SpaceXAI Explores Buying Customer Data From Troubled Startups to Train Grok AI Models

SpaceXAI Explores Acquiring Third-Party Data

SpaceXAI has held informal internal discussions about buying customer and operational records from troubled startups to train its AI models, according to Bloomberg1

1

2

2

. The talks focus on acquiring digital assets of defunct startups as a cost-effective solution for obtaining AI training data. While these discussions remain preliminary and may not result in deals, they signal a strategic shift in how SpaceXAI sources information to train its AI models, particularly Grok.

The AI division needs cheaper data options urgently. SpaceX announced a $15.8 billion commitment to its AI efforts in its first earnings report, even as the company's AI division ended the quarter with a $1.3 billion net operating loss2

2

. Elon Musk has set ambitious targets for Grok to compete with rival models from OpenAI and Anthropic, with plans to unveil Grok 5 before year-end, claiming it will reach AGI-level capabilities.

Bankruptcy Auctions as Data Marketplaces

The template for buying data from struggling or bankrupt startups already exists. Google paid $10 million in a bankruptcy auction for Spirit Airlines' internal records after the discount carrier collapsed1

1

3

3

. That archive contained 100 million emails, 500 million Microsoft Teams messages, aircraft operations data, employee productivity records, 1 million time-card records, 17 million individually owned Microsoft OneDrive items, and hundreds of thousands of employee records including tax forms and employment contracts.

Chapter 11 bankruptcy proceedings treat customer databases and digital estates the same as office furniture—assets to be auctioned to the highest bidder. Nobody asks the customers for permission because by the time the gavel drops, there's no company left for them to complain to3

3

. SpaceXAI wants operational data and customer information from startups in exactly that position to keep training Grok on how real businesses actually run.

Privacy Concerns and Legal Challenges

The strategy raises significant ethical concerns and privacy concerns. Ireland's Data Protection Commission opened a statutory inquiry in April 2025 into whether EU and EEA users' public posts were lawfully processed to train the Grok models1

1

. The regulator had already used emergency High Court powers against X in August 2024, questioning whether data protection laws were being followed.

Under the European Data Protection Board's 2024 opinion, lawful processing turns partly on what people would reasonably expect. A customer of a company that no longer exists has no expectations about their data being used to train Grok1

1

. Consent does not travel with a database either—the buyer becomes a new controller and must find its own lawful basis for processing.

The Association of Flight Attendants filed an objection with the U.S. bankruptcy court regarding Google's Spirit Airlines data purchase, arguing that deidentification standards only cover consumers, leaving employees' private data vulnerable2

2

. The union stated that pseudonymized datasets can still reveal which employees generated grievances, performance data, investigation subjects, and internal communications about management. The bankruptcy court has postponed approval of the purchase.

SpaceXAI's Evolving Data Strategy

This represents a change of direction for SpaceXAI. The unit has previously trained Grok on posts from X and on in-house specialists called AI tutors1

1

. Hiring for the AI tutors team was paused in June, with Starlink staffer Jack Garabedian taking over operations. A dead company's customer list represents a different kind of data with a different kind of history compared to social media posts or curated training datasets.

Elon Musk has been transparent about broader ambitions. In August, he told SpaceX staff the company would train Grok on the sum total of all SpaceX information, telling employees the model would "inherit your thoughts" and absorb their outlook along with their work3

3

. However, he did not specify which categories of employee data would actually be used or how.

Industry-Wide Implications

Facilitating failed companies' sale of employee data to AI labs has become a niche market. As agentic AI makes its way into the workplace, AI labs increasingly value industry-specific knowledge across sectors2

2

. Any company that users gave their data to and is now struggling financially might end up selling it to train AI models.

SpaceXAI's European privacy notice says it relies on legitimate interests to improve its models and names a US entity as the controller for European users, with all processing happening in American data centres1

1

. The company released Grok 4.5 in July as its first model since the xAI merger closed, with a $60 billion Cursor acquisition and a delayed new version of Grok coming up3

3

. Watch how bankruptcy courts rule on employee data sales and whether European regulators expand their investigations beyond social media data to include purchased digital estates.