Three AI bubbles are testing Wall Street’s faith in the boom
From Anthropic’s anticipated IPO to the semiconductor sell-off and mounting pressure on neocloud companies, investors are reassessing the cost of the AI revolution. Strong earnings could revive the rally, but the market increasingly needs proof that massive investments can generate sustainable returns.
This correction should have come sooner. For weeks, U.S. Treasury bonds have been trading at yields above 5%, yet the stock market largely shrugged off the warning signs. That changed when higher-than-expected U.S. inflation reinforced concerns about interest rates and helped drain momentum from Wall Street. By the end of a turbulent week, the Nasdaq had plunged 4.5%, wiping out $2.2 trillion in market value, while the S&P 500 fell 2%. Only the Dow Jones Industrial Average managed to finish the week in positive territory, as investors began looking beyond AI-related stocks. The market’s “fear index” spiked, and the usual doomsayers were quick to ask whether October could bring another “Black Monday.” But with the final quarter of the year underway, could the correction instead offer a rare opportunity to buy ahead of a traditional year-end rally?
The reasons for the market’s loss of momentum, particularly in AI infrastructure, have been apparent for some time. Until now, however, Wall Street has tended to pause briefly before resuming its climb. Is this time different?
It depends on whom you ask. One notable change is that hedge funds have been selling stocks aggressively over the past week, a shift that could intensify the downturn if it continues. Their influence over market movements makes their positioning worth watching.
Research firm Gartner has also warned that the economics of the AI boom are increasingly difficult to reconcile. The investments already made in infrastructure, together with those still planned, appear out of proportion to projected revenues. For the overall investment to pay off, revenue forecasts would need to more than double. If the numbers fail to add up, the companies making those investments could suffer, and their investors would feel the consequences.
Others remain more optimistic. Many economists at major banks argue that the market is not in a bubble and that stock valuations are not particularly stretched. Dan Simkowitz, co-president of Morgan Stanley, told Calcalist last week that valuation multiples had not risen since the beginning of the year because companies continued to improve their profitability.
Michael Hartnett, chief investment strategist at Bank of America, has similarly argued that the bank’s “sell signal,” issued in May, has expired. He expects bond yields to fall below 5% again, even as the outlook for Federal Reserve policy remains a key factor for investors.
To understand what is happening in the markets, and why the debate over an “AI bubble” has intensified, it helps to separate the market into three distinct areas of risk.
1. The Anthropic bubble
One of the biggest factors shaping sentiment is the anticipated public offerings of Anthropic and OpenAI, two AI companies whose debuts investors have been eagerly awaiting. Both may have preferred to go public earlier, but SpaceX’s plans crowded the market for major offerings, encouraging them to wait in the hope that they would be closer to profitability by the time they listed.
Investors have been setting aside capital to participate in what could be historic IPOs. But expectations have begun to shift. OpenAI CEO Sam Altman has said the company will not pursue an IPO in 2026, leaving Anthropic in the spotlight.
Anthropic has been preparing investors for a potential public offering, but questions have emerged about its growth and the quality of its reported profitability. The company reported third-quarter revenue of nearly $17 billion, although its growth rate has slowed. Its claims of profitability also rely on adjusted figures that exclude certain costs and other items.
Against this backdrop, some investors have begun discussing a valuation of around $1.5 trillion, below the roughly $2 trillion previously anticipated. Even at the lower figure, Anthropic would rank among the largest publicly traded companies by market value.
The stakes extend beyond Anthropic itself. A public offering could provide a real-world test of how much investors are willing to pay for the AI business model, including its enormous infrastructure costs and uncertain long-term margins. If demand for the shares falls short of expectations, the impact could spread to other AI companies and the wider technology market.
The prospectus and subsequent investor reaction will therefore be important signals. A disappointing reception would not necessarily mean that AI demand is collapsing, but it could force investors to reassess the valuations assigned to the companies expected to lead the next phase of the technology boom.
2. The chip bubble
Semiconductor stocks staged a meteoric rally between March and June this year. The Philadelphia Semiconductor Index, known as the SOX, doubled in value within a few months, and the second quarter of 2026 marked its strongest quarterly performance on record. Now the index has retreated 23% from its peak.
The decline is painful for investors who bought near the top. Yet even another drop of a similar magnitude would leave the index roughly around its level at the beginning of the year, illustrating how far and how quickly the sector had climbed.
