SpaceX Weighs Buying Failed Startups’ Data for AI

SpaceXAI was created after SpaceX acquired Elon Musk’s xAI business in February, and the talks are occurring within that newly combined artificial-intelligence operation.
The proposed purchases would expand on SpaceXAI’s existing use of X posts and in-house “AI tutors,” whose training work has included specialized areas such as finance, science and humor; hiring for the tutor team was paused in June, and former Starlink employee Jack Garabedian later took over the group.
Garabedian reportedly introduced clearer processes, meetings and goals for training data, while an internal newsletter said annotation, preference alignment and validation were essential to turning a new AI model and coding agent into finished products.
The Spirit Airlines archive that Google reportedly sought contained about 100 million emails and 500 million Microsoft Teams messages, illustrating the scale and sensitivity of the type of corporate data that AI developers may pursue.
Elon Musk told SpaceX employees that Grok would be trained on “the sum total of all SpaceX information,” including information generated by the company’s employees, extending the proposed data strategy beyond external startup acquisitions.
SpaceX's AI division is exploring a controversial strategy: buying customer and operational data from failed startups to train Grok and other AI models. GuruFocus reported that the talks remain preliminary and no deals have been announced. The datasets could include customer emails, internal documents, code and workflow records — far beyond what SpaceX already uses from X posts and internal AI trainers.
The approach mirrors Google's reported effort to acquire Spirit Airlines' 100 million emails and 500 million Teams messages. But it raises serious privacy questions, especially as Irish regulators investigate whether European X users' posts were legally used to train Grok without permission.
SpaceXAI was created after SpaceX acquired Elon Musk's xAI business in February. PCMag noted the newly combined team currently trains Grok using X posts and in-house "AI tutors" who teach finance, science and humor. The tutor hiring paused in June, then former Starlink employee Jack Garabedian took over the program.
Garabedian introduced clearer processes and goals for training data. An internal newsletter emphasized that annotation, preference alignment and validation are essential to finishing new AI models and coding agents. The startup data purchases would dramatically expand this pool of training material.
Elon Musk went further with SpaceX staff, saying Grok would be trained on "the sum total of all SpaceX information" — including data from employees themselves. This extends the AI training strategy beyond buying defunct startup records to harvesting internal company knowledge.
The Spirit Airlines archive Google pursued contained roughly 100 million emails and 500 million Teams messages. This scale shows how much sensitive corporate data AI companies are chasing. Failed startups hold similar treasure troves of customer records, financial documents and proprietary processes.
PCMag reported that SpaceX is specifically targeting customer and operational data from troubled companies. The deals would be affordable ways to build massive training datasets. But regulators worldwide are watching how AI firms obtain and use private information.
Ireland is already investigating whether X users' public posts were lawfully used to train Grok. Bitdefender noted this adds risk to SpaceXAI's aggressive data-buying plans. If regulators rule that X training violated privacy laws, similar restrictions could apply to startup data purchases.
Publishers
17
Articles
35
Reach
52