Elon Musk is reshuffling xAI's engineering ranks from top to bottom as the company races to catch rivals before SpaceX takes it public at a potential multi-trillion-dollar valuation.
Michael Nicholls, the SpaceX executive now serving as xAI's president, did not mince words. In an internal memo viewed by Business Insider, he told staff the company is "clearly behind" competitors and that its compute training performance is "embarrassingly low." Those are striking admissions from a venture backed by some of the most deep-pocketed investors in technology, and they signal just how much ground xAI needs to make up before SpaceX heads to the public markets.
SpaceX acquired xAI in February, folding Musk's artificial intelligence startup into a corporate structure expected to pursue an initial public offering that could value the combined entity at over $2 trillion, based on reports from multiple financial outlets. That valuation depends heavily on xAI proving it can stand toe to toe with OpenAI, Anthropic, and Google, all of which have released increasingly capable models over the past eighteen months. Right now, xAI's Grok assistant remains a niche product integrated primarily into the X social media platform, with limited external distribution.
The reorganization places specific technical leaders in charge of each phase of the model development pipeline. Devendra Chaplot, who joined xAI last month after stints at Facebook and Thinking Machines Lab, will run pre-training, the foundational stage where models absorb patterns from massive internet-scale datasets. Aman Madaan takes over model factory and tooling, essentially the plumbing that keeps training infrastructure running smoothly.
Post-training duties are split. Aditya Gupta will head general post-training and reinforcement learning, the critical fine-tuning phase where raw models are shaped into useful products. Beibin Li, formerly of Microsoft and Meta, will handle post-training specifically for Grok Code, xAI's programming assistant. Xuhui Jia from Google DeepMind and Yukun Zhu will oversee video and image training, areas where competitors like OpenAI's DALL-E and Sora have established strong early leads.
On the product side, Andrew Milich and Jason Ginsburg, both recruited from AI coding startup Cursor in March, will manage Grok Main, Grok Voice, and Grok Imagine. Physical infrastructure falls to Jake Palmer, while Daniel Dueri, a SpaceX software engineering director, handles compute infrastructure. Matt Monson, who directs Starlink software at SpaceX, adds data leadership at xAI to his responsibilities.
The Cost of Constant Turmoil
This is not the first shakeup. xAI has been in near-constant organizational flux since the SpaceX acquisition, and the instability has extracted a real cost. Eight of the engineers who co-founded the company alongside Musk have departed since January, including Ross Nordeen, one of Musk's closest deputies, Guodong Zhang, Manuel Kroiss, and Toby Pohlen. At times, Musk has managed dozens of direct reports directly, a structure that rarely scales well for a company pushing into some of the most complex engineering challenges in software.
Bringing in SpaceX and Tesla veterans to fill the gaps makes operational sense given the shared corporate family, but it also raises questions about whether xAI is building genuine AI expertise or simply reshuffling trusted lieutenants from other Musk companies. The distinction matters, particularly for investors evaluating the company's technical depth ahead of an IPO.
Nicholls has promised significant improvements in training performance within the next two months, an aggressive timeline that reflects the urgency of the moment. If xAI cannot demonstrate credible progress in model capability and infrastructure efficiency before the public offering, the risk is that SpaceX arrives on the public market carrying an AI division that looks more like a work-in-progress than a competitive threat to the industry's leaders.
For startups and investors watching from the sidelines, the xAI saga is a useful reminder that capital and brand recognition alone do not win in artificial intelligence. The sector rewards technical velocity and talent retention above all else, and right now xAI is trying to rebuild both under extreme pressure.
