls short of expectations, the next few years will be critical in determining whether this boom becomes a bubble – or a foundation for lasting economic change.