Citadel Leaders Split on AI Bubble Risks and Stock Outlook

Conflicting assessments of the artificial intelligence investment boom are emerging as Citadel Securities strategist Scott Rubner anticipates a recovery in AI stocks after the US midterm elections, while Citadel CEO Ken Griffin warns that the market may already be in a bubble. In parallel, CITIC Securities maintains a positive long-term outlook for Chinese and US AI models, highlighting commercial adoption and valuation potential.

Rubner, head of equity and derivatives strategy at Citadel Securities, outlined his outlook on Bloomberg Television on October 20, 2026. He said institutional clients were delaying purchases until election uncertainty subsides, while retail investors were already increasing their exposure to US equities, particularly semiconductor stocks.

According to Rubner, individual investors on Citadel's platform recorded $9.1 billion in net purchases of US equities during the latest reported week, the largest such weekly total in 15 weeks. Semiconductor stocks accounted for $2.6 billion of those purchases, highlighting the concentration of retail demand in AI-related businesses.

Rubner described the midterm elections as a "clearing event for risk." He attributed the recent underperformance of AI stocks to broader market instability rather than a breakdown in the underlying investment case.

He also identified corporate share buybacks and third-quarter earnings announcements as potential sources of additional demand beginning in early November. Historical data cited in the report show that the S&P 500 has delivered average fourth-quarter gains of approximately 5.5% to 5.6% during midterm election years since 1930.

Griffin offered a more cautious assessment at the Future of Global Markets conference in New York. In an interview with Bloomberg, he questioned whether companies were generating sufficient economic benefits to justify their growing expenditures on artificial intelligence systems.

"We are almost certainly in the middle of an AI bubble," said Griffin, CEO of Citadel.

Griffin pointed to substantial differences in companies' ability to translate AI investments into measurable productivity improvements. He also highlighted the challenge businesses face in evaluating the costs of models developed by leading AI companies, including Anthropic and OpenAI.

"Every company is holding the Anthropic and OpenAI bills and saying, 'We have to figure out how much productivity we can wring out of this bill,'" Griffin said.

The Wall Street Journal reported on October 20 that Anthropic could spend $750 billion over the coming years to secure computing capacity. Griffin's assessment of the company's prospects hinges on whether AI models become essential tools across industries.

Despite his concerns, Griffin acknowledged that speculative excess can accompany major technological shifts. He predicted that AI would generate new US businesses, with the strongest survivors potentially maintaining dominant positions long after a market correction.

"Radical economic transformation usually goes hand in hand with a bubble," Griffin said.

"If you want to use the word bubble, those bubbles often coincide with enormous technological transformation," Griffin added.

Drawing a comparison with the dot-com era, Griffin emphasized the potential longevity of successful companies that emerge from a speculative boom. He referred to Amazon as an example of a business investors could have benefited from holding through that period.

"The companies that survive on the other side of the bubble go on to dominate the market for decades after it bursts," Griffin said.

"Back then, you should have held on to your Amazon shares rather than selling," Griffin added.

Meanwhile, CITIC Securities argued that the long-term commercial potential of large AI models from China and the United States remains substantial. The brokerage cited accelerating model development and a revenue-subscription-infrastructure cycle, alongside continued growth in annual recurring revenue at OpenAI and Anthropic.

CITIC Securities compared the gap between AI usage and monetization with the early commercialization phase of the mobile internet. It argued that broader adoption across companies, applications and geographic markets could create additional revenue opportunities.

For China, the brokerage said rising demand for data labeling was helping ease constraints on the data required to improve models after initial training. It also identified the fourth quarter of 2026 as a key period for assessing large-scale corporate adoption of overseas AI models.

The brokerage noted that Anthropic's quarter-on-quarter annual recurring revenue growth had slowed somewhat amid competition following the August-September launch of GPT-6-Astra and reverse-distillation factors. However, it expected growth to recover after October, citing the late-September release of Opus5.5 and expectations surrounding next-generation models such as Fable 5.5 ahead of Anthropic's initial public offering.

CITIC Securities said Chinese AI models' annual recurring revenue remained comparable to levels overseas models had reached approximately 18 months earlier, while their valuations stood at about one-tenth those of overseas counterparts. The brokerage said improvements in model performance, combined with low revenue bases and rapid growth, could support a broader valuation reassessment.