Yann LeCun's attack on xAI lands because the hard facts are worse than the insult: the founders are gone, and SpaceX is now renting huge chunks of its AI compute to the companies Grok was meant to fight.
Yann LeCun did not need to dress this up. In a June 18 CNBC interview, the former Meta chief AI scientist called Elon Musk's xAI "kind of a failure" and said he was not positive about its chances against the frontier labs. You can dismiss some of that as rivalry. You shouldn't dismiss the evidence underneath it.
Business Insider reported that all of xAI's non-Musk co-founders have now left the company. Ross Nordeen, the final holdout, was cut off from company systems and disappeared from a large internal group chat, according to the same report. Manuel Kroiss, who led pretraining work, had already left in March, making him the 10th of 11 original co-founders to go. Earlier departures included Tony Wu and Jimmy Ba, who left in February as the company was being pulled into SpaceX.
That is not a normal churn story. In AI, the founding team is not decoration around the brand. It is the product, the recruiting engine, the memory of why certain technical choices were made. When every original researcher and operator around Musk has gone, you are looking at more than a management reshuffle. You are looking at a lab that has lost the people who made its first claim to seriousness.
LeCun's explanation was blunt. Musk has made it "very, very difficult" to hire top AI people, he told CNBC, because he has not behaved well toward the previous team. That is a hard thing to prove from the outside, and LeCun is not neutral. But the departure record does a lot of work on its own. If you are trying to beat OpenAI, Anthropic and Google, you cannot make your founding bench disappear and pretend nothing has changed.
The money story is just as revealing. SpaceX acquired xAI in February 2026 in an all-stock transaction that valued SpaceX at $1 trillion and xAI at $250 billion, for a combined $1.25 trillion valuation, according to Reuters and Business Insider reporting at the time. By June, SpaceX had gone public, and its AI infrastructure was being sold to the same outside labs Musk had spent years trying to outbuild.
Anthropic announced in May that it would use SpaceX's Colossus One data center near Memphis for more than 300 megawatts of computing capacity across more than 220,000 Nvidia GPUs, Business Insider reported. A later SpaceX filing cited by Business Insider said Anthropic was paying $1.25 billion a month through May 2029 for compute capacity from SpaceX's Colossus data centers. Google then agreed to pay SpaceX $920 million a month from October 2026 through June 2029 for about 110,000 Nvidia GPUs, CPUs, memory and related infrastructure, according to a SpaceX SEC filing reported by Business Insider.
LeCun put the point plainly: SpaceX has huge infrastructure and is renting it out because that is how it can recoup the cost. Frankly, that is a better business than pretending idle GPUs are a strategy. The uncomfortable part is who is paying. Anthropic's Claude models compete directly with Grok. Google competes directly with Grok. If your AI challenger is now selling compute to its challengers, the pitch has changed.
There is a version of this story where Musk still wins. Amazon Web Services became one of the most important businesses in technology by renting infrastructure to everyone else. CoreWeave built a public-market story around Nvidia-backed AI compute. If SpaceX can turn Colossus into a durable cloud business, investors may not care whether Grok ever catches Claude or ChatGPT.
But do not confuse a good infrastructure business with proof that xAI worked as an AI lab. xAI was launched in 2023 as a frontier-model company, not as a landlord for Nvidia racks in Memphis. It absorbed X in 2025, folded into SpaceX in 2026, lost its original co-founders, and is now using Colossus to serve outside customers. That is a lot of movement for a company still trying to explain what it is.
LeCun has his own bet to sell
You also need to read LeCun with his incentives in view. Reuters reported in March that his new company, Advanced Machine Intelligence Labs, raised $1.03 billion at a $3.5 billion pre-money valuation. The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions. AMI Labs is built around LeCun's belief that large language models are not the final path to machine intelligence, and that world models, systems that understand physical cause and effect, matter more.
So when LeCun says xAI cannot compete, you are hearing a founder with a live commercial thesis. He is not a detached professor watching from the balcony. He is building a rival company whose entire argument gets stronger if the LLM race looks overheated, expensive and badly managed. His criticism may still be right. It is not disinterested.
He also warned CNBC that AI companies risk a "big bubble explosion" if they do not cut costs and raise prices. That warning lands differently from someone who has just raised more than $1 billion to pursue an alternative architecture. If the current model race breaks under its own spending, AMI Labs benefits. Keep that in mind.
Still, the xAI facts are hard to wave away. Business Insider reported that SpaceX and xAI did not respond to requests for comment on LeCun's remarks. That silence does not prove his case, but the company filings, the compute contracts and the co-founder exits already say plenty. Grok may improve. SpaceX may make billions from Colossus. But the original xAI story was about beating the frontier labs, and right now the clearest revenue line comes from renting hardware to them.
Whether you call that failure depends on what you thought xAI was supposed to be. If it was meant to become a cloud infrastructure arm inside SpaceX, the strategy is working. If it was meant to build the world's strongest AI lab around Musk, LeCun's insult is not the problem. The facts are.
Also read: Waymo's fourth recall in two years exposes a pattern the robotaxi industry can't afford to ignore • Rumble bets its future on GPU cloud infrastructure after closing its $767 million Northern Data deal • Hyundai takes full control of Boston Dynamics as SoftBank exits for $325 million
