How Imminent is the AI Bubble Burst ? by Tulika Majumder
The AI investment boom of 2023 to 2026 shows classic bubble traits: record capital spending, extreme market concentration, and valuations that assume near perfect execution. History and current data still suggest a sudden economy wide burst is not imminent in late August 2026. A correction, a rotation across sectors, or a series of rolling bubbles is more likely over the next twelve to twenty four months. A larger unwind could come later if productivity and revenue fail to catch spending.
As of late August 2026, AI related stocks have already fallen about 20 percent from their June highs. Investors are asking harder questions about returns, interest rates, and cheaper competing models. The underlying buildout has not stopped. The live debate is no longer whether there is exuberance. It is how soon disappointment turns into a financing pullback.
What makes this look like a bubble is the scale of spending relative to proven commercial returns. The Bank for International Settlements, often called the central bank for central banks, compared the AI capital expenditure surge to nineteenth century canal and railway manias, 1920s electrification, and the 1990s dot com boom. Each episode began with a genuine technological advance that attracted more capital than commercial returns could ultimately justify. Each ended with an investment reversal and, in several cases, a broader recession.
The five largest hyperscalers Amazon, Microsoft, Google, Meta, and Oracle are on track to spend more than one trillion dollars on AI related capital expenditure across 2025 and 2026. That pace outstrips free cash flow and is pushing more debt issuance. Amazon has guided toward about 200 billion dollars, Microsoft about 190 billion, Google about 180 billion, and Meta up to 140 billion for 2026. Global AI investment is running around 850 billion dollars in 2026, hundreds of billions above the pre AI trend.
Veteran investor Jeremy Grantham has called this the biggest investment bubble in American history. He has advised investors to avoid U.S. stocks and warned that the most speculative AI names could fall 70 percent. He points to extreme valuations, including a Shiller CAPE near 40 and a modified Buffett Indicator around 235 percent, plus market concentration in which the top ten S&P 500 names account for roughly 40 percent of the index versus about 27 percent at the 2000 peak. He notes that bubbles often form around the most important technologies. Railroads transformed the world, yet their stocks collapsed after overinvestment.
OpenAI chief executive Sam Altman said in 2025 that investors as a whole were overexcited about a kernel of truth, comparing the moment to the late 1990s internet boom. He argued that when bubbles happen, smart people get overexcited, while still insisting the underlying technology is transformative. More recently he has acknowledged that economic adoption and disruption are arriving more slowly than he expected because of institutional inertia.
Other voices sound the same alarm. Wharton finance professor Itay Goldstein said there are many indications that we are in a bubble and that overpricing looks likely. Allianz chief investment officer Ludovic Subran pointed to SpaceX borrowing soon after its listing as evidence markets had entered bubble territory. Economist Steve Keen has argued that unsustainable data center spending and a mismatch between costs and serious user revenue could produce a burst within a year, more as a cash flow and supply chain crisis than a classic banking collapse.
A widely cited MIT Project NANDA study found that the large majority of custom enterprise generative AI pilots produced no measurable profit and loss impact, with only a small fraction reaching successful production deployment. The popular 95 percent failure headline has been debated and applies most sharply to bespoke enterprise systems rather than all AI use. Even so, it underscores a persistent gap between spending and proven returns at scale. Economy wide productivity data has not yet shown the dramatic acceleration many valuations assume.
A full burst may still not be imminent. A Barron’s analysis published in late August 2026 argued that capital spending bubbles almost always end badly but rarely when investors most expect them to. Using a historical rule of 25, the point at which transformative technology spending approaches 25 percent of economic output, the piece concluded AI is years away from that threshold. With U.S. GDP around 30 trillion dollars, the danger zone would require several more trillion dollars of domestic outlays. Current trajectories point toward the early 2030s. Melius Research’s Ben Reitzes said this will be bigger than anybody thinks, for longer than anybody thinks.
Macquarie analyst Viktor Shvets described the cycle as a sequence of rolling bubbles rather than one crash: first large language models, then facilitators such as chips, power, and data centers, then applications. As one segment cools, another can pick up the slack. Large contract backlogs and physical orders for chips, memory, and construction distinguish this from pure vaporware speculation.
Goldman Sachs Research has noted important differences from the late 1990s tech boom. The current investment wave is narrower, the largest companies remain highly profitable, and the path of earnings looks different. That does not eliminate risk. An earnings bubble could still form if profit expectations prove too optimistic. It does reduce the chance of an immediate 2000 style wipeout. The hyperscalers generating the bulk of spending have fortress balance sheets and existing cash cow businesses, unlike many dot com era startups.
Jeff Bezos has framed the episode as a good bubble that finances useful infrastructure even if some investors lose money. The physical assets data centers, power plants, and specialized chips will remain after any valuation reset and can support later waves of application layer growth.
A sudden synchronized collapse that tanks the global economy in 2026 is the lower probability scenario. More plausible is a multi year deflation or rotation. AI infrastructure stocks give way to software and applications. Weaker players and highly leveraged private credit vehicles face stress. Capital expenditure growth slows as companies demand clearer return on investment. Energy and chip bottlenecks ease or intensify depending on policy and supply.
The BIS warning is not that collapse is certain tomorrow. It is that disappointment on returns could trigger a financing pullback and turn the boom into a protracted investment bust with knock on effects on financial conditions. Because much of the funding now flows through less regulated private credit and circular equity ties, any unwind could be sharper than a traditional bank led crisis. Supply side constraints electricity, grid connections, and advanced chips add another layer of vulnerability and inflationary pressure.
History offers a mixed lesson. Railroads, electrification, and the internet all produced lasting infrastructure and productivity after painful busts. The companies that survived, and the utilities built around the new technology, created enormous long term value. The same pattern is possible here. Overinvestment in compute and data centers today could enable cheaper, more capable AI tomorrow, even if many current equity holders are disappointed.
The AI bubble is real in the sense that capital has raced far ahead of demonstrated commercial returns and economy wide productivity gains. Expert consensus from the BIS, Grantham, and others is that risks of a meaningful correction are elevated and rising. The most careful historical and spending analyses still suggest the music has not stopped. A 2026 to 2027 period of volatility, sector rotation, and tighter scrutiny of return on investment is more probable than an imminent 70 percent market wipeout.
The technology itself is not going away. What is in question is the timing and magnitude of the financial adjustment required to bring valuations and spending into line with what businesses and consumers will actually pay for. Policymakers and investors should treat the BIS urgency message seriously. The longer the mismatch persists and the more leverage builds, the more painful the eventual reckoning will be. For now the bubble has further to run, but the air is already leaking.