Ken Griffin, founder of Citadel, said investors are “almost certainly” in an artificial-intelligence bubble, while arguing that the companies and technologies that survive such a period can shape markets for decades. His comments add a note of caution to an investment boom that has made AI infrastructure a major force across technology and equity markets.
Speaking at Citadel’s Future of Global Markets Conference in New York on October 20, Griffin stopped short of calling for investors to abandon AI-related shares. According to Decrypt, he said major economic transformations are often accompanied by bubbles, pointing to the internet boom as an earlier example. He also said he regretted not holding on to his Amazon stock.
Why Griffin sees a bubble
A market bubble generally describes a period when asset prices rise much faster than the underlying businesses appear able to justify, creating the risk of a sharp reassessment when expectations meet reality. Griffin said the existence of a bubble was not necessarily the most important issue when viewed against the potential scale of the technological change.
“What’s left in the wake of these bubbles, if we want to use that word, is often game changing for the world,” he said. In his view, businesses that remain standing after a bubble can become market leaders for decades. He did not predict that the AI market would collapse.
Griffin offered Citadel’s own experience as an example of the technology’s potential. He said AI agents—software designed to complete multistep tasks independently—had reduced the time needed for some high-level research at the firm from several weeks to a few hours.
The stakes extend beyond specialist technology investors. AI investment has helped support equity markets over the past year, meaning index funds and retirement accounts may also have exposure to companies benefiting from the buildout. Citadel itself is a hedge fund, a private investment firm serving wealthy clients and institutions.
GPUs face a supply-and-demand test
Griffin compared graphics processing units, or GPUs, with airline seats. GPUs are widely used to train and operate AI models, while airline seats are costly to provide initially but have a relatively small additional cost when one more customer uses available capacity. If supply grows beyond demand, prices for both can fall sharply.
He drew a parallel with the telecommunications expansion around 2000. Companies installed more fiber-optic cable than the market needed, and communication prices subsequently dropped. Griffin questioned whether chips that are five or six years old and currently generate what he described as extraordinary rental rates will continue earning those rates for another five or six years.
That warning matters because AI companies, cloud providers and other infrastructure businesses are committing large sums to chips, servers and data centers. Gartner has forecast worldwide AI spending of nearly $1.5 trillion for 2025, including hardware and AI-enabled phones. Nvidia chief executive Jensen Huang said in January that the sector was only a few hundred billion dollars into what he called the largest infrastructure buildout in history, with trillions of dollars still to come.
If customers become less willing to pay premium prices, the effects would be felt first by chip and cloud companies that depend on continued spending.
Corporate buyers weigh cheaper alternatives
Griffin also said businesses were becoming more selective about how they use AI. Companies are comparing premium models from OpenAI and Anthropic with open-source models that can be downloaded and run on their own computers. The decision depends on whether a particular task justifies the higher cost of a premium service.
He said corporate America had achieved uneven results in turning AI into measurable productivity. Companies are examining their bills from AI providers and asking where the technology is producing sufficient value. That scrutiny could pressure AI laboratories to demonstrate that their spending and pricing can be supported by customer demand.
The possibility of slowing investment is already part of the market debate. Goldman Sachs Research said analysts expect capital-spending growth by the largest AI infrastructure builders to ease from 75% in the third quarter of 2025 to 25% by the end of 2026.
Griffin’s position is also notable because Citadel has participated in the market it is warning about. On July 30, the hedge fund bought most of the public stock portfolio of Situational Awareness, a fund founded by former OpenAI researcher Leopold Aschenbrenner, at about a 10% discount. The portfolio’s value had reportedly fallen roughly 67% in July.
Situational Awareness had used leverage—borrowing to increase its market exposure—of as much as four times its capital. As prices declined, lenders demanded additional collateral. Selling under those conditions is known as forced selling, and it can intensify market declines as sales push prices lower. Griffin wrote in an August letter to Citadel clients that the firm had since reduced more than 80% of the risk in that portfolio.
Moneyiar market picture
Moneyiar market data recorded at approximately 23:28-23:29 on October 22, 2026, showed mixed performance among the listed digital assets:
- Compound (COMP): $24.3408109457, down 5.18% over 24 hours but up 9.37% over seven days.
- Rootstock Infrastructure Framework (RIF): $0.0720136497, down 4.64% in 24 hours and 11.69% over seven days.
- Decimal (DEL): $0.458442584, up 1.85% over 24 hours and 1.61% over seven days.
- USD.AI (CHIP): $0.0440224449, down 8.03% over 24 hours and 7.27% over seven days.
These figures cover digital assets and do not directly measure the performance of AI infrastructure companies. They nevertheless show the mixed short-term conditions visible in the Moneyiar data while investors assess whether enthusiasm for emerging technologies is translating into durable economic value.
