AI Bubble’s 2026 Burst Could Spark a U.S. Downturn

Economist Ruchir Sharma has argued that the current artificial intelligence boom fits his four-part bubble test of overinvestment, overvaluation, over-ownership and rising leverage, based on recent data on capital spending and markets reported by Business Insider.

At the same time, International Monetary Fund chief economist Pierre-Olivier Gourinchas has compared the AI surge with the late-1990s internet stock boom. He stressed that AI-linked investment has so far added less than 0.4 percent of United States GDP since 2022, compared with about 1.2 percent for dot-com investment in the late 1990s, according to an interview with Al Jazeera.

Sharma estimates that AI-related capital expenditure and the stock-market wealth effect now account for roughly 40 to 60 percent of 2025 U.S. growth. This means a reversal could have macroeconomic consequences that reach well beyond the technology sector, as described in his remarks reported by Times of India.

Bubble Signals Intensify

In those comments, Sharma describes a concentration of spending and market gains around a small group of large technology firms. He notes that technology investment has reached about 5 percent of U.S. GDP, a level similar to 2000. Furthermore, by some measures AI-related stocks generated about 80 percent of this year's equity-market gains, while helping drive 40 to 60 percent of economic growth.

A separate analysis of global markets finds that total technology investment now exceeds 6 percent of U.S. GDP and has surpassed the record set at the height of the dot-com bubble. U.S. households hold roughly 52 percent of their financial wealth in equities, according to an article by Sharma in the Financial Times that was republished by the Financial Post in 2025.

The same analysis reports that AI-related spending by the so-called Magnificent Seven technology firms more than doubled between 2023 and 2025 to about 380 billion dollars. This underscores how much of the current investment cycle depends on a narrow group of corporate balance sheets.

Sharma has warned that the United States economy has effectively become a single large bet on AI, with relatively weak growth outside that segment. This leaves little room for disappointment if adoption or productivity gains fall short.

He has also argued that past speculative cycles rarely ended because of the technology itself. Instead, they tended to break after borrowing costs rose, saying that higher interest rates and tighter financial conditions have been the most consistent historical triggers for crashes across previous bubbles.

"Every single bubble or mania in history has been pricked by just one factor - when interest rates finally go up."

– Ruchir Sharma

Lessons from the Dot-Com Crash

The late-1990s dot-com episode remains the closest historical parallel for an AI-driven investment cycle. Between 1995 and its peak in March 2000, the technology-heavy Nasdaq Composite rose by about 600 percent before falling roughly 78 percent by October 2002, according to historical data summarized by Investopedia.

That collapse erased around 5 trillion dollars in stock-market value and was followed by a shallow U.S. recession in 2001. IMF staff now cite this sequence when discussing the risks from an eventual AI correction, as reported by Al Jazeera.

Analyses of the dot-com bust describe how overbuilding in telecommunications and hardware left some networks and data infrastructure underused for years, even as they later supported the wider internet economy. Investors and workers in those sectors experienced a long period of adjustment while asset prices, payrolls and business models reset.

This pattern illustrates a recurring tension in technology-led booms. Investment surges can generate both misallocated capital and durable infrastructure. The same question now applies to AI data centers, networks and specialized chips that may outlast an initial speculative phase.

Macroeconomic Shock Scenarios

Because recent U.S. growth appears unusually dependent on AI-linked capital spending and market gains, a sharp slowdown in that spending could have outsize macroeconomic effects. Sharma has estimated that AI-related investment alone now contributes roughly 40 percent of U.S. growth, and that including the wealth effect from higher technology valuations lifts the share closer to 60 percent.

If that contribution fell back toward zero in a short period, overall growth would slow even if other sectors did not deteriorate. Given that non-tech investment has already weakened in recent years, as noted in IMF discussions of the U.S. outlook, the risk is that an AI reversal would coincide with soft underlying demand rather than offsetting it.

Research firm BCA has compared the current AI cycle with earlier booms in railways, electrification, oil and the internet. They argue that those episodes typically involved slower-than-expected adoption, overestimated revenues and heavy debt before investment collapsed and hurt the wider economy. In commentary relayed by market news account Walter Bloomberg on X, BCA said the AI cycle may be near its top and could fade within 6 to 12 months if expectations are not met.

In such a scenario, layoffs at AI-focused firms and associated service providers would likely coincide with reduced construction, equipment orders and stock-market wealth. This would weigh on consumer spending in regions where technology employment and data center projects are concentrated.

Sharma has also highlighted another channel of risk. Some techno-optimist forecasts assume that AI systems will replace up to 40 percent of existing work tasks and could push unemployment rates as high as 20 percent. He links this to the possibility of political backlash if it materialises, according to his Financial Times commentary republished by the Financial Post.

