The corporate structure of Elon Musk's technology empire has undergone another round of consolidation, and the result is a set of Russian nesting dolls that would challenge even the most diligent securities lawyer to diagram.
xAI, the artificial intelligence company Musk founded in 2023 to compete with OpenAI, has been dissolved as a standalone entity and absorbed into SpaceX.
The social media platform X, which xAI had previously swallowed, is now a feature within a feature of SpaceX. The combined entity is being referred to, at least in some press materials, as SpaceX AI, though it is unclear whether this is the name of the entire company or merely a branding exercise for its AI division.
The consolidation was announced alongside a compute partnership with Anthropic, the AI company behind Claude, under which SpaceX will provide Anthropic with access to Colossus 1, the roughly 220,000-GPU data centre in Memphis that was originally built for xAI's model training.
Musk said xAI's own training workloads had moved to Colossus 2, a larger cluster under construction, freeing up the original facility for commercial use. The Anthropic deal includes a clause allowing Musk to reclaim the compute resources if he determines that Anthropic's AI "engages in actions that harm humanity," a provision that has no obvious precedent in commercial data centre agreements and raises the question of who, exactly, would make that determination.
The arrangement is remarkable for several reasons. First, it represents a pivot for xAI from frontier AI research to infrastructure-as-a-service. Musk launched xAI with the stated ambition of building a truth-seeking AI that could compete with OpenAI's GPT and Anthropic's Claude. Renting out the compute capacity originally built for that purpose suggests the competitive AI model effort has been deprioritised in favour of a cloud computing business.
Second, the corporate structure is now genuinely unusual. SpaceX is a private aerospace company with government contracts, national security clearances and a dominant position in the global launch market. It now also contains a social media platform with hundreds of millions of users, an AI division that was until recently a separate company, and a commercial data centre operation serving one of the most prominent AI startups in the world. The regulatory, governance and conflict-of-interest implications of housing all of these businesses under a single private corporate umbrella are substantial and largely untested.
Third, the partnership with Anthropic is a notable shift in tone. Musk has previously called Anthropic "misanthropic" and publicly attacked its leadership. The rapprochement, which Musk framed as the result of being personally impressed by Anthropic's staff, suggests that commercial pragmatism has overtaken ideological rivalry, at least for now.
Related reading
- X rolls out Grok-powered translation and AI image editing tools
- xAI lands permit for Mississippi power plant
- Elon Musk's ego may be the biggest obstacle to his OpenAI lawsuit
For Anthropic, the deal provides an immediate solution to a pressing problem: its paying subscribers have been complaining about rate limits and degraded service during peak hours, and its other infrastructure agreements with Amazon, Google and Microsoft will not deliver meaningful capacity until late 2026 or 2027. Access to 220,000 GPUs within weeks fills a gap that no other partner could address on that timeline.
The question is what it means for the longer term. Anthropic is now operationally dependent on infrastructure controlled by one of the most unpredictable figures in the technology industry, under a contract that includes a unilateral termination clause based on a subjective assessment of harm to humanity. That is not a standard cloud computing arrangement, and the risks it introduces for Anthropic's service continuity are not trivial.