SpaceX has held internal discussions about buying customer and operational data from troubled or defunct startups to improve its AI models, according to Bloomberg reporting that cites people familiar with the matter. The idea is to find a more affordable source of data.

The key limitation: these are reportedly informal discussions, not announced transactions. Bloomberg said they may not lead to any deals. The report does not identify potential sellers, describe specific datasets or explain which models would use the information.

Why a failed startup’s data could still have value

A company can run out of funding or fail to find a sustainable business model while still holding useful information. Depending on the business, that could include records of customer interactions, product usage or internal workflows. Those are possible examples, not details established by the report.

For an AI developer, such records might offer insight into real-world tasks and business processes. A startup winding down could also have an incentive to sell assets that otherwise would remain unused.

But a low purchase price does not necessarily make a dataset economical. Reviewing its origins, removing sensitive information and preparing inconsistent records for use can add substantial costs.

Buying data is not the same as securing permission to train on it

The central question in any potential transaction would be what rights the seller can actually transfer. Possessing customer information does not automatically establish permission to sell it or authorize every downstream use.

A buyer would need to examine privacy notices, customer contracts, applicable law and any restrictions attached to third-party material. A shutdown or financial crisis does not, by itself, eliminate those constraints.

Customer data and operational data also raise different concerns. Operational records may contain confidential business information, while customer records may include personal details. Some datasets could contain both.

What would make a dataset worth acquiring?

Beyond the asking price, a practical evaluation would focus on four issues:

• Provenance: Is there a reliable record of how the information was collected?
• Usage rights: Is the intended AI use permitted?
• Quality: Are the records accurate, consistent and sufficiently documented?
• Relevance: Does the information address a specific model-development need?

More data is not automatically better data. Outdated, duplicated or poorly documented records may offer little benefit, even if they are inexpensive.

What to watch next

The next meaningful development would be evidence of an actual agreement, along with details about the data involved and its permitted uses. For now, the report describes a possible sourcing strategy—not a completed acquisition or a demonstrated improvement in AI performance.

This article was inspired by SpaceX May Buy Data From Failed Startups for AI Models from Bloomberg Tech. Please visit the original video for the creator’s full presentation and context.