Collateral Damage in Supply Chains

AI infrastructure spending has already transformed capital flows into data centers and electricity networks. Market-analysis firm IoT Analytics estimates that global data center equipment and infrastructure spending reached about 290 billion dollars in 2024 and could approach 1 trillion dollars by 2030, driven largely by hyperscaler capital expenditure, according to its 2025 report on the data center infrastructure market posted by IoT Analytics.

Within that total, Alphabet, Microsoft, Amazon and Meta invested nearly 200 billion dollars in capital expenditure in 2024, with growth expected to exceed 40 percent in 2025 as they race to add AI-ready capacity. This surge supports a complex chain of suppliers, from server manufacturers and networking-equipment vendors to construction firms and power-equipment providers.

A separate construction-industry analysis projects that U.S. data center construction spending could peak at about 89 billion dollars in 2026, according to the 2025 data center report published by MOCA Systems. That level of activity affects not only specialized builders but also local real estate markets, permitting processes and grid-connection projects in host communities.

If AI-related capital budgets were reduced sharply, many of these projects could be postponed or cancelled. Contractors that had expanded to meet hyperscaler demand might face periods of underutilized equipment and staff. Local governments that anticipated new property-tax and utility revenues could see those expectations revised.

On the hardware side, IoT Analytics expects the data center infrastructure market to grow rapidly through the end of the decade, with GPU-based servers and high-density cooling systems taking a rising share of spending. A reversal in AI workloads would not eliminate this demand but could leave chipmakers and component suppliers with more capacity than is immediately needed, forcing price adjustments and cutbacks in new investment.

Utilities that have upgraded transmission lines and generation assets to accommodate power-hungry AI clusters would also be exposed. If planned data centers do not materialize or operate below capacity, some of that infrastructure could remain underused for several years, even if longer-term electrification trends eventually absorb it.

What Could Cushion the Fall?

Despite the bubble risks, both Gourinchas and Sharma have emphasized that an AI correction would differ from 2000 in important ways. The IMF chief economist has noted that most current AI investment is funded from the cash flows of large technology companies rather than by highly leveraged borrowers. He also pointed out that AI-related capital formation has so far added less than 0.4 percent of GDP, compared with 1.2 percent in the dot-com years, according to his remarks reported by Al Jazeera.

This smaller share suggests that even a sizeable markdown in AI-related equity values might have a more limited direct impact on banks and the broader financial system than the housing bust of 2008 or other credit-driven crises. Losses would fall mainly on shareholders and on companies that had committed to aggressive expansion plans.

At the same time, history indicates that the physical and organizational assets built during speculative phases can continue to support growth after valuations reset. In the dot-com case, fiber networks, server farms and software platforms laid the groundwork for e-commerce, social media and cloud computing, even though many of the original firms behind those investments failed.

Something similar could occur with AI infrastructure. High-density data centers, advanced chips and trained technical staff could be repurposed for more targeted or lower-cost applications once capital becomes scarcer. This could potentially improve productivity in sectors that did not participate in the initial boom.

However, this adjustment process would take time and would not erase the near-term disruption to workers, suppliers and local economies tied closely to AI projects. The main uncertainty is how far interest rates will rise from current levels and how sensitive AI-driven equity valuations and capital spending are to those changes.

Closing Outlook

"This big bet on AI better work out for America - because if it does not work out, then I think there is a lot of trouble for this country ahead."

– Ruchir Sharma

The same AI investments that currently support a large share of U.S. growth could become a source of drag if confidence in AI-driven returns weakens and the cost of capital rises. A reversal on that scale would not necessarily replicate the dot-com crash, but the combination of concentrated equity ownership, heavy capital spending and high expectations creates conditions for a pronounced adjustment.

Whether that adjustment takes the form of a brief reset or a more extended downturn will depend on the path of interest rates, the pace at which AI applications generate measurable productivity gains and how quickly surplus servers, chips and grid capacity can be redeployed. For now, both investors and policymakers are focused on a familiar variable from earlier cycles: the price of borrowing.

Sources

• Sharma, R. "AI Boom Has 4 Bubble Signs Could Burst 2026 Economist." Business Insider, 2025.
• Sharma, R. "AI Boom or Bubble Is the AI Bet Driving US Growth Into Risky Territory." Times of India, 2025.
• Sharma, R. "Four Os Shape Bubble." Financial Post, 2025.
• Gourinchas, P.-O. "IMF Says AI Investment Bubble Could Burst Comparable to Dot-Com Bubble." Al Jazeera, 2025.
• Investopedia Staff. "Dot-Com Bubble Definition." Investopedia, 2025.
• MOCA Systems. "Data Center Infrastructure Market Report." MOCA Systems, 2025.
• IoT Analytics. "Data Center Infrastructure Market Analysis." IoT Analytics, 2025.
• BCA Research. "AI Cycle Analysis via Walter Bloomberg." X Platform, 2025.