Citadel founder Ken Griffin told a New York audience on October 20 that AI is “almost certainly” in a bubble.He stopped short of calling a crash but said firms that emerge from it will lead markets for decades.Some Citadel research shrank from weeks to two or three hoursGriffin was speaking at the Future of Global Markets Conference, an annual event hosted by Citadel Securities, on Tuesday night. “Any radical transformation of our economy is generally intertwined with a bubble,” he said.He cited the dot-com era and said he wished he had held on to his Amazon stock. The companies that survive a bubble “are the market leaders for decades to come,” Griffin said, adding the froth was no reason to sell.Nvidia shares closed at $216.47 on Oct. 20, hours before the talk, then leapt 2.5% to $221.97 the following day.On October 21, Invesco QQQ Trust (QQQ) increased 1.1% to $725.20.Citadel’s own desks show the upside. AI agents, programs that carry out multistep jobs without a human at the helm for each step, trim some of the firm’s research from six to eight weeks to two or three hours.Griffin’s view of AI flipped in 2026. He was dismissive of it at Davos in January, then “unsettled” by its capabilities at Stanford’s business school in May.AI is opening up a strong stretch for entrepreneurs, he also said, and already threatens white-collar jobs.Corporate America gets uneven value from OpenAI and Anthropic billsHe said that GPUs, the chips that train and run AI models, are like airplane seats. Since it doesn’t cost much to sell one more of either, the price can crash when there are more of them than people want.He brought up the telecom boom around the year 2000, when companies laid a lot more fiber-optic cable than anyone needed and call and data rates went down.Chips that are five or six years old still rent for what he called extraordinary rates. He didn’t think they would keep renting for that much for another five or six years.Buyers are getting more selective, said Griffin. They’re figuring out which jobs need premium models from OpenAI and Anthropic, and which can be handled with cheaper open-source models.“Corporate America has been very uneven in its ability to extract value from its AI models,” he said. He spoke of finance teams reviewing the lab bills, asking where the productivity gains are.Cryptopolitan reported that Google brought out Gemini 4 Argon on October 13, priced at an introductory rate of $2 per million input tokens and $10 per million output tokens.Anthropic’s Claude Opus 5.5 costs $4 and $20, while OpenAI’s GPT-6 Astra costs $10 and $50. Meta then published Muse Spark Open, an open-weight model with 816 billion parameters that a number of developers are able to download and use for free under limited commercial terms.Nvidia sells the GPUs Griffin was talking about. It has $279 billion in supply commitments, mostly for memory for its Vera Rubin systems.The company reports its third-quarter results on November 17, with guidance of $108.0 billion.If you're reading this, you’re already ahead. Stay there with our newsletter.