SpaceX in Talks to Buy Defunct Startup's Data to Train Grok
nashnova research
SpaceX is internally discussing buying customer and operational data from failed startups to train its AI model Grok, Bloomberg reports. This means Grok's data pipeline is expanding beyond Musk's own ecosystem — and the AI training-data race has entered a scavenger phase.
Why buy data from dead companies?
SpaceX's AI unit, SpaceXAI, is in early internal talks to acquire customer and operations data from struggling or defunct startups.
This means → Bankrupt companies have no leverage and no one maintaining their data assets — acquisition costs drop sharply.
Google tried the same playbook: it offered $10 million for business data from Spirit Airlines, the defunct budget carrier. The move sparked privacy complaints from former crew members.
In plain terms = the company is gone, but the data survives. Buy it cheap before anyone else does.
What data has Grok relied on so far?
SpaceXAI previously drew on two internal sources: data from Musk's social platform X, and a team of in-house specialists known as "AI tutors."
The tutor team handled labeling, preference alignment, and verification — essentially telling the AI which answers are good and which are bad — across domains from finance to humor.
This reflects a model that was heavily dependent on Musk's own ecosystem. Pursuing outside data marks a strategic pivot.
What changed inside the AI team?
SpaceXAI froze hiring for its AI-tutor program in June, then replaced the team's leader.
Diego Pasini, a young engineer, was succeeded by Jack Garabedian — a veteran of the Starlink satellite-internet division.
Garabedian focused on stabilizing the group, introducing clear processes, regular meetings, and defined data-training targets.
An internal July briefing cited progress on a new AI model and a coding agent, noting: "Without the team's daily labeling and verification, none of these could ship as finished products."
What did Musk say?
At a recent internal meeting, Musk told SpaceX employees: "We're going to train Grok on all SpaceX information — it will be trained on you."
This means → Beyond buying defunct-startup data, employee data across Musk's empire is also being fed into the training pipeline.
In plain terms = Musk's position is blunt — every usable dataset goes in, whether it comes from his own staff or from someone else's collapsed business.
Why does this matter?
SpaceXAI competes with Anthropic and OpenAI for enterprise clients. High-quality training data is the critical variable for model competitiveness.
Google's Spirit Airlines data purchase already triggered privacy backlash. If SpaceX follows through, it faces the same data-privacy compliance risk.
This signals a broader shift: the AI data race has moved from "who has the most public data" to "who can acquire the most unique private data at the lowest cost" — and defunct companies are the new hunting ground.
市场有风险，内容仅供研究参考，不构成投资建议。