AI stocks have surged since the spring, leaving investors caught between conflicting warnings and bullish predictions. The question of whether there’s a bubble in AI stocks has become impossible to ignore — especially as certain index funds grow increasingly concentrated in a handful of technology giants. Are we witnessing a genuine productivity revolution, or reprising the dot-com script?

 

Everyone’s talking about a possible bubble in A stocks. Sam Altman, CEO of OpenAI, recently warned that “investors as a whole are overexcited about AI”, predicting someone will “lose a phenomenal amount of money.” 

Lisa Su, CEO of AMD, which makes AI-capable chips, disagreed. AI sceptics, she said, are “thinking too small” about what she calls a multi-year AI “supercycle”.

For investors, the stakes couldn’t be much higher. Getting this wrong means either missing historic gains or suffering catastrophic losses.

This article won’t predict what AI stocks will do next. Nobody can. But it will give you an evidence-based framework for thinking through the question, rooted in academic research, historical precedent, and the uncomfortable realities even well-diversified investors now face.

 

Why AI stocks surged in 2025

Before assessing whether there’s a bubble, it’s important to understand why AI stocks have risen so sharply. 

The primary force: enormous capital expenditure by large technology firms, often called hyperscalers. These companies are investing tens to hundreds of billions of dollars collectively to build AI datacentres and cloud computing capabilities. This spending drives demand for firms providing the underlying hardware.

Crucially, there’s real, rapidly growing demand for AI computing power to train and run increasingly complex models. This creates multi-year commitments, often pre-paid. Unlike previous speculative booms, current AI spending is funded by operating cash flow and profits of established technology companies — genuine financial strength, not venture capital froth.

Companies holding critical positions in the AI ecosystem have seen valuations soar. Nvidia’s dominance in GPU design creates significant barriers to entry. Cloud platform providers profit from both infrastructure and applications. Chip manufacturing remains concentrated in a handful of firms — Taiwan Semiconductor Manufacturing Company produces the advanced chips, while ASML Holding is the sole provider of extreme ultraviolet lithography machines essential for making them.

Companies such as these are projected to deliver high future revenue and earnings growth, leading investors to bid up prices today.

The question isn’t whether these drivers are real. They are. The question is whether current valuations already reflect — or overshoot — the economic value this technology will ultimately create.

 

What history teaches us about technology and speculation

To evaluate today’s AI market, we need historical context. The most relevant comparison remains the dot-com bubble of the late 1990s.

The internet genuinely transformed everything — commerce, communication, media, education. The technology was real. The revolution was real. Yet the dot-com bubble became one of the most catastrophic periods for investor wealth destruction in modern history.

During the mania, investors poured money into virtually any company with “.com” in its name, regardless of business model, revenues, or path to profitability. The core belief: the internet would transform everything, and getting in early was paramount. Traditional valuation metrics — profits, revenue growth, even coherent business plans — were dismissed as obsolete thinking.

The market peaked in March 2000. When companies failed to deliver profits matching their valuations, and investors realised many business models were unsustainable, the collapse began. The NASDAQ fell 78% from its peak. Many stocks went to zero. An early 2000s recession followed.

The lesson: revolutionary technology doesn’t automatically translate into profitable investments during speculative excess.

Historical bubbles share remarkably consistent patterns. There’s always a “this time is different” mentality — a belief that the new technology is so revolutionary that traditional valuation metrics no longer apply. Investors prioritise the potential size of the total addressable market over current financial reality. There’s a flood of new, unproven companies with the right buzzwords but little substance. Speculation feeds on itself through herd behaviour. Fear of missing out overwhelms discipline.

These patterns demonstrate that genuine technological revolutions frequently coincide with speculative manias that destroy investor wealth. The internet transformed the world, yet most 1999 internet investors still lost money.

 

The case for a bubble in AI stocks

Growing numbers of economists and industry observers argue that the AI stock boom shows classic signs of speculative excess.

Jeffrey Sonnenfeld and Steven Henriques, in their recent analysis for Yale and Fortune, highlighted how investor enthusiasm has pushed expectations far ahead of what the technology currently delivers. They described investor behaviour as displaying the “tell-tale psychology of mania”: narrow market leadership, investors chasing whatever carries an AI label, capital flooding into projects long before commercial payoffs appear.

