The AI Sell-Off: Is the Market Finally Distinguishing Winners From Losers?

Since ChatGPT's launch in November 2022, artificial intelligence has been a major force behind the stock market's advance. The so-called Magnificent Seven, although not exclusively AI companies, accounted for 63% of the S&P 500's gains in 2023 and 55% in 2024. The enthusiasm continued into 2026, but began to fade over the summer. By the end July, semiconductor stocks had fallen more than 20% from their mid-June highs. In September, concerns about AI financing spread into the credit markets. October brought particularly severe losses among specialized cloud providers, followed by a broader technology sell-off in early November.

On November 5, the Nasdaq fell about 12% below its October 6 record, entering correction territory. Concerns over AI spending, elevated borrowing costs and potential restrictions on data-center construction contributed to the decline.

01 There Is No Single AI Trade

The companies associated with artificial intelligence have very different business models.

• Chipmakers such as NVIDIA and AMD supply the processors needed to power AI.

• Hyperscalers, including Microsoft, Amazon, Alphabet and Meta, invest heavily in computing infrastructure while operating established, profitable businesses.

• Neoclouds, such as CoreWeave and Nebius, specialize in renting computing capacity, frequently financed through substantial borrowing.

• AI model developers such as Anthropic and OpenAI build the underlying models, while companies across virtually every industry are beginning to use AI to improve productivity and develop new products.

The distinction matters. A dollar spent on AI infrastructure generates revenue for a chipmaker but represents an investment that someone else must eventually recover. Ultimately, the question is not simply whether AI demand will grow, but which companies can convert that demand into attractive returns on capital.

02 October Exposed the Weakness of the Neocloud Model

The October correction was particularly severe among neoclouds. According to Dealroom, CoreWeave declined 37%, Nebius 35% and IREN 41%. By comparison, the Nasdaq fell approximately 5%. The problem is not necessarily insufficient demand. CoreWeave, for example, reported approximately $104 billion in contracted revenue backlog. Rather, the concern is how profitably that demand can be served.

Neoclouds must invest enormous sums upfront in GPUs (Graphic Processing Units, the processors empowering AI) and data centers, often borrowing heavily, while customers pay over several years. Consider a company borrowing $1 billion to purchase GPUs expected to generate rental income over five years. If newer, more efficient chips reduce rental prices after two years, the company could face declining revenues while its debt obligations remain unchanged. Neouclouds also have shorter cycle usage-based revenue, while they bear high long-dated fixed costs.

Delays in construction, higher interest rates and customer concentration compound the risks. These concerns became very visible as certain CoreWeave bonds yielded approximately 16%. A large backlog may provide revenue visibility, but it does not guarantee profitability or sufficient cash flow to repay debt.

03 Are Hyperscalers Being Unfairly Punished?

In his November 2 shareholder letter, famed investor Bill Ackman of Pershing Square offered a compelling counterargument. He noted that the four leading hyperscalers' combined annual capital expenditures have risen from approximately $140 billion in 2023 to an estimated $740 billion in 2026, with more than $1 trillion expected in 2027.

Yet through September, their shares had risen only 8% on average in 2026, despite approximately 27% earnings-per-share growth. Ackman argues that investors are focusing on the spending while underestimating its potential returns. His analysis suggests that new AI infrastructure could generate pretax returns on invested capital of 15% to 25%.

These are estimates, not demonstrated results. Actual returns will depend on customer demand, equipment utilization, pricing and replacement costs. Nevertheless, hyperscalers have important advantages: enormous existing recurring cash flows, diversified businesses, established customer relationships and balance sheets capable of supporting substantial investment.

Their valuations are not necessarily cheap in absolute terms. But unlike highly leveraged infrastructure providers, they can finance much of their growth and benefit directly from using AI in their existing businesses.

04 Anthropic's IPO Provides Another Test

Anthropic's anticipated November IPO illustrates the difficulty of valuing extraordinary technological growth. Its third-quarter revenue approached $17 billion, up 47% sequentially, but the company reported a $567 million operating loss. Its recently disclosed computing and infrastructure commitments total approximately $895 billion over multiple years, with about $70 billion due within 12 months.

NYU professor Aswath Damodaran estimates Anthropic's equity value at approximately $1.45 trillion. His valuation assumes annual revenue eventually reaching $1 trillion by 2036.Anthropic may become an outstanding business. Whether it becomes an outstanding investment depends on the price investors pay and the profits it eventually delivers.

05 What History Suggests About the Correction

The recent selling has been unusually severe. Bespoke Investments Group's November 5 analysis found that the Nasdaq 100 ETF (QQQ) was more than three standard deviations below its 50-day moving average, an extreme historically exceeded on only 0.2% of trading days since 1999. In 16 previous comparable instances, QQQ was higher one year later 14 times.

When both QQQ and the semiconductor ETF SMH reached similarly oversold conditions, subsequent QQQ returns averaged 13.4% over three months and 25.1% over six months. These figures are encouraging, but the simultaneous signal occurred only four times (such a small sample cannot reliably predict future returns).

06 Our Perspective

We do not view the correction as evidence that AI's potential has diminished. Rather, investors are questioning whether that potential will translate into satisfactory financial returns.

Some companies may deserve substantially lower valuations because their financing risks or business economics have deteriorated. Others may have become more attractive as share prices declined while their long-term earnings prospects remained intact. History suggests that technological advances create enormous economic value, but not always for the investors who finance their development.

The opportunity today is not simply to buy the AI sell-off. It is to identify businesses capable of earning sustainable returns on capital, with strong competitive positions, manageable financial obligations and valuations that leave room for disappointment.

THIS BLOG POST HAS BEEN WRITTEN BY OUR TEAM, IT HAS BEEN EDITED AND ENHANCED BY AI. ALL FACTS AND FIGURES HAVE BEEN VERIFIED BY OUR TEAM. THE VISUALIZATIONS HAVE BEEN CREATED BY AI AND VERIFIED BY OUR TEAM.