Yann LeCun, the founder of AMI Labs, stated in a CNBC interview that xAI's core founding team has largely departed, making it extremely difficult for Elon Musk to recruit top-tier AI talent. Consequently, xAI will be unable to compete effectively with OpenAI and Anthropic in cutting-edge AI development.

He further pointed out that the operational costs of AI services are not declining as rapidly as anticipated. Major labs are currently subsidizing user usage with investor funds, a model he deems unsustainable. LeCun's comments have reignited market skepticism about the valuation of leading AI firms and cast a shadow over the fervent investment climate in the sector. His own venture, AMI Labs, secured a $1 billion funding round in March at a pre-money valuation of $3.5 billion, focusing on what he believes is the next frontier for AI: "world models."

Declaring xAI a "Failure"

LeCun's criticism of xAI centers on its talent crisis. "xAI is honestly a failure in some sense because the founding team left," he remarked. Over the past year, several of xAI's co-founders have departed. LeCun believes Musk's past treatment of his teams has significantly diminished his appeal within the elite AI talent pool. "Elon now is in a very, very difficult position to hire top AI people because of the way he treated the previous team... it was not great," LeCun said.

Commercially, xAI also faces pressure. In February, Musk merged SpaceX with xAI in a deal valuing xAI at a staggering $1.25 trillion. However, financial data shows that the AI business segment (including xAI) under SpaceX reported an operating loss of $2.5 billion for the quarter ending March 31.

LeCun also noted that xAI operates two large data centers in Memphis, Colossus 1 and Colossus 2, and rents out computing power to companies like Google and Anthropic, "because that's the only way he [Musk] can recoup the cost." He expressed a pessimistic outlook on xAI's future.

Unsustainable Business Models

LeCun raised deeper concerns about the broader AI industry's business model. He observed that while the price of AI services is rising, operational costs are not falling fast enough, leading to persistent losses where investors effectively subsidize user consumption. "That can't last very long, right?" he questioned.

He warned that leading labs like OpenAI and Anthropic face three choices: raise prices, cut costs, or face a "bubble explosion." This assessment echoes internal industry concerns. According to CNBC, OpenAI CEO Sam Altman stated in a company livestream this month that enterprise clients are increasingly discussing AI spending, calling AI costs "a huge problem."

Diverging Technical Visions

LeCun holds reservations about the current dominant AI technical path. He has long criticized the limitations of Large Language Models (LLMs), which predict the next word by learning language patterns and excel in reasoning and programming. He argues this architecture cannot support truly general-purpose AI systems.

In contrast, he champions the "world model" approach, which aims to enable AI to build an understanding of how the real or a simulated world operates, encompassing objects, causality, and action logic. "I personally think that we will not have general-purpose, reliable agent systems before we have world models," LeCun stated.

Currently, from Anthropic to OpenAI, major AI firms are heavily investing in AI agents—systems capable of autonomously performing complex tasks. LeCun acknowledges the practical utility of LLMs in areas like programming and mathematics but stresses that "the cost of running these systems at that level of performance is much higher than what people are willing to pay for." He identifies this mismatch as central to the industry's bubble risk.

The $1 billion funding secured by AMI Labs in March represents LeCun's significant bet on the world model path and marks his latest divergence from the prevailing AI narrative.