SpaceX explores buying data from failed companies to train Grok

16:00

By: Azzat Manal

SpaceX explores buying data from failed companies to train Grok

SpaceX is reportedly exploring the possibility of acquiring customer and operational data from struggling or defunct companies as it seeks additional material to train its artificial intelligence systems, including Grok. According to reports citing people familiar with the discussions, the idea remains at an informal stage and may not result in any transaction.

The discussions reflect the growing value of proprietary corporate information in the artificial intelligence industry. While public data has traditionally played a major role in training large language models, AI developers are increasingly looking for real-world business records that can help systems understand specialized workflows, workplace communication and industry-specific processes.

For SpaceX and its AI operations, access to large and diverse datasets could support efforts to improve Grok and develop increasingly capable AI systems. Elon Musk has previously described ambitious plans for Grok, while the broader AI industry continues to compete over computing power, talent and high-quality training data.

The potential acquisition of data from bankrupt companies also raises significant questions about privacy and how information originally collected for one purpose can be used after a business collapses. A company’s financial failure does not necessarily erase the data it accumulated during its operations, leaving customer and employee information potentially subject to asset sales conducted through bankruptcy proceedings.

A recent case involving Spirit Airlines illustrates the emerging market. Google won a bankruptcy auction for $10 million to acquire a large collection of the airline’s internal corporate data, including about 100 million emails and 500 million Microsoft Teams messages. The package also included business documents, operational information, software and other records. Passenger profiles and loyalty-program information were excluded from the proposed transaction.

Google said the information it receives will be processed to remove personally identifiable information before being transferred and that the company intends to use the material to improve its products and artificial intelligence models. The transaction, however, has generated objections from former Spirit employees and their representatives, who have raised concerns about the protection of sensitive workplace information.

The Spirit case demonstrates why corporate data can be particularly attractive to AI developers. Unlike much of the information available on the public internet, internal business archives contain examples of how employees communicate, manage operations, solve problems and make decisions in real working environments. Such material could be useful for developing AI agents designed to perform professional tasks.

At the same time, removing names and other obvious identifiers does not automatically eliminate every privacy concern. Workplace emails, messages and documents can contain contextual information that may make individuals identifiable, particularly when large datasets are combined with information from other sources. Privacy advocates and employee representatives have therefore questioned whether existing safeguards are sufficient for the growing use of corporate data in AI development.

The emerging interest in failed companies’ data points to a broader shift in the AI industry. As developers seek increasingly specialized training material, information generated through everyday business activity is becoming a potential commercial asset. Bankruptcy proceedings could consequently become an unexpected channel through which valuable corporate datasets reach technology companies and AI developers.

For SpaceX, any future purchase would have to navigate the legal and privacy obligations attached to the data involved. The current discussions reportedly remain preliminary, meaning there is no confirmed agreement for the acquisition of customer records from a particular failed company.

The debate nevertheless highlights a growing challenge for the digital economy: determining what happens to personal and workplace information when the company that originally collected it disappears. As AI developers place greater value on proprietary data, questions surrounding consent, ownership, anonymization and the permissible reuse of information are likely to become increasingly important.