Yann LeCun, one of the most influential figures in artificial intelligence and often called the “godfather of AI,” has launched a sharp attack on Elon Musk’s xAI and warned that the wider sector could face a financial crisis. In an interview with CNBC, LeCun said xAI is effectively a “failure” and argued that the leading AI companies are built on an economic model that cannot last.
LeCun, formerly Meta’s chief AI scientist and now head of AMI Labs, has been publicly sparring with Musk for years over the future of AI. He said the dispute has ranged from the capabilities of AI models to Musk’s social media posts. “Frankly, xAI is sort of a failure, because the founding team left,” he said, adding that Musk is now struggling to attract top researchers because “he did not behave particularly well toward the previous team.” He noted that several xAI founders left the company over the past year.
Musk continues to spend heavily on computing infrastructure as he tries to compete with OpenAI, Anthropic and Google. In February, SpaceX and xAI completed a merger that valued the combined company at about $1.25 trillion. LeCun questioned whether xAI can justify the enormous investment, saying the company has “huge infrastructure” and rents out some of its computing power to other firms. “That’s the only way he can cover the costs,” he said, referring in part to the Colossus 1 and Colossus 2 data centers xAI is building in Memphis. Google and Anthropic have already leased computing capacity there.
LeCun also said he is “not very optimistic” about xAI’s future and does not expect it to compete with OpenAI and Anthropic. More broadly, he said many top AI companies are losing money because investor capital is subsidizing users while revenues still do not cover operating costs. “It cannot go on much longer,” he said, warning that firms will soon need to raise prices, cut costs, or face “a big bubble burst.”
He reiterated his long-standing skepticism of large language models, the technology behind ChatGPT and Claude, and said true advanced AI will require “world models” that understand environment, cause and effect, and how the world works. He said reliable general autonomous systems will not arrive until they are based on that approach. While he accepts that language models are useful in areas such as programming and mathematics, he said their operating costs remain too high for what users are willing to pay.