The central concern: the disconnect between investment and tangible return. Sonnenfeld and Henriques pointed to research showing organisations had poured tens of billions into generative AI initiatives while reporting little to no measurable productivity impact. 

Economist Michael Roberts made a similar point, arguing that AI’s extraordinary share of recent economic growth reflects investment intensity rather than the technology’s current ability to lift productivity across the wider economy. Today’s valuations, in Roberts’s view, embed hopes of unprecedented productivity gains that data simply don’t support.

This gap invites comparison with earlier speculative overreach. Pedro Carvão, writing in Harvard Business Review, warned that financial structures emerging around AI — large cross-investments, revenue booked through related parties, deals designed to signal momentum rather than fund operations — echo patterns from the dot-com run-up. These circular arrangements, Carvão argued, inflate perceived opportunity without creating underlying economic value.

The behavioural dynamics add weight. Sam Altman’s warning that “investors as a whole are overexcited” came from the chief executive of the company at the centre of the boom. Jeff Bezos, hardly a technology sceptic, described an environment where “every experiment gets funded” — implicitly comparing it to earlier periods when fear of missing out overwhelmed discipline.

Even the Bank of England has taken notice, warning that expectations around AI have become a potential market vulnerability.

These voices suggest a market driven less by present fundamentals and more by unanchored expectations, herd behaviour, and hopes of extraordinary future payoffs. The risk: AI’s promise — real though it is — has been priced in too early and too aggressively.

 

The bullish case: why high valuations may be justified

Others, including prominent economists and technologists, argue high valuations reflect reality: AI is a general-purpose technology with potential to reshape the global economy over decades.

AMD CEO Lisa Su pushed back strongly against “overhype narratives”, arguing sceptics are thinking too small. She described AI as being at the start of a multi-year supercycle comparable to earlier industrial shifts.

This optimism mirrors a long academic tradition. William Janeway, an economist specialising in innovation cycles, noted that speculative excitement often accompanies transformative technologies. Episodes of over-enthusiasm can coexist with — and indeed help finance — the infrastructure and experimentation required for genuine breakthroughs.

Railways, electrification, and the early internet all went through periods of over-investment that funded infrastructure far ahead of demand. What looked like a bubble at the time turned out to be a necessary overshoot enabling long-run productivity and growth. Applied to AI, high valuations may reflect not irrational mania but rational confidence in a general-purpose technology still in its infancy.

Unlike loss-making dot-com companies of the late ’90s, leading AI-related companies are established, profitable, and already generating commercial returns from cloud services, data-centre hardware, and AI-driven enterprise tools. Even Sonnenfeld and Henriques, generally cautious, acknowledged that the earnings power of today’s dominant technology platforms sets them apart from companies that collapsed in earlier speculative periods.

A second counter-argument concerns financial stability. Pierre-Olivier Gourinchas, the IMF’s chief economist, pointed out that the current AI boom is equity-financed, not debt-financed. Bubbles become dangerous when leverage amplifies losses, as in the 2008 housing crisis. Without that debt element, a sharp fall in AI equity prices — while painful — would be far less likely to trigger systemic damage. Even a dot-com-style correction would be economically manageable.

Finally, proponents argue use-cases are emerging faster than in previous technological cycles — from AI-assisted software development to protein design, logistics optimisation, and scientific modelling. If commercial adoption continues accelerating, today’s valuations could increasingly be supported by earnings rather than expectations alone.

The case against the “AI bubble” narrative isn’t that valuations aren’t high. They clearly are. It’s that underlying technological and economic forces may be strong enough to justify them over a long horizon.

 

Why index investors should check their AI exposure

Index investors have done particularly well from the AI stock boom. If you’ve held a global tracker or S&P 500 fund through 2025, you’ve captured the rally automatically — exactly as you should.

But there’s an important consideration: the handful of technology companies deeply involved in AI infrastructure now constitute an unusually large share of the S&P 500’s total market capitalisation. Because the index is market-cap weighted, the unprecedented size of these few companies means their movements have an outsized influence on your returns.

