NINE OF ELEVEN co-founders have now departed xAI, the artificial intelligence start-up Elon Musk launched in San Francisco barely two years ago with the stated ambition of understanding "the true nature of the universe." The latest casualties — Zihang Dai and Guodong Zhang, both senior technical leaders — were pushed out this week after Musk grew frustrated with the poor performance of xAI's coding products relative to rivals. Managers from SpaceX and Tesla have been parachuted in to audit employees' work, and some have been fired on the spot. The start-up that was meant to challenge OpenAI is instead being rebuilt from inside by people who build rockets.
The backdrop makes the turmoil especially consequential. In February, SpaceX acquired xAI in an all-stock deal valuing the combined entity at $1.25 trillion — the largest corporate merger in history. The clock is now ticking toward a June IPO that could raise up to $50 billion at a valuation of $1.5 trillion, according to the Financial Times. Musk needs xAI to look like a formidable AI competitor, not a talent sieve. Yet in the AI coding market — the most lucrative battleground in enterprise software right now — xAI's Grok products have failed to gain meaningful traction with paying users or businesses. Anthropic's Claude Code, by contrast, has reached roughly $2.5 billion in annualized revenue and commands an estimated 54% of the coding-agent market. OpenAI's Codex holds around 21%. Grok does not register.
Founder keepers
But the deeper problem at xAI is not technical lag; it is structural instability dressed up as urgency. Musk's playbook — fire fast, reorganize relentlessly, import "fixers" from his other companies — has worked spectacularly at SpaceX, where hardware iteration rewards a culture of controlled demolition (sometimes literally). Applying the same template to a research-intensive AI lab is a different proposition. AI talent is scarce, portable, and courted daily by every well-funded lab in Silicon Valley. The co-founders who departed did not vanish into retirement; several have hinted at launching their own ventures, and at least one, Kyle Kosic, jumped directly to OpenAI.
The revolving door has created a perverse dynamic. Recruiters at xAI are now contacting candidates who were previously rejected, often offering improved financial terms. Musk himself posted an apology on X for past hiring decisions, promising to revisit the company's interview history. This is not the posture of a company with the wind at its back. Meanwhile, the "Macrohard" project — Musk's grandly named initiative to build digital agents capable of replicating entire software companies — has already lost its first leader, Toby Pohlen, who departed just sixteen days after being put in charge. His replacement is Ashok Elluswamy, head of AI software at Tesla, who will now split his attention across two of Musk's entities.
Staff describe a morale crisis. The constant reorganizations prevent teams from building the sustained focus that frontier AI development demands. Training competitive large language models is not a sprint; it requires months of uninterrupted work on data quality, architecture, and evaluation — precisely the areas where xAI has been weakest. One identified reason for the coding product's poor performance, according to people familiar with the matter, is the quality of data used to train its models. That is not a problem solved by firing researchers; it is solved by retaining them long enough to fix it.
The irony is that Musk has the raw infrastructure to compete. The Memphis data center houses more than 200,000 specialized AI chips, with plans to expand to one million GPUs. xAI benefits from the firehose of data generated by X, which Musk merged into the company last year. And his ability to recruit remains potent — this week xAI poached two engineers from Cursor, the popular AI coding application. The inputs are formidable; the question is whether the organism can metabolize them before the next reorganization scatters the team again.
For investors weighing the SpaceX IPO, the xAI question is not peripheral — it is central. At a $250 billion attributed valuation within the merged entity, xAI represents roughly a fifth of what public-market buyers will be purchasing. A start-up burning an estimated $1 billion per month, with no meaningful consumer or enterprise foothold in its target market, and a founding team reduced to two survivors, is a substantial bet embedded inside a rocket company. Musk has compared the current upheaval to Tesla's early chaos, which is fair enough — Tesla nearly died several times before becoming the world's most valuable automaker. But Tesla's turnaround required retaining a core engineering team through the crisis, not replacing it wholesale. Whether xAI's remaining builders can outrun the next round of fixers may determine whether Musk's trillion-dollar AI ambitions reach orbit — or, like so many of his co-founders, quietly drift away. ■
For more, join 75,000 subscribers getting tech's favorite brief here
