SpaceX weighs failed-startup data purchases for AI

SpaceX has held informal discussions about buying customer and operational data from troubled or defunct startups to train artificial intelligence models, people familiar with the talks say, as the company looks for lower-cost information beyond material available on the public internet.

The discussions are taking place within SpaceXAI, the artificial intelligence operation created after SpaceX acquired Elon Musk’s xAI business in February. No transaction has been announced, the talks are preliminary and may not lead to any purchases. SpaceX has not publicly detailed what types of companies it could target, how much it might spend or what safeguards would apply to any acquired datasets.

The proposal would give SpaceXAI access to records generated inside real businesses, potentially including customer interactions, support material, internal documents, workflow data and other operational information. Such material is increasingly valuable to developers seeking to train AI agents that can perform practical workplace tasks rather than merely generate text from internet-based knowledge.

Demand for this kind of information has expanded as developers search for datasets showing how employees solve problems, communicate with customers, write software, manage projects and make decisions. Companies specialising in data acquisition have approached founders of failed or sold startups about licensing code bases, emails, documents, support tickets and internal communications, typically with identifying information removed before the material is supplied for model development.

For SpaceXAI, acquiring datasets directly could provide another source of training material for Grok and other models while reducing dependence on expensive bespoke data-generation programmes. The company already uses computing infrastructure and has integrated artificial intelligence more deeply into SpaceX since the xAI acquisition.

Regulatory filings show that SpaceX completed the all-stock merger with xAI in February, bringing the Grok developer and the X social-media platform into the wider group. The artificial intelligence activities were subsequently reorganised under the SpaceXAI name, which now covers the company’s AI products and services.

Any attempt to purchase data from failed businesses would nevertheless raise questions over ownership, consent and privacy. A startup shutting down may retain valuable databases and internal records, but the right to sell or license those assets can depend on contracts, privacy notices, bankruptcy proceedings and applicable data-protection laws. Customer records may also contain personal or commercially sensitive information that requires deletion, anonymisation or other restrictions before reuse.

The distinction is important because SpaceXAI’s current enterprise terms state that user content supplied by business customers is not used to train foundation models, large language models or other AI systems, subject to disclosed settings and agreements. A separate purchase of corporate datasets would therefore involve a different legal and contractual route from using ordinary customer interactions with SpaceXAI services.

SpaceXAI’s privacy policy, effective from August 24, says the company collects and uses personal information connected with its consumer services and describes separate treatment for data processed on behalf of business customers. The policy does not by itself establish what rules would govern datasets bought from third parties, which would depend on the source, contractual rights and jurisdiction.

The possible data-buying strategy emerges as SpaceX turns artificial intelligence into a larger part of its business alongside rockets, Starlink and computing services. Regulatory filings this year have described AI as a material part of the combined company’s strategy and highlighted the integration of xAI into SpaceX.

SpaceX has also been expanding its role as a provider of computing capacity. A filing with the US Securities and Exchange Commission disclosed a June agreement under which Google is due to pay $920 million a month for access to computing capacity that includes about 110,000 Nvidia graphics processors, with the arrangement running through June 2029 subject to termination provisions. The filing states that the customer retains ownership of its content, AI models and related data.