Elon Musk’s AI startup, xAI, is in turmoil behind the scenes. 

The company is constantly playing catch-up with rivals Anthropic, with Musk personally obsessed with having his chatbot Grok match Anthropic’s Claude, Bloomberg reports. And its merger with SpaceX, followed by the latter going public in a historic IPO, led to dozens of employees leaving the company, throwing daily operations into chaos.

To get the house in order, Musk promoted Michael Nicolls to lead xAI (which was recently renamed SpaceXAI). But his work was cut out for him. As part of Musk’s plan to rebuild the company, xAI was seeking to slash up to 30 percent of its workforce in March, and that led to employees being fired without anyone informing them, according to Bloomberg.

Its efforts to hire new talent were equally fraught. Promising candidates would get interviewed and not hear back, because the understaffed human resources department couldn’t finish the necessary paperwork, according to the reporting.

Another challenge was Musk’s obsession with Claude. Reportedly, every time the Anthropic chatbot got an update, Musk demanded that Grok should match it. Numerous projects referenced Claude and some Slack channels were even named after it. Keeping up with Claude had become the startup’s explicit mission.

“Our near-term goals are to match the performance of Claude,” Nicolls wrote in a memo to staffers after becoming xAI president, per Bloomberg.

Some began to doubt Musk’s leadership, believing that the centibillionaire, whose success was built on making cars and engineering rockets, was out of his depth. One of his most ambitious projects, Macrohard, aimed to create a powerful AI agent system capable of emulating an entire software company — something that experts are skeptical is even possible. Soon, some employees complained that creating large language models was much different from creating hardware, per Bloomberg, believing Musk didn’t understand AI.

Other controversies further sowed chaos and dampened morale. At the beginning of the year, Grok sparked outrage for being used to generate millions of conconsensual nudes of real people, including children, disturbing some employees, according to Bloomberg. xAI also told employees that it would pay them to feed their tax returns to train Grok, but never did.

xAI cofounders soon gave a clear vote of no confidence in the company’s direction. In February, some began leaving as Musk prepared to merge xAI with SpaceX ahead of its IPO, and soon after the merger was completed, all eleven of xAI’s cofounders had fled the company.

As the IPO date loomed ever closer, xAI company found itself in such a precarious position that it sold some of its glut of computing power to Anthropic — an ironic development given Musk’s desperate attempts to have xAI emulate its AI chatbot, Claude.

The largest AI companies, including Anthropic, have all struggled to find or build the computing capacity necessary to train and run their powerful systems. xAI is the rare exception, with its model using just 11 percent of its vast array of computing power in April, according to Bloomberg. It also agreed to sell computing power to Google and another AI startup called Reflection.

These agreements secured a multibillion dollar revenue stream, but directly benefited xAI’s biggest rivals — and in a way, waved a big white flag. xAI was practically admitting it wasn’t on the same level as the industry’s top dogs, and was contenting itself with selling hardware, and not building the best AI model.

More on xAI: Grok’s Foul-Mouthed AI “Translation” Feature Puts Unspeakably Ghoulish Words Into Users’ Mouths