Top holdings in many broad market indices now represent a larger share of total market capitalisation than at almost any point in history. The S&P 500 is currently more concentrated than it has been historically, with a significant percentage of its movement traceable to a handful of companies linked by the AI theme.

This isn’t necessarily a problem — it’s simply a feature of how market-cap weighted indices work. When certain sectors surge, they naturally become a larger part of your portfolio. You’ve captured those gains, which is precisely what index investing is designed to do.

The question is one of awareness rather than alarm: do you know how much of your total wealth is now tied to AI-related companies?

If you own a global tracker, a US index fund, and perhaps a growth fund or technology allocation, you might discover you have 25% to 30% of your wealth concentrated in a handful of AI-related technology companies across those holdings. That’s not an error in your strategy — it’s a natural consequence of how these companies have grown. But it may be more concentration in a single theme than you realise or intend.

The solution isn’t to abandon index investing. It’s to periodically check your total exposure and ensure it aligns with your risk tolerance and financial plan.

 

What the evidence-based framework actually says

Here at rockwealth, we strongly believe in what we call evidence-based investing — grounding every decision we make in what the peer-reviewed academic evidence tells us.  

So what guidance does the research give us when it comes to the AI stock boom?

 

Market efficiency: if it’s obvious, it’s priced in

Eugene Fama’s efficient market research demonstrates that publicly available information gets reflected in stock prices without delay. Microsoft, Nvidia, Google, and Amazon aren’t hidden opportunities. They’re the most analysed, most discussed, most obviously positioned companies for the AI revolution on the planet.

The efficient market hypothesis doesn’t claim prices are always “correct” — they’re based on imperfect forecasts and they’re wrong constantly. But it does suggest we can’t systematically identify when they’re too high or too low. Every investor, analyst, and pension fund manager knows about AI. If it’s obvious to you, it’s already in the price.

 

The thematic investing trap

Bradford Cornell and Aswath Damodaran identified a recurring pattern across transformative technologies — railways in the 1840s, electronics in the 1960s, the internet in the 1990s.

The pattern is remarkably consistent: investors exaggerate the total addressable market, underestimate how many firms will fight for it, obsess over growth metrics rather than profitability requirements, and watch share prices divorce from earnings, revenue, and book values.

The critical insight: even genuinely revolutionary technologies often destroyed more investor wealth than they created during bubble periods. Railways transformed society. Most railway investors lost money. The internet transformed commerce. Most dot-com investors lost money.

 

The concentration disaster risk

The Sequoia Fund offers a sobering case study. Founded by protégés of Warren Buffett and built on his principles, with a stellar multi-decade track record, the fund allowed Valeant Pharmaceuticals to grow to over 30% of their portfolio.

When Valeant collapsed from $257 to $9 per share, it savaged the previously strong performance record. The fund trailed the S&P 500 by over three percentage points annually across the following decade.

That’s what concentration does. It creates uncontrollable disaster risk that can undo decades of endeavour. The risk is entirely unnecessary.

 

The brutal statistics

Research examining long-term market performance reveals an extreme fact: only two to four per cent of stocks drive all net wealth creation over time. The vast majority fail to create meaningful wealth over their lifetime.

Success requires capturing that tiny minority of long-term winners — impossible in a concentrated portfolio but probable in a broadly diversified one. As one researcher observed, it’s like roulette. Do you want chips across the table, or everything on red 17?

 

The behavioural gauntlet

Investors navigating the AI boom face a perfect storm of psychological biases. Fear of missing out is overwhelming when everyone discusses AI profits. Research shows social influence amplifies behavioural biases rather than reducing them.

Recency bias leads us to assume recent trends continue indefinitely. Analyst forecasts for AI-focused companies consistently prove too optimistic, with analysts constantly revising downward their earlier bullish projections.

Perhaps most troubling: the median individual investor spends just six minutes researching each trade, focusing primarily on price charts — the least predictive information available.