The concerns surrounding chips are rooted in a familiar feature of the industry: its extreme cyclicality. Periods of shortages can give way to oversupply, causing prices, earnings expectations and share valuations to swing sharply. The current upcycle has lasted long enough to make investors increasingly nervous, reviving memories of the dot-com crash in 2000.
Still, a major collapse is not inevitable. Bank of America has argued that leading AI-chip companies such as Nvidia, Micron and Broadcom are trading at relatively modest forward price-to-earnings multiples. Comparisons with Cisco, a symbol of the dot-com bubble, are instructive, although the companies and their market conditions are not identical.
The key distinction between the current boom and the dot-com era is the extent to which the infrastructure is being used. During the earlier cycle, a massive wave of spending on telecommunications equipment left much of the installed capacity underused. Today, much of the existing AI infrastructure is operating at high utilization, reflecting real demand for computing power.
That does not guarantee that all planned investment will generate adequate returns. It does, however, shift the question from whether the infrastructure is needed to whether the companies building and financing it can earn enough to justify the cost.
That brings us to the third potential bubble: the so-called neocloud companies.
3. The neocloud bubble
One of the central investment theses of the past two years has been that the main constraint on the AI revolution is supply rather than demand. On that assumption, investors and lenders poured money into new data centers, power infrastructure and companies designed to rent computing capacity to customers that could not afford to build their own facilities.
These neocloud providers effectively aggregate access to graphics processing units, or GPUs, and other AI infrastructure for smaller players. Leading companies in the segment include Nebius, which has significant operations in Israel and has made several local acquisitions, and U.S.-based CoreWeave.
To expand rapidly, CoreWeave has relied heavily on external financing. Its debt has grown to roughly $50 billion, and last week the yield on some of its bonds surged to around 15%. Such yields signal that investors are demanding much higher compensation for the risk of lending to the company.
In the first half of the year, CoreWeave recorded $4.6 billion in revenue and a net loss of $1.4 billion. Those figures highlight the challenge of turning rapid expansion into sustainable profits and cash flow, particularly as borrowing costs rise and the company must continue funding its infrastructure buildout.
Nvidia is both a major supplier of GPUs to CoreWeave and one of its financial backers, creating a close relationship between the chipmaker and the infrastructure provider. After CoreWeave’s shares fell sharply over the weekend, Nvidia CEO Jensen Huang defended the company, arguing that its position was strengthening and that its cost of capital was declining.
The pressure is not limited to CoreWeave. Microsoft has withdrawn from a planned data-center agreement with Nscale for a site in Essex, England, after delays and difficulties securing sufficient power. Nscale has said its relationship with Microsoft remains strong and that it is exploring alternative power sources for the project. Nevertheless, the development highlights a central risk facing neocloud providers: even when demand for computing capacity exists, delays in construction, grid connections and financing can undermine the economics of individual projects.
The sector’s vulnerability is therefore not simply a question of whether AI will be used. It is whether the companies building the infrastructure can bring capacity online on time, secure customers on viable terms and generate enough cash to service their debt.
So, who is right?
Market volatility could persist in the coming weeks, with U.S. midterm elections and a busy earnings season adding to uncertainty. Major technology companies are due to report results, beginning with Meta and Tesla. Strong earnings, and especially upbeat forecasts, could help ease the pressure. Weak guidance or signs that AI spending is failing to produce adequate returns could have the opposite effect.
According to the latest projections from Harel Finance, four companies, Micron, Nvidia, Meta and Broadcom, are expected to account for 37.5% of projected earnings growth among S&P 500 companies. That concentration underscores how heavily the broader market’s outlook depends on a relatively small group of companies tied to AI and its infrastructure.
Geopolitical risks are another source of uncertainty, including the possibility of renewed escalation in the Middle East following comments by former U.S. President Donald Trump about Iran. Such risks can further unsettle markets already sensitive to inflation, interest rates and corporate guidance.
Even so, a correction does not necessarily spell the end of the bull market. If corporate earnings remain resilient and the AI investments underway translate into sustainable revenue and profits, investors may once again look toward a year-end rally. But the burden of proof is shifting. Companies can no longer rely on AI enthusiasm alone; they will need to show that the enormous sums being spent can generate returns.
The AI revolution is likely to have far-reaching consequences, although its full impact remains difficult to predict. In 2000, few could fully imagine how the internet would transform daily life, work and business models. AI could prove similarly transformative. The challenge for investors is to distinguish that long-term potential from the valuations, financing structures and expectations that may have run too far ahead of reality.
