AI Market Correction: Bubble Burst or Healthy Reset?

AI-related equities became the dominant driver of index performance in the period following ChatGPT's launch in late 2022. According to JP Morgan Asset Management, AI stocks accounted for roughly 75% of S&P 500 returns and approximately 80% of earnings growth over that stretch. That concentration was extraordinary by any historical measure.

Valuations followed the narrative upward. By late 2025, the S&P 500 was trading at around 23 times forward earnings, while the Case-Shiller cyclically adjusted price-to-earnings ratio exceeded 40 a level not seen since the peak of the dot-com era, according to analysis cited by multiple financial institutions including Deutsche Bank and Citigroup. The top ten U.S. stocks alone represented roughly 40% of S&P 500 market capitalisation and about 25% of global equity value, per Goldman Sachs Asset Management's 2026 investment outlook.

What changed next was a cascade of earnings disappointments. Oracle's December 2025 results, which missed revenue forecasts and projected a 40% increase in AI-related capital expenditure to $50 billion for fiscal 2026, shook investor confidence. Highly speculative AI-focused companies saw valuations fall 30–50% or more in the final quarter of 2025. Then, in late January 2026, a single-day selloff wiped an estimated $440 billion from Microsoft's market value alone.

The clearest warning light is not a single scandal or a sudden reversal in AI capability. It is market structure combined with a widening gap between investment and demonstrable return. An August 2025 report by MIT's Nanda research group found that despite $30–40 billion in enterprise investment into generative AI, 95% of organisations reported zero measurable return. A February 2026 study from the National Bureau of Economic Research reinforced that finding: 90% of firms reported no AI impact on workplace productivity, even as executives projected future productivity gains a pattern that echoes what economists call the productivity paradox.

The capital expenditure picture adds another dimension. Citigroup estimated that hyperscaler AI capex could reach approximately $490 billion by 2026, and Morgan Stanley projected global data centre spending of around $3 trillion between 2025 and 2028, roughly half of it funded by private credit. OpenAI alone projected operating losses of $74 billion in 2028, with revenues far short of its committed infrastructure spending. Concerns about circular financing — where major AI players hold equity stakes in each other, creating interlocking valuations were raised by multiple analysts, including those at Yale School of Management.

Here is the kicker: AI-related investment accounted for an estimated half of U.S. GDP growth during the first half of 2025. A reversal of that spending would not merely weigh on stock prices. It would risk dragging broader economic growth with it.

The distinction between a structural bust and a cyclical correction matters enormously for how investors should respond. A structural bust implies that the underlying technology overpromised and that the capital deployed will not generate acceptable returns the dot-com parallel that many analysts have drawn. A cyclical correction implies that spending is front-loaded relative to revenue capture, creating a temporary valuation gap that resolves as products mature and monetisation improves.

The weight of current evidence leans toward cyclical, with structural tail risks that remain real. AI adoption at the enterprise level has grown sharply Stanford's 2025 AI Index noted a significant jump in organisational AI use in 2024. The technology's usefulness is not in dispute; the dispute is over timing and who captures the value. As Blue Whale Growth Fund's chief investment officer Stephen Yiu argued in December 2025, the market had broadly failed to distinguish between companies with strong free cash flows and those burning capital to build infrastructure without a clear path to profitability.

Why this matters: the Jevons effect offers a plausible scenario where efficiency improvements in AI models actually expand, rather than contract, total compute demand provided the models deliver on measurable productivity. If they do not, that compounding loop breaks, and the infrastructure buildout becomes a stranded-asset problem. That is the structural scenario that investors should keep in their scenario set, not their base case.

The following comparison distils the key indicators across the two scenarios.

The scorecard does not produce a clean verdict. Individual AI infrastructure companies and loss-making startups show classic bubble anatomy circular financing, profitless growth, and valuations built on optionality rather than earnings. The broader index, however, has absorbed the AI-sector volatility with relative composure, which is not a feature of a systemic bust.

2026 is shaping up as the year when investors stop paying for AI's potential and start demanding proof of AI's profitability. The market is beginning to fragment between what CNBC's reporting on Blue Whale's analysis called "AI spenders" and "AI monetisers" those writing the capex cheques and those actually generating recurring revenue from AI-powered products. That differentiation, overdue as it is, is what a healthy correction looks like.

For investors, the practical implication is not to exit AI entirely but to scrutinise cash flow yield, debt structure, and the revenue visibility behind any AI-linked position. Deutsche Bank's analysis noted that across dozens of global indices, markets entering a year at high valuations tended to underperform those starting from cheaper levels. That pattern does not guarantee a crash, but it does imply that the bar for multiple expansion is higher from here.

The DeepSeek shock in January 2025 when Nvidia's shares fell 17% in a single session after a Chinese startup released a competitive model reportedly trained at a fraction of U.S. costs offered an early glimpse of how quickly AI competitive assumptions can be disrupted. Whether that represents a one-off or a preview of sustained margin compression across the sector is the central unresolved question of the current correction. The answer will determine whether history looks back on 2025–2026 as the moment the AI market matured, or the moment it peaked.