 

What evidence-based investing demands

The research points to clear principles:

Maintain broad global diversification. Own the whole market — all of capitalism, not just the exciting bits. That means globally diversified portfolios capturing US stocks, international developed markets, emerging markets, and appropriate bond allocations for your risk tolerance. You’ll own the AI winners naturally, at market-cap weights rather than emotional overweights. Crucially, you’ll own thousands of companies that benefit from AI indirectly, plus industries that thrive regardless of AI’s success.

Actively monitor your concentration risk. Check what percentage of your portfolio is actually exposed to AI mega-caps. If you own a global tracker, a US index fund, any growth funds, or any technology funds, add up your total exposure to Microsoft, Nvidia, Apple, Google, Meta, Amazon, and Tesla. You might discover you have 20%, 30%, or even 40% of your wealth tied to a handful of companies in a single sector, all riding the same AI narrative. This violates every diversification principle.

Resist specialised thematic products. The asset management industry has responded to the passive investing revolution by launching increasingly niche products. Research shows these specialised ETFs charge significantly higher fees while systematically underperforming. They’re designed to exploit your behavioural biases, not improve your outcomes.

Focus on the controllables. The evidence is overwhelming: we can’t reliably identify which sectors or stocks will outperform. But we can control costs (every pound in fees is permanently lost compounding potential), diversification (broad exposure guarantees you’ll own the winners), behaviour (staying invested through volatility), and tax efficiency.

 

The uncomfortable reality

Here’s the bottom line: the AI revolution might be entirely real. The technology might transform everything. AI might be as important as electricity or the internet.

And you still shouldn’t make a concentrated bet on it.

Why? Because if it’s obvious, it’s already priced in. You can’t identify which specific companies will win the competitive battles ahead. Most transformative technologies destroy more investor wealth than they create during bubble periods. Concentration creates uncontrollable disaster risk that can wipe out decades of careful wealth building.

Broad diversification guarantees you’ll capture whatever happens, whoever wins, whichever companies survive and thrive.

The paradox of evidence-based investing: it tells us to own everything precisely because we can’t predict which parts will deliver exceptional returns. Market efficiency doesn’t mean prices are always correct — they’re based on uncertain forecasts and they’re frequently wrong. But it means we can’t systematically identify when they’re too high or too low.

Whether or not there’s a bubble in AI stocks, your exposure should be roughly proportional to AI companies’ weight in global markets — no more, no less. If that feels disappointing or insufficient, that’s your emotions talking, not your evidence-based analysis.

Betting on red 17 might occasionally work. But spreading your chips across the table is the only rational long-term strategy. Even when red 17 looks like the surest thing in the world.

 

Why this demands professional guidance

The honest truth: this isn’t a simple question with a simple answer.

Getting it right requires assessing your total AI exposure across multiple funds, understanding how correlations behave during market stress, implementing rebalancing discipline when it feels most wrong, making tax-efficient adjustments, and maintaining diversification through volatility that will test your conviction.

This is precisely why evidence-based financial planning exists.

A financial planner with an evidence-based investment philosophy can audit your total portfolio for hidden concentration risks, help you understand how much AI exposure is appropriate for your specific circumstances, implement systematic rebalancing discipline, keep you from behavioural mistakes when markets turn, and ensure your portfolio serves your financial plan rather than your emotions.

The question of a bubble in AI stocks won’t be the last time markets test your discipline. Working with an adviser ensures you have a framework that survives whatever comes next.

 

If you want to make long-term financial decisions supported by independent empirical evidence, why not book an appointment with us?

 

Financial Planner in Norwich

rockwealth Norwich is an evidence-based and fee-only Financial Planning firm situated in the City of Norwich, Norfolk.

About: rockwealth Norwich IFA, is based in the centre of Norwich. As an independent financial planning adviser, we can help with many financial services, from independent financial advice, pension and retirement advice, investment advice and inheritance tax planning.

Interested to work with us?: We offer an Initial Discovery Consultation, completely free of charge and without any obligation. You can visit us at our office, schedule a video call, or call us on:  01603 542080.

Learn why Norwich IFA has never been busier >

Find rockwealth Norwich IFA at: rockwealth, St Georges Works, 51 Colegate, Norwich, Norfolk, NR3 1